Train bottom fault detection intelligent sleeper based on image recognition

Abstract

The invention discloses a train bottom fault detection intelligent sleeper based on image recognition, and relates to the technical field of train fault detection. The three image acquisition modules are arranged in the middle and at the two ends of the sleeper body correspondingly, so that pictures, collected by the three image acquisition modules, of the bottom of a train can cover the whole picture of the bottom of the train after being spliced. The power supply module is mounted on the sleeper body and is used for supplying power to the image acquisition module; the data management module is installed on the sleeper body and used for uploading the images collected by the image collection module to a train detection system terminal. According to the intelligent sleeper, faults in the whole train bottom range can be detected through the image collection modules, images obtained through linear array image collection formed by the three image collection modules can be directly spliced, special processing is not needed, the image authenticity is guaranteed, the faults of the train bottom in the train running process can be monitored in real time, and the fault detection efficiency is improved. And vehicle bottom defects possibly formed under the dynamic action are analyzed, so that the early warning effect is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of train fault detection, and more specifically, to an intelligent sleeper for train underbody fault detection based on image recognition. Background Art

[0002] In railway transportation, the train underbody is the part with the most complex structure and the most numerous components, and ensuring its safety is extremely crucial. At present, the inspection of the train underbody mainly relies on the maintenance in vehicle depots and factories, which is not only costly, but also may miss problems, and the detection technology is relatively outdated. Traditional manual inspection has many deficiencies, mainly relying on train inspectors using primitive inspection means, such as visual inspection, tactile inspection, and using tools such as tapping with a hammer for assistance. These methods not only have low efficiency, but also bring a heavy workload to train inspectors. Generally speaking, this inspection method completely depends on the working state of train inspectors. When train inspectors are in a poor state due to high work intensity, safety problems may occur.

[0003] Although the existing under-rail detection trolley can extend the detection arm into the underbody for detection, the detection time is only when the train is not running and in a stationary state, and it cannot reflect in real time the faults existing during the train operation and give early warnings of the possibility of fault occurrence.

[0004] The existing equipment for monitoring faults during train operation is mainly installed beside the track, and the monitoring angle of the under-rail is limited, and it cannot accurately and comprehensively obtain the fault conditions of the train underbody. Moreover, before the acquired images are merged, they are processed through perspective transformation and distortion correction, etc., which causes image distortion to a certain extent and cannot accurately reflect the underbody fault conditions.

[0005] Therefore, how to be able to perform real-time detection of train underbody faults without stopping the train and meet a more comprehensive detection effect, so as to provide a simple and efficient structure, is an urgent problem to be solved by those skilled in the art. Summary of the Invention

[0006] In view of this, the present invention provides an intelligent sleeper for train underbody fault detection based on image recognition, aiming to solve the above technical problems.

[0007] To achieve the above object, the present invention adopts the following technical solutions:

[0008] An intelligent sleeper for train underbody fault detection based on image recognition, comprising:

[0009] A sleeper body;

[0010] An image acquisition module, the number of the image acquisition modules is three, and they are respectively arranged in the middle and at both ends of the sleeper body, so that the pictures of the train underbody collected by the three image acquisition modules can cover the entire panorama of the train underbody after splicing;

[0011] A power supply module, the power supply module is installed on the sleeper body and is used to supply power to the image acquisition module;

[0012] A data management module, the data management module is installed on the sleeper body and is used to upload the pictures collected by the image acquisition module to the train detection system terminal.

[0013] Through the above technical solutions, the intelligent sleeper proposed by the present invention can detect faults within the entire range of the underbody through the image acquisition module. The images obtained by the three image acquisition modules forming a linear array image acquisition can be directly spliced without special processing, ensuring the authenticity of the images. It can monitor the faults of the train underbody in real time during the train operation, and analyze the possible underbody defects formed under dynamic actions to achieve a warning effect.

[0014] Preferably, in the above intelligent sleeper for train underbody fault detection based on image recognition, the image acquisition module includes an image acquisition box, a high-speed camera and a fill light fixed inside the image acquisition box; the image acquisition box is fixed on the sleeper body, and the top surface of the image acquisition box is made of transparent material.

[0015] Preferably, in the above intelligent sleeper for train underbody fault detection based on image recognition, the image acquisition box includes a box body and a cover body, the box body and the cover body are fixedly connected by bolts, and the cover body is provided with special transparent glass.

[0016] Preferably, in the above intelligent sleeper for train underbody fault detection based on image recognition, a dust-proof wiper is installed on the cover body, and the dust-proof wiper is powered by the power supply module.

[0017] Preferably, in the above intelligent sleeper for train underbody fault detection based on image recognition, the high-speed camera is installed at the center of the box body, the number of the fill lights is two, and they are respectively installed on both sides of the high-speed camera.

[0018] Preferably, in the above intelligent sleeper for train underbody fault detection based on image recognition, a power / data transmission interface is provided on the side wall of the box body, and the power / data transmission interface is used to realize the electrical connection between the high-speed camera and the fill light inside the box body and the power supply module and the data management module.

[0019] Preferably, in the above-mentioned intelligent sleeper for train underbody fault detection based on image recognition, the power supply module and the data management module are respectively installed in two spaced spaces of the three image acquisition modules.

[0020] Preferably, in the above-mentioned intelligent sleeper for train underbody fault detection based on image recognition, the image acquisition module extracts and matches feature points through the SIFT algorithm to achieve image stitching.

[0021] Preferably, in the above-mentioned intelligent sleeper for train underbody fault detection based on image recognition, the sleeper body is placed between two adjacent railway sleepers and arranged parallel to them. The sleeper body is buried in the ballast, and the burial depth is the same as that of the railway sleeper.

[0022] Preferably, in the above-mentioned intelligent sleeper for train underbody fault detection based on image recognition, the positions of the three image acquisition modules on the installed sleeper body are respectively located between the two rails and on the outside.

[0023] Through the above technical solutions, compared with the prior art, the present invention discloses an intelligent sleeper for train underbody fault detection based on image recognition. The underbody image is obtained by a line array camera, and a deep learning network model is used to identify the faults existing in the image, realizing the real-time monitoring function of train operation safety. The intelligent sleeper also adopts a wireless transmission technology to transmit the underbody fault information to the terminal of the train detection system. The terminal of the train detection system analyzes and processes the image, realizing the real-time monitoring of the train underbody state, which is beneficial for the staff to timely understand the train safety status and timely formulate or adjust the maintenance plan. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only the embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained according to the provided drawings without creative efforts.

[0025] Figure 1 The drawings are the top view of the structure of the intelligent sleeper for train underbody fault detection based on image recognition provided by the present invention;

[0026] Figure 2 The drawings are the schematic structural diagrams of the image acquisition module provided by the present invention;

[0027] Figure 3 The drawings are the top view of the internal structure of the image acquisition module provided by the present invention;

[0028] Figure 4The attached drawing is a top view of the installation state of the intelligent sleeper for train underbody fault detection based on image recognition provided by the present invention.

[0029] Among them:

[0030] 1 - Sleeper body;

[0031] 2 - Image acquisition module;

[0032] 21 - Image acquisition box; 211 - Box body; 2111 - Power / data transmission interface; 212 - Cover body; 2121 - Special transparent glass; 22 - High - speed camera; 23 - Fill light;

[0033] 3 - Power module;

[0034] 4 - Data management module;

[0035] 5 - Railway sleeper;

[0036] 6 - Rail. Specific implementation manners

[0037] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the attached drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0038] Refer to the attached Figure 1 , the embodiments of the present invention disclose an intelligent sleeper for train underbody fault detection based on image recognition, including:

[0039] Sleeper body 1;

[0040] Image acquisition module 2, the number of image acquisition modules 2 is three, and they are respectively arranged in the middle and at both ends of the sleeper body 1, so that the pictures of the train underbody collected by the three image acquisition modules 2 can cover the entire picture of the train underbody after splicing;

[0041] Power module 3, the power module 3 is installed on the sleeper body 1 and is used to supply power to the image acquisition module 2;

[0042] Data management module 4, the data management module 4 is installed on the sleeper body 1 and is used to upload the pictures collected by the image acquisition module 2 to the train detection system terminal.

[0043] In this embodiment, the train detection system terminal receives the underbody images taken by the intelligent sleeper system through the local area network, splices the images of the three image acquisition modules 2, uses the deep learning algorithm to identify faults, and realizes accurate fault positioning.

[0044] See the appendix Figure 2 and the appendix Figure 3 As shown in FIGS. Figure 3 and , the image acquisition module 2 includes an image acquisition box 21, a high-speed camera 22 and a fill light 23 fixed inside the image acquisition box 21; the image acquisition box 21 is fixed on the sleeper body 1, and the top surface of the image acquisition box 21 is made of a transparent material.

[0045] To further optimize the above technical solution, the image acquisition box 21 includes a box body 211 and a cover body 212. The box body 211 and the cover body 212 are fixedly connected by bolts, and the cover body 212 is provided with special transparent glass 2121.

[0046] To further optimize the above technical solution, a dust-proof wiper is installed on the cover body 212, and the dust-proof wiper is powered by the power module 3.

[0047] The dust-proof wiper provided in this embodiment has the same structure as the existing car wiper, only the size is changed, and its specific structure will not be described in detail here. It should be noted that the dust-proof wiper is started for cleaning according to the recognition effect of the dirt on the captured image. When the dirt recognition rate of the obtained image reaches more than 10%, the dust-proof wiper is started for cleaning.

[0048] To further optimize the above technical solution, the high-speed camera 22 is installed at the center of the box body 211, and the number of fill lights 23 is two, and they are respectively installed on both sides of the high-speed camera 22.

[0049] To further optimize the above technical solution, a power / data transmission interface 2111 is provided on the side wall of the box body 211. The power / data transmission interface 2111 is used to realize the electrical connection between the high-speed camera 22 and the fill light 23 in the box body 211 and the power module 3 and the data management module 4.

[0050] To further optimize the above technical solution, the power module 3 and the data management module 4 are respectively installed in two spaced spaces of the three image acquisition modules 2.

[0051] To further optimize the above technical solution, the image acquisition module 2 uses the SIFT algorithm to extract and match feature points to achieve image stitching.

[0052] In image stitching, the SIFT (Scale-Invariant Feature Transform) algorithm is used to extract and match feature points. The SIFT algorithm is an algorithm for image feature extraction and matching. It has scale invariance, rotation invariance and partial brightness invariance, and is very useful in image stitching.

[0053] 1. Steps of the SIFT algorithm:

[0054] (1) Scale - space Extremum Detection: The SIFT algorithm first detects extrema in images at different scales. For a given image I(x, y), a scale - space L(x, y, σ) is first generated, where σ is the scale parameter. The scale - space can be obtained by Gaussian filtering the image:

[0055] L(x, y, σ) = G(x, y, σ) * I(x, y)

[0056] where G(x, y, σ) is the Gaussian kernel function:

[0057]

[0058] Next, the difference D(x, y, σ) of the scale - space is calculated:

[0059] D(x, y, σ) = L(x, y, kσ) - L(x, y, σ)

[0060] where k is the scale interval. By finding extrema in D(x, y, σ), the key points in the image can be obtained.

[0061] (2) Key - point Localization: After finding the extrema, we use Taylor expansion to accurately locate the position of the key points. For each extremum (x, y, σ), we calculate its gradient and Hessian matrix in D(x, y, σ):

[0062]

[0063] Then, use Taylor expansion to find the accurate position of the extremum:

[0064]

[0065] If the values of x′, y′ and σ′ are very close to the values of x, y and σ, then this extremum is considered a key point.

[0066] (3) Key - point Orientation Assignment: For each key point, we calculate the gradient direction histogram around it and select the maximum value in the histogram as the direction of the key point. This makes the SIFT algorithm rotation - invariant.

[0067] (4) Key - point Description: Finally, we generate a descriptor for each key point. The descriptor is a vector that contains the gradient direction and intensity information around the key point. By comparing the descriptors, we can find the matching key points.

[0068] 2. Innovations and Advantages of the SIFT Algorithm:

[0069] Scale invariance: The SIFT algorithm can detect key points in images of different scales, making it invariant to image scaling.

[0070] Rotation invariance: The SIFT algorithm assigns a direction to each key point, making it invariant to image rotation.

[0071] Partial brightness invariance: The SIFT algorithm uses gradient information to describe key points, making it partially invariant to image brightness changes.

[0072] High matching accuracy: The descriptors generated by the SIFT algorithm are very unique, so key points can be precisely matched.

[0073] 3. Stitching effect:

[0074] Seamless stitching: There are no obvious stitching seams in the stitched image, looking natural.

[0075] Panoramic effect: The stitched image can fully reflect the entire scene, providing a broader view.

[0076] Robustness: Even in the case of scale changes, rotation changes, and illumination changes in the image, the stitching effect is still good.

[0077] See the appendix Figure 4 , the sleeper body 1 is placed between two adjacent railway sleepers 5 and arranged parallel to them. The sleeper body 1 is buried in the ballast, and the burial depth is the same as that of the railway sleeper 5.

[0078] To further optimize the above technical solution, the positions of the three image acquisition modules 2 on the installed sleeper body 1 are respectively located between and outside the two steel rails 6.

[0079] The main functions of the system provided in this embodiment include:

[0080] (1) Vehicle information acquisition function

[0081] It can acquire vehicle number information, automatically measure speed and count axles and vehicles, forming complete multiple-unit vehicle information.

[0082] (2) Linear array image acquisition function

[0083] It can automatically capture images of visible parts such as coupler buffer fittings, suspension systems, braking systems, basic air braking devices, car body frames, bogies, support beams, and steel plates, and splice images of multiple parts as needed for display on the inspection terminal computer. Since the linear array camera is integrated in the sleeper and located under the track, a compensation light source is required to provide a stable light source to ensure imaging quality. The captured pictures are compared with normal exposure pictures, and the light source compensation intensity is adjusted in real time to avoid overexposure and underexposure.

[0084] (3) Fault analysis and early warning

[0085] The terminal of the train detection system uses deep learning algorithms to identify fault information and give alarms, analyze potential faults in the operating state, and implement early warnings.

[0086] The early warning information includes the fault type and the occurrence probability P. When P≥0.7, a fault may have occurred at the bottom of the car body, and a red warning is output; when 0.4≤P<0.7, the fault at the bottom of the car body is in the development stage, and a yellow warning is output; when P<0.4, the fault situation at the bottom of the car body can be ignored, and a green warning is output.

[0087] Main performance of the system:

[0088] (1) Adaptation speed

[0089] The high-speed camera can adapt to train speeds of 0 - 120 km / h.

[0090] (2) Server and storage capacity

[0091] The original digital information of fault detection and fault images are stored in the system for no less than 1 year; other image information is stored for no less than 30 days.

[0092] (3) Axle counting and vehicle counting error

[0093] It can automatically count axles and vehicles. The axle counting error is less than 3x10 -6 , and the vehicle counting error is less than 3x10 -5 .

[0094] (3) Vehicle orientation identification

[0095] It can automatically identify the A and B ends of EMU vehicles, and the identification rate is not less than 99.9%; automatically obtain the tag information of locomotives and vehicles, and automatically extract information such as train number, car number, and vehicle type.

[0096] (4) Identification of key parts

[0097] Accurately locate and capture images of key components at the bottom of the train car body (such as brake beams, bogies, couplers, bolts, nuts, rivets, etc.).

[0098] In another implementation, in order to prevent the influence of train vibration on the acquisition system, a collaborative damping structure is further provided:

[0099] First, add a layer of rubber pads at the bottom of the sleeper body 1 and between the cover 212 and the lower structure, which can absorb the low-frequency vibration caused by the roadbed during train operation.

[0100] Then, the image acquisition module 2 is connected to the sleeper body 1 through an elastic support, forming a "suspended" structure, allowing the image acquisition module 2 to have a small displacement during vibration. In this embodiment, the elastic support can be a silicone shock absorber or an air spring, and at the same time, a rigid limiter is used to prevent excessive deviation.

[0101] A silicone shock absorber is a component that utilizes the elastic properties of silicone material to achieve shock absorption function. Silicone has excellent elasticity, aging resistance, and temperature resistance, and can undergo elastic deformation when subjected to external forces, thereby absorbing and dissipating vibration energy.

[0102] An air spring is a component that utilizes the elastic properties of compressed air to achieve shock absorption and buffering functions. It adjusts the stiffness and shock absorption performance by changing the air pressure inside the airbag.

[0103] Finally, in combination with the acceleration sensor installed inside the image acquisition box 21, the vibration frequency is monitored in real time, and the damping parameters of the suspension system are dynamically adjusted. An image stabilization algorithm is embedded in the data management module 4, and real-time displacement compensation is performed on the image using the acquired acceleration data to eliminate blurring or misalignment phenomena.

[0104] Specifically, the acceleration sensor monitors the vibration and acceleration changes suffered by the image acquisition module during operation in real time. The acceleration sensor can usually measure the acceleration values in three directions (X, Y, Z axes).

[0105] Sampling frequency: The acceleration sensor acquires data at a relatively high frequency (such as 100 Hz or higher) to ensure that high-frequency vibration information can be captured.

[0106] Data output: The sensor transmits the acquired acceleration data to the data management module 4 in the form of digital signals.

[0107] The data management module 4 is the core processing unit of the entire system. It receives the acceleration data and image data, and processes the image through the embedded image stabilization algorithm. The following are the specific implementation steps of the algorithm:

[0108] (1) Preprocessing of acceleration data

[0109] Filtering process: The acceleration data may contain noise and interference signals, so it is necessary to first remove high-frequency noise through a low-pass filter (such as a Butterworth filter) and retain the low-frequency signals related to vibration.

[0110] Data smoothing: Smooth the filtered acceleration data to reduce data fluctuations and improve the accuracy of subsequent processing.

[0111] (2) Calculation of vibration parameters

[0112] Displacement calculation: Based on the acceleration data, the displacement of the image acquisition module 2 in the X, Y, and Z directions is calculated through integral operations. The formula is as follows:

[0113] Δx(t) = ∫∫a x (t)dt 2

[0114] Δy(t) = ∫∫a y (t)dt 2

[0115] Δz(t) = ∫∫a z (t)dt 2

[0116] Where a x (t), a y (t), a z (t) are the accelerations in the X, Y, and Z directions respectively, and Δx(t), Δy(t), Δz(t) are the corresponding displacements.

[0117] Rotation angle calculation: If the rotational vibration of the image acquisition module needs to be considered, the rotation angle can be calculated through the acceleration data. For example, the direction change of the acceleration vector is used to estimate the rotation angle.

[0118] (3) Image displacement compensation

[0119] Image registration: Based on the calculated displacement and rotation angle, the acquired images are subjected to registration processing. The purpose of registration is to adjust the images into a unified reference coordinate system to eliminate the displacement and rotation deviations caused by vibration.

[0120] Interpolation algorithm: During the registration process, an interpolation algorithm (such as bilinear interpolation or bicubic interpolation) is used to fill the pixel gaps generated by displacement compensation to ensure the integrity of the image.

[0121] Real-time processing: The image stabilization algorithm needs to perform real-time processing while the image is being acquired to ensure that the output image is clear, stable, without blurring or misalignment.

[0122] Through displacement compensation, the image blurring phenomenon caused by vibration is reduced, and the clarity of the image is improved. The image registration process eliminates the image misalignment phenomenon caused by vibration to ensure the accuracy of image stitching. The algorithm can perform real-time processing while the image is being acquired without affecting the overall operation efficiency of the system.

[0123] In another embodiment, to improve the adaptability of the high-speed camera, a multi-degree-of-freedom adjustment bracket and an automatic adjustment system are further provided:

[0124] First, a universal ball joint + screw fine-tuning mechanism is adopted to support the pitch angle, horizontal offset, and height adjustment of the camera module, and the position is locked by a dial to adapt to the underbody height and clearance of different vehicle models. In this embodiment, the screw fine-tuning mechanism is a mechanical device that uses the principle of screw drive to achieve precise position adjustment. It converts rotational motion into linear motion by rotating a screw or nut, thereby achieving fine adjustment of the position of an object. The screw fine-tuning mechanism is widely used in occasions that require high-precision positioning, such as optical instruments, precision machinery, measuring equipment, etc. In this embodiment, it is driven by a stepper motor.

[0125] Then, a laser ranging module is integrated into the image acquisition box 21 to real-time feedback the underbody distance to the servo motor, and automatically adjust the lens focal length to ensure clear imaging.

[0126] The horizontal position of the camera module is automatically adjusted by a stepper motor to ensure full-width coverage of the underbody.

[0127] In other embodiments, the shock absorption design can also be combined with the dynamic adjustment system. Through the synergistic effect, the present invention not only effectively reduces the impact of vibration on image acquisition, but also further improves the clarity, stability, and integrity of the image through real-time adjustment and optimization. This synergistic mechanism significantly improves the overall performance of the system.

[0128] The various embodiments in this specification are described in a progressive manner. The key point of each embodiment is to illustrate the differences from other embodiments. The same or similar parts among the various embodiments can be referred to each other. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the description of the method part.

[0129] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but will be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A smart sleeper for train underbody fault detection based on image recognition, characterized in that: include: Sleeper body(1); Image acquisition modules (2), the number of the image acquisition modules (2) being three and being respectively arranged at the middle and both ends of the sleeper body (1), so that the images of the train bottom collected by the three image acquisition modules (2) can cover the entire picture of the train bottom after being spliced; A power module (3), the power module (3) being mounted on the sleeper body (1) and being used to supply power to the image acquisition module (2); A data management module (4), the data management module (4) is installed on the sleeper body (1) and is used to upload the pictures collected by the image collection module (2) to a train detection system terminal.

2. According to claim 1, a smart sleeper for train underbody fault detection based on image recognition is characterized in that: The image acquisition module (2) comprises an image acquisition box (21), and a high-speed camera (22) and a fill light (23) fixed inside the image acquisition box (21); the image acquisition box (21) is fixed on the sleeper body (1), and the top surface of the image acquisition box (21) is made of a transparent material.

3. According to claim 2, a smart sleeper for train underbody fault detection based on image recognition is characterized in that: The image acquisition box (21) comprises a box body (211) and a cover body (212); the box body (211) and the cover body (212) are fastened together by bolts; and the cover body (212) is provided with special transparent glass (2121).

4. According to claim 3, a smart sleeper for train underbody fault detection based on image recognition is characterized in that: A dustproof wiper is installed on the cover body (212), and the dustproof wiper is powered by the power module (3).

5. According to claim 3, a smart sleeper for train underbody fault detection based on image recognition is characterized in that: The high-speed camera (22) is installed at the center of the box body (211), and the number of the fill lights (23) is two, which are respectively installed and arranged on both sides of the high-speed camera (22).

6. According to claim 3, a smart sleeper for train underbody fault detection based on image recognition is characterized in that: A power supply / data transmission interface (2111) is provided on the side wall of the box body (211), and the power supply / data transmission interface (2111) is used to realize electrical connection between the high-speed camera (22) and the fill light (23) in the box body (211) and the power supply module (3) and the data management module (4).

7. According to claim 1, a smart sleeper for train underbody fault detection based on image recognition is characterized in that: The power supply module (3) and the data management module (4) are respectively installed in two compartments of the three image acquisition modules (2).

8. The intelligent sleeper for train underbody fault detection based on image recognition according to claim 1 is characterized in that: The image acquisition module (2) extracts and matches feature points through the SIFT algorithm to achieve image stitching.

9. The intelligent sleeper for train underbody fault detection based on image recognition according to claim 1 is characterized in that: The sleeper body (1) is placed between two adjacent railway sleepers (5) and arranged in parallel therewith; the sleeper body (1) is buried in the crushed stone ballast, and the burying depth is consistent with that of the railway sleepers (5).

10. The intelligent sleeper for train underbody fault detection based on image recognition according to claim 9 is characterized in that: After installation, the three image acquisition modules (2) on the sleeper body (1) are located between and outside the two steel rails (6), respectively.

Images