A measurable intelligent identification method for turnout surface defects

Abstract

An embodiment of the present invention discloses a measurable intelligent identification method for apparent defects of turnouts. A track inspection vehicle is used to obtain top images and oblique images of a base rail and a point rail, and the top image is input into a top width analysis model. The top width analysis model includes a point rail identification module and a top width identification module. The point rail identification module is used to identify the profile of the point rail. The top width identification module determines the position of the point rail of a preset width based on the profile of the point rail. The height difference analysis model includes an image transformation module and a height difference calculation module. The image transformation module transforms the oblique image into a lateral image. The height difference calculation module determines the position of the point rail and the base rail at a preset height difference. The output position data of the top width analysis model and the height difference analysis model are compared. When the two position data are inconsistent, a defect identification result is output.

Description

Technical Field

[0001] The present invention relates to the field of turnout apparent defect recognition, and in particular to a measurable intelligent recognition method for turnout apparent defects. Background Art

[0002] Turnouts are critical branching devices in railway track systems, designed to solve the problem of train path selection in multi-track networks. They are characterized by complex construction, short service life, large number of components, and limitations on train speed. As the weak link in the track, their performance directly affects line capacity, wheel-rail relationship, and maintenance costs.

[0003] The switch rail is a crucial component in the turnout. Its movement directs the train onto the mainline or siding, making it the key to the turnout. When a train passes over the switch rail, significant lateral and longitudinal impact forces are generated. Consequently, the switch rail often suffers from defects such as wear and deformation, poor fit between the switch rail and the base rail, and poor fit between the switch rail and the slide bed. These problems accelerate the wear and tear of the switch rail, leading to excessive dynamic response during train operation and resulting in dynamic irregularities, impacting the safety, stability, and comfort of the train as it passes through the switch.

[0004] Compared with the base rail, the point rail needs to have a certain reduction value. When the train passes through the switch, the reduction value of the tip of the point rail (for example, 23mm) is to ensure that the wheel can gradually increase the contact area with the point rail, thereby smoothly transferring the load. If the tip does not have an appropriate reduction value, the wheel will produce a violent impact at the tip of the point rail; and at the position where the point rail has a certain width (for example, 20mm), the base rail and the point rail jointly bear the vertical load of the wheel. Because the wheel tread has a slope, the point rail should have a certain reduction value (for example, 2mm) at this section. At this width section, the point rail has better anti-deformation ability. If the width of the point rail does not meet expectations, it will bear the wheel load together with the base rail, which will cause the point rail to be overloaded, bringing hidden dangers of point rail disease.

[0005] Existing methods for measuring rail point drop primarily rely on manual contact measurement using feeler gauges, calipers, or simple measuring instruments. This method suffers from low efficiency, poor accuracy, and significant subjectivity. Traditional methods are also susceptible to interference from ambient light and oil stains on the rail surface, making it prone to missing localized wear or cracks. This results in discontinuous data and fails to meet the high-precision, real-time maintenance requirements of high-speed railways. Furthermore, measuring only the rail point drop is insufficient. Without considering the rail's cross-sectional width, it fails to reflect the rail's stress-bearing condition.

[0006] Therefore, designing an accurate, fast and measurable intelligent identification method for turnout surface defects is an urgent problem to be solved by those skilled in the art. Summary of the Invention

[0007] In view of this, an embodiment of the present invention provides a measurable intelligent identification method for apparent defects of turnouts. A track inspection vehicle is used to obtain top images and oblique images of the base rail and the point rail. The position of the point rail of a preset width is determined through the top image. The position of the point rail of a preset height difference is determined by perspective transformation of the oblique image. The position of the point rail of a preset width and the position of the point rail of a preset height difference are compared to determine whether they are consistent, thereby identifying defects. This improves the convenience and accuracy of identifying apparent defects of turnouts and reduces the cost of identification.

[0008] This application discloses a measurable intelligent identification method for apparent defects of turnouts, comprising the following steps:

[0009] Use a camera located just above the stock rail and the switch rail in the track inspection vehicle to obtain top images of the stock rail and the switch rail;

[0010] Oblique images of the stock rail and the point rail are obtained using a camera located obliquely above the stock rail and the point rail in the track inspection vehicle;

[0011] Inputting the top image into a top width analysis model, the top width analysis model comprising a point rail recognition module and a top width recognition module, the point rail recognition module being used to recognize the point rail profile, and the top width recognition module determining the position of the point rail of a preset width based on the point rail profile;

[0012] Inputting the oblique image into a height difference analysis model, the height difference analysis model comprising an image transformation module and a height difference calculation module, the image transformation module transforming the oblique image into a horizontal lateral image, the height difference calculation module being configured to calculate the height difference between the point rail and the base rail based on the lateral image, and determining the positions of the point rail and the base rail at a preset height difference;

[0013] The output position data of the top width analysis model and the output position data of the height difference analysis model are compared, and when the two position data are inconsistent, a disease identification result is output.

[0014] Preferably, the point rail recognition module includes a feature extraction unit and a contour prediction unit. The feature extraction unit adopts a U-Net structure. The output features of the encoders in each layer are transmitted to the decoders in the same layer and across layers to the decoders in different layers. The decoder sequentially passes the input features through an edge-aware attention module, a transposed convolution layer, a bilinear interpolation layer, and a refined convolution layer to output contour features. The edge-aware attention module performs Sobel edge extraction on the features from the decoder in the previous layer to output edge features, performs content feature extraction on the features from the encoder to output content features, calculates attention weights based on the edge features and the content features, and performs feature weighting based on the attention weights to output enhanced features.

[0015] Preferably, the contour feature is input to the contour prediction unit, and the contour prediction unit includes a parallel semantic segmentation path and an edge enhancement path. The contour feature passes through the semantic segmentation path and the edge enhancement path respectively and then is fused to output the point rail contour. The contour feature passes through multiple convolution operations in stages in the semantic segmentation path and then outputs the point rail contour probability. The contour feature passes through depthwise separable convolution, residual block, and depthwise separable convolution in the edge enhancement path and then outputs a distance field map.

[0016] Preferably, the top width identification module includes a contour parameterization unit, a width calculation unit, and a width matching unit. The contour parameterization unit is used to convert the point rail contour into a point rail contour point sequence. The width calculation unit determines the width array of each contour point based on the point rail contour point sequence. The width matching unit determines the contour point of the preset width based on the width array.

[0017] Preferably, the profile parameterization unit establishes a coordinate system with the tip of the point rail as the origin and the direction of the point rail as the x-axis, determines the coordinates of each point of the point rail profile, and establishes a point rail profile point sequence. The width calculation unit calculates the normal direction point by point along the point rail profile point sequence, projects a ray to the side of the point rail close to the base rail, determines the intersection of the ray and the side, calculates the Euclidean distance between each point in the point rail profile point sequence and the intersection, and establishes a width array. The width matching unit searches the width array for a profile point closest to a preset width.

[0018] Preferably, the image transformation module includes a track feature detection unit, a coordinate system construction unit, a perspective transformation unit, and a remapping unit, wherein the track feature detection unit detects the tip of the switch rail and the working surface of the stock rail, and determines the feature points of the switch rail and the stock rail;

[0019] The coordinate system construction unit uses the tip of the point rail as the origin of the coordinate system, uses the direction of the fitted straight line parallel to the working surface of the base rail as the x-axis, and uses the direction perpendicular to the x-axis as the y-axis to establish the coordinate system;

[0020] The perspective transformation unit constructs and solves a perspective transformation matrix by matching feature points in the oblique image with feature points in a reference image;

[0021] The remapping unit compensates for the longitudinal offset of the image caused by vehicle vibration by using the pitch angle deviation value measured in real time by the camera, corrects the perspective transformation matrix, and uses the corrected perspective transformation matrix to map any point in the oblique image to the corresponding position in the lateral image, thereby realizing the conversion of the image perspective to obtain the lateral image.

[0022] Preferably, the height difference calculation module includes an edge extraction unit, a reference plane unit, a height difference calculation unit, and a positioning unit. The edge extraction unit detects the rail top edge point set of the sharp rail and the base rail based on the lateral image. The reference plane unit models the rail tops of the sharp rail and the base rail based on the rail top edge point set of the sharp rail and the base rail, and determines the edge point sequence of the sharp rail and the base rail. The height difference calculation unit calculates the distance between the sharp rail and the base rail point by point in the vertical direction based on the edge point sequence of the sharp rail and the base rail to form a height difference array. The positioning unit searches for the contour point closest to the preset height difference in the height difference array.

[0023] Preferably, the horizontal distance between the contour point output by the top width analysis model and the tip of the point rail is determined, and the horizontal distance between the contour point output by the height difference analysis model and the tip of the point rail is determined. The two horizontal distances are compared, and if the deviation is greater than a preset threshold, the defect identification result is output.

[0024] This application proposes a measurable intelligent identification method for apparent defects of turnouts, which uses a track inspection vehicle to obtain images of the point rail, does not rely on manual measurement, reduces detection costs, and improves recognition accuracy; measures the height difference after perspective transformation of the oblique image, which solves the limitation of the track inspection vehicle that cannot directly obtain lateral images, and improves the applicability of the track inspection vehicle in point rail detection scenarios; through the design of the point rail recognition module, the outline of the point rail is accurately determined, providing a solid data basis for the measurement of the point rail width; the recognition and comparison of the preset width position and the preset height difference position realizes the joint detection of the point rail reduction value and the point rail width, and more accurately discovers the hidden dangers of point rail defects. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The above and other objects, features and advantages of the present invention will become more apparent through the following description of the embodiments of the present invention with reference to the accompanying drawings, in which:

[0026] Figure 1 This is a flow chart of a method for intelligently identifying measurable apparent defects of turnouts provided by an exemplary embodiment of the present invention;

[0027] Figure 2 is a schematic structural diagram of a top width analysis model provided by an exemplary embodiment of the present invention;

[0028] Figure 3 It is a structural diagram of a height difference analysis model provided by an exemplary embodiment of the present invention. DETAILED DESCRIPTION

[0029] The present application is described below based on the following embodiments, but the present application is not limited to these embodiments. In the detailed description of the present application below, certain specific details are described in detail. Those skilled in the art can fully understand the present application without the description of these details. To avoid obscuring the essence of the present application, well-known methods, processes, procedures, components, and circuits are not described in detail.

[0030] Furthermore, persons of ordinary skill in the art will appreciate that the figures provided herein are for illustration purposes only and are not necessarily drawn to scale.

[0031] Unless the context clearly requires otherwise, words like “include”, “comprising” and the like throughout this application should be interpreted as including rather than exclusive or exhaustive; that is, as meaning “including but not limited to”.

[0032] In the description of this application, it should be understood that the terms "first", "second", etc. are used for descriptive purposes only and should not be understood to indicate or imply relative importance. In addition, in the description of this application, unless otherwise specified, "plurality" means two or more.

[0033] This embodiment provides a measurable intelligent identification method for turnout surface defects, such as Figure 1 As shown, the method provided in this embodiment includes the following steps:

[0034] Step 1: Use the camera located directly above the stock rail and the switch rail in the track inspection vehicle to obtain the top image of the stock rail and the switch rail.

[0035] Existing methods for measuring the drop of the point rail are typically performed manually using a dedicated measuring instrument, using high-precision sensors for contact measurement, or using a point rail drop gauge to measure the distances of various sections of the point rail. This still relies primarily on manual on-site measurement, which requires significant additional labor costs and measurement work. Furthermore, since the point rail is attached to the inside of the base rail, installing image acquisition equipment or laser measurement equipment in the middle of the track will inevitably interfere with train operation, while installing equipment next to the track will be affected by passing trains. Therefore, using a track inspection vehicle to capture images of the point rail and using this image data as the data basis for defect identification is an efficient and inexpensive method that avoids the tedious and high cost of manual contact measurement and also improves identification accuracy.

[0036] Oblique images of the stock rail and the point rail are obtained using a camera located obliquely above the stock rail and the point rail in the track inspection vehicle.

[0037] The cameras on a track inspection vehicle are typically located directly beneath the vehicle floor, capturing vertical images of the rail surface to detect planar defects such as wear and cracks. They are also located on both sides of the vehicle floor, obliquely aiming at the rail working surface and rail waist, with a shooting angle of 30 to 45 degrees, allowing for simultaneous consideration of both the rail head profile and the rail base. To measure the drop of the point rail, the optimal image is a horizontal profile of the point and base rail. However, to improve the applicability of the track inspection vehicle, oblique images are used directly as the data basis for point rail measurement, and subsequent algorithm design ensures accurate measurement.

[0038] Step 2: Input the top image into the top width analysis model 100. The top width analysis model 100 includes a point rail recognition module 110 and a top width recognition module 120. The point rail recognition module 110 is used to identify the point rail profile, and the top width recognition module 120 determines the point rail position of a preset width based on the point rail profile.

[0039] Step 3: Input the oblique image into the height difference analysis model 200. The height difference analysis model 200 includes an image transformation module 210 and a height difference calculation module 220. The image transformation module 210 transforms the oblique image into a horizontal lateral image. The height difference calculation module 220 is used to calculate the height difference between the point rail and the base rail based on the lateral image and determine the positions of the point rail and the base rail at a preset height difference.

[0040] Step 4: Compare the output position data of the top width analysis model and the height difference analysis model. When the two position data are inconsistent, output the disease identification result.

[0041] In step 2, if Figure 2 As shown, the point rail recognition module 110 includes a feature extraction unit 111 and a contour prediction unit 112. The feature extraction unit 111 adopts a U-Net structure, and the output features of the encoders at each layer are transmitted to the decoders at the same layer and across layers to decoders at different layers.

[0042] U-Net is a classic convolutional neural network architecture that effectively extracts image features through the collaborative work of an encoder and decoder. In the U-Net structure, each encoder layer outputs a feature map. These feature maps not only contain the features extracted by the current layer but also retain certain local information. These feature maps are directly transmitted to the corresponding layers in the decoder. For example, the feature map output by the first layer of the encoder is transmitted to the first layer of the decoder, the feature map output by the second layer of the encoder is transmitted to the second layer of the decoder, and so on.

[0043] In addition to same-layer transmission, the U-Net structure in this embodiment also transmits the encoder's feature maps across layers to different layers of the decoder. The role of the skip connection is to directly transmit the features extracted from the encoder to deeper layers in the decoder, thereby enhancing the transmission and fusion of features. For example, the feature map output by the first layer of the encoder may be transmitted to the second or third layer of the decoder, and the feature map output by the second layer of the encoder may be transmitted to the third or fourth layer of the decoder, and so on. Because feature maps at different levels contain feature information at different scales, cross-layer transmission can fuse features of different scales together, thereby better capturing the multi-scale information in the image.

[0044] The decoder outputs contour features after inputting features through an edge-aware attention module, a transposed convolution layer, a bilinear interpolation layer, and a refined convolution layer in sequence. The edge-aware attention module performs Sobel edge extraction on the features from the previous decoder layer to output edge features, and performs content feature extraction on the features from the encoder to output content features. The attention weight is calculated based on the edge features and the content features, and the feature weighting is performed based on the attention weight to output enhanced features.

[0045] Among them, the edge-aware attention module uses the Sobel operator to extract the edge features output by the previous layer decoder as the basis for attention generation, while capturing features of different coarseness from the encoder outputs of the same layer and across layers, and automatically adjusts the attention weight according to the feature content, thereby enhancing the model's perception of the track contour.

[0046] The transposed convolution layer uses the minimized convolution kernel (3×3) for transposition, and performs bilinear interpolation after the transposed convolution instead of the traditional pre-transposition. The purpose of this design is to make the transposed convolution establish semantic rationality first, and bilinear interpolation is responsible for geometric smoothness. Finally, the refinement convolution layer uses 1×1 convolution to correct the local distortion missed in the first two stages and output contour features.

[0047] The contour features are input to the contour prediction unit 112, which includes a parallel semantic segmentation path and an edge enhancement path. The contour features are fused after passing through the semantic segmentation path and the edge enhancement path to output the point rail contour. The contour features are sequentially subjected to multiple staged convolution operations in the semantic segmentation path to output the point rail contour probability. The contour features are sequentially subjected to depthwise separable convolution, residual block, and depthwise separable convolution in the edge enhancement path to output a distance field map.

[0048] Among them, the main function of the semantic segmentation path is to generate a contour probability map, that is, to predict the probability of each pixel belonging to a contour, and gradually restore the spatial resolution of the feature map in stages to make it consistent with the resolution of the input image, providing semantic-level contour information for subsequent contour synthesis. At the same time, the staged convolution operation learns different features and patterns in each stage, thereby enhancing the generalization ability of the model. By gradually extracting and refining features, the model can better adapt to the input data and task requirements.

[0049] The semantic segmentation path consists of multiple convolutional operations, followed by a transposed convolution, a 3×3 convolution and batch normalization, a transposed convolution, a 3×3 convolution and batch normalization, and finally an output of the track outline probability. The transposed convolution is used to increase the spatial size of the feature map while reducing the number of channels. This helps restore the spatial resolution of the feature map and reduces the computational burden. The 3×3 convolution further processes the feature map, adding nonlinear features, while batch normalization stabilizes the training process.

[0050] The edge enhancement pathway generates sub-pixel edges by extracting edge information through a distance field regression network. This pathway focuses on accurate edge extraction and enhancement, providing high-precision edge information for subsequent contour synthesis.

[0051] The edge enhancement path first uses depthwise separable convolution to perform depthwise separable convolution on the input feature map, reducing the amount of computation while extracting edge features. Residual learning is then introduced to enhance feature extraction and avoid the vanishing gradient problem. Finally, the feature map is further processed using depthwise separable convolution to generate a distance field map. The value of each pixel in the distance field map represents the distance from the pixel to the nearest edge, as shown below:

[0052]

[0053] d is the pixel distance to the nearest edge.

[0054] Ultimately, the semantic segmentation path provides semantic-level contour information, that is, the probability that each pixel belongs to a contour, and the edge enhancement path provides high-precision edge information, that is, the distance from each pixel to the nearest edge or the probability of the edge. By fusing the outputs of these two paths, we can fully utilize the semantic information and edge information to generate more accurate and robust contour prediction results.

[0055] In step 2, the top width identification module 120 is used to accurately locate the cross-section position where the top width of the switch rail reaches a preset value (such as 20 mm). Figure 3As shown, the top width identification module 120 includes a contour parameterization unit 121, a width calculation unit 122, and a width matching unit 123. The contour parameterization unit 121 is used to convert the point rail contour into a point rail contour point sequence, the width calculation unit 122 determines the width array of each contour point according to the point rail contour point sequence, and the width matching unit 123 determines the contour point of the preset width according to the width array.

[0056] The profile parameterization unit 121 establishes a coordinate system with the tip of the sharp rail as the origin and the direction of the sharp rail as the X-axis, determines the coordinates of each point of the sharp rail profile, and establishes a sharp rail profile point sequence. The width calculation unit 122 calculates the normal direction point by point along the sharp rail profile point sequence, projects a ray to the side of the sharp rail close to the base rail, determines the intersection of the ray and the side, calculates the Euclidean distance between each point in the sharp rail profile point sequence and the intersection, and establishes a width array. The width matching unit 123 searches the width array for the profile point closest to the preset width.

[0057] In step 3, the image transformation module 210 uses pure visual geometric transformation technology to convert the side image of the turnout taken at an oblique angle into a standard horizontal perspective, eliminate perspective distortion, and provide standardized input for height difference calculation.

[0058] Image transformation module 210 includes a track feature detection unit 211, a coordinate system construction unit 212, a perspective transformation unit 213, and a remapping unit 214. Track feature detection unit 211 uses a depthwise separable convolutional network to detect the tip of the switch rail and the working surface of the base rail, and determines the feature points of the switch rail and base rail. Feature points of the switch rail and base rail are unique and stable points in the switch rail image, and can include the tip of the switch rail, the corner point of the slide table, and the inner point of the base rail. By finding corresponding feature points in the source and target images, the geometric transformation relationship between the two images can be determined. The correspondence between feature point pairs is the basis for calculating the perspective transformation matrix.

[0059] The coordinate system construction unit 212 establishes a coordinate system with the tip of the point rail as the origin of the coordinate system, a direction parallel to the fitting line of the base rail working surface as the X-axis, and a direction perpendicular to the X-axis as the Y-axis.

[0060] The perspective transformation unit 213 constructs and solves a perspective transformation matrix by matching feature points in the oblique image with feature points in a reference image.

[0061] Among them, the oblique image is the source image, and the reference image is the side reference image. Feature points are detected in the oblique image, and multiple matching feature point pairs between the reference images are found. Using the multiple matching feature point pairs, the perspective transformation matrix of each image relative to the reference image is calculated, as shown below.

[0062]

[0063] Among them, (x2, y2) is the coordinate of the feature point in the reference image, (x1, y1) is the coordinate of the feature point in the source image, and H is the perspective transformation matrix.

[0064] The remapping unit 214 compensates for the longitudinal offset of the image caused by vehicle vibration by using the pitch angle deviation value measured in real time by the camera, corrects the perspective transformation matrix, and uses the corrected perspective transformation matrix to map any point in the oblique image to the corresponding position in the lateral image, thereby realizing the conversion of the image perspective to obtain the lateral image.

[0065] The remapping unit 214 is responsible for fine-tuning the image after perspective transformation to ensure that the output image meets the requirements of high-precision height difference calculation.

[0066] In order to eliminate the longitudinal offset of the image caused by vehicle vibration, the remapping unit 214 obtains the pitch angle deviation Δθ (unit: degree) measured in real time by the camera and corrects the third row parameters of the perspective transformation matrix:

[0067]

[0068] Among them, k is the calibration coefficient, Δθ is the pitch angle deviation of the camera, and h 31 and h 32 are the first two elements of the third row of the original perspective transformation matrix, h' 31 and h' 32 These are the first two elements of the third row of the corrected perspective transformation matrix.

[0069] h 31 and h 32 These are the first two elements of the third row of the perspective transformation matrix, which affect the scale and position of the point by adjusting h 31 and h 32 You can control the perspective effect of the image.

[0070] Finally, the corrected perspective transformation matrix is ​​used to map any point in the oblique image to the corresponding point in the side image, thereby obtaining the transformed side image, as shown below.

[0071]

[0072] Among them, (x i ,y i ) are the coordinates of each point in the oblique image, (x j ,y j ) are the coordinates of each point in the transformed lateral image, is the corrected perspective transformation matrix, and ω is the scaling factor.

[0073] In step 3, the height difference calculation module 220 includes an edge extraction unit 221, a reference plane unit 222, a height difference calculation unit 223, and a positioning unit 224. The edge extraction unit 221 detects the rail top edge point set of the point rail and the base rail based on the lateral image. The reference plane unit 222 models the rail tops of the point rail and the base rail based on the rail top edge point set of the point rail and the base rail, and determines the edge point sequence of the point rail and the base rail. The height difference calculation unit 223 calculates the distance between the point rail and the base rail point by point in the vertical direction based on the edge point sequence of the point rail and the base rail to form a height difference array. The positioning unit 224 searches the height difference array for the contour point closest to the preset height difference.

[0074] Among them, the reference plane unit 222 determines the working surface of the point rail and the base rail, which is the basis for height difference calculation. By accurately modeling the top shape of the base rail, a spatial reference system is provided for the height difference calculation of the point rail. Through modeling, the influence of interference objects such as fasteners and rust spots can be eliminated, and the smooth constraint of the rail surface geometry can be maintained.

[0075] The reference plane unit 222 models the top of the point rail and stock rail based on their top edge point sets, including both base modeling and point rail modeling. The stock rail modeling includes: 1. Selecting edge points from the stock rail working surface to the rail head centerline; 2. Using a cubic polynomial model, as shown below:

[0076]

[0077] in, is the surface height of the base track at position x, x is the position coordinate along the base track, a0, a1, a2, a3 are the coefficients of the polynomial, which are used to define the shape of the polynomial, and ε is the error term, which represents the difference between the model prediction value and the actual value;

[0078] 3. Perform robust fitting using the RANSAC algorithm, output the fitting coefficients a0, a1, a2, a3, obtain a cubic polynomial model, and determine the edge point sequence of the basic track;

[0079] The switch rail modeling uses piecewise spline interpolation fitting, including: 1. Setting a control point every 10 mm along the switch rail working surface at the edge of the switch rail, and setting a control point every 2 mm at the tip of the switch rail and in the switch area; 2. Setting a curve function for the spline between two adjacent control points, as shown below:

[0080]

[0081] Among them, S i(x) is the height of the curved surface of the switch rail at the x position between the i-th to the i+1-th control point, x is the position coordinate along the switch rail direction, i is the serial number of the control point, a i , b i , c i , d i are the coefficients of the spline function;

[0082] 3. In order to avoid excessive curvature changes in the curve, maintain the smoothness and continuity of the curve, limit the curvature change rate, and ensure that the absolute value of the curvature change rate is less than 0.01mm -2 , as shown below:

[0083]

[0084] in, is the curvature of each spline segment;

[0085] 4. Use the least squares fitting algorithm to construct the objective function as shown below:

[0086]

[0087] Where m is the number of control points, is the data fitting term, which represents the relationship between the spline curve s(x) and the control point (x j ,y j ), is a regularization term used to control the smoothness of the spline curve, and λ is a smoothing factor;

[0088] Solve the objective function and determine the coefficient a of the spline curve when the objective function takes the minimum value i , b i , c i , d i , obtain various spline curve functions and determine the edge point sequence of the basic track.

[0089] In step 4, the horizontal distance between the contour point output by the top width analysis model 100 and the tip of the point rail is determined, and the horizontal distance between the contour point output by the height difference analysis model 200 and the tip of the point rail is determined. The two horizontal distances are compared. If the deviation is greater than a preset threshold, the defect identification result is output.

[0090] Among them, the position data output by the top width analysis model 100 is the coordinate position of the coordinate system established in the top surface image, and the position data output by the height difference analysis model 200 is the coordinate position of the coordinate system established in the side image after perspective transformation. The coordinate systems of the two are different, but both coordinate systems use the tip of the point rail as the origin. Then, the x-axis coordinates of the two coordinate positions can represent the distance between the two contour points and the tip of the point rail. On the point rail, whether in the top surface perspective or the side perspective, the same distance from the tip of the point rail represents the same cross-sectional position. Therefore, in step 4, the x-axis coordinates of the contour points output by the top width analysis model 100 are compared with the x-axis coordinates of the contour points output by the height difference analysis model 200. If the deviation between the two is greater than the preset threshold, it indicates that the point rail position of the preset width is inconsistent with the point rail position of the preset height difference, which means that the point rail does not bear the required wheel load at the position of the predetermined width, and there is a hidden danger of disease. At this time, it is necessary to output the disease identification result.

[0091] The foregoing is merely a preferred embodiment of the present application and is not intended to limit the present application. Persons skilled in the art will readily appreciate that various modifications and variations are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present application are intended to be within the scope of protection of the present application.

Claims

1. A measurable intelligent identification method for apparent defects of turnouts, characterized by: The following steps are involved: Use a camera located just above the stock rail and the switch rail in the track inspection vehicle to obtain top images of the stock rail and the switch rail; Oblique images of the stock rail and the point rail are obtained using a camera located obliquely above the stock rail and the point rail in the track inspection vehicle; Inputting the top image into a top width analysis model, the top width analysis model comprising a point rail recognition module and a top width recognition module, the point rail recognition module being used to recognize the point rail profile, and the top width recognition module determining the position of the point rail of a preset width based on the point rail profile; Inputting the oblique image into a height difference analysis model, the height difference analysis model comprising an image transformation module and a height difference calculation module, the image transformation module transforming the oblique image into a horizontal lateral image, the height difference calculation module being configured to calculate the height difference between the point rail and the base rail based on the lateral image, and determining the positions of the point rail and the base rail at a preset height difference; Comparing the output position data of the top width analysis model and the output position data of the height difference analysis model, and outputting a disease identification result when the two position data are inconsistent; The top width recognition module includes a profile parameterization unit, a width calculation unit, and a width matching unit. The profile parameterization unit is used to convert the point rail profile into a point rail profile point sequence. The width calculation unit determines a width array of each profile point based on the point rail profile point sequence. The width matching unit determines a profile point of a preset width based on the width array. The profile parameterization unit establishes a coordinate system with the tip of the switch rail as the origin and the direction of the switch rail as the x-axis, determines the coordinates of each point of the switch rail profile, and establishes a switch rail profile point sequence. The width calculation unit calculates the normal direction point by point along the switch rail profile point sequence, projects a ray to the side of the switch rail close to the base rail, determines the intersection of the ray and the side, calculates the Euclidean distance between each point in the switch rail profile point sequence and the intersection, and establishes a width array. The width matching unit searches the width array for a profile point that is closest to a preset width.

2. The method according to claim 1, characterized in that The point rail recognition module includes a feature extraction unit and a contour prediction unit. The feature extraction unit adopts a U-Net structure. The output features of the encoders in each layer are transmitted to the decoders in the same layer and across layers to the decoders in different layers. The decoder sequentially passes the input features through an edge-aware attention module, a transposed convolution layer, a bilinear interpolation layer, and a refined convolution layer, and then outputs contour features. The edge-aware attention module performs Sobel edge extraction on the features from the decoder in the previous layer and outputs edge features. It performs content feature extraction on the features from the encoder and outputs content features. The attention weight is calculated based on the edge features and the content features, and the features are weighted according to the attention weight to output enhanced features.

3. The method according to claim 2, characterized in that The contour features are input to the contour prediction unit, which includes a parallel semantic segmentation path and an edge enhancement path. The contour features are fused after passing through the semantic segmentation path and the edge enhancement path to output the point rail contour. The contour features are sequentially subjected to multiple staged convolution operations in the semantic segmentation path to output the point rail contour probability. The contour features are sequentially subjected to depthwise separable convolution, residual blocks, and depthwise separable convolution in the edge enhancement path to output a distance field map.

4. The method according to claim 1, wherein The image transformation module includes a track feature detection unit, a coordinate system construction unit, a perspective transformation unit, and a remapping unit. The track feature detection unit detects the tip of the switch rail and the working surface of the stock rail and determines the feature points of the switch rail and the stock rail. The coordinate system construction unit uses the tip of the point rail as the origin of the coordinate system, uses the direction of the fitted straight line parallel to the working surface of the base rail as the x-axis, and uses the direction perpendicular to the x-axis as the y-axis to establish the coordinate system; The perspective transformation unit constructs and solves a perspective transformation matrix by matching feature points in the oblique image with feature points in a reference image; The remapping unit compensates for the longitudinal offset of the image caused by vehicle vibration by using the pitch angle deviation value measured in real time by the camera, corrects the perspective transformation matrix, and uses the corrected perspective transformation matrix to map any point in the oblique image to the corresponding position in the lateral image, thereby realizing the conversion of the image perspective to obtain the lateral image.

5. The method according to claim 1, wherein The height difference calculation module includes an edge extraction unit, a reference plane unit, a height difference calculation unit, and a positioning unit. The edge extraction unit detects a rail top edge point set of the point rail and the base rail based on the lateral image. The reference plane unit models the rail tops of the point rail and the base rail based on the rail top edge point set of the point rail and the base rail, and determines an edge point sequence of the point rail and the base rail. The height difference calculation unit calculates the distance between the point rail and the base rail point by point in the vertical direction based on the edge point sequence of the point rail and the base rail to form a height difference array. The positioning unit searches the height difference array for a contour point closest to a preset height difference.

6. The method according to claim 1, characterized in that Determine the horizontal distance between the contour point output by the top width analysis model and the tip of the point rail, determine the horizontal distance between the contour point output by the height difference analysis model and the tip of the point rail, compare the two horizontal distances, and output the defect identification result if the deviation is greater than a preset threshold.

Images