Turnout steel rail light band identification method, system and equipment and storage medium

Abstract

The invention provides a turnout steel rail light band recognition method, system and device and a storage medium, and relates to the technical field of intelligent detection and image recognition, and the method comprises the steps: obtaining first information which comprises turnout steel rail light band image data, turnout steel rail geometric morphology data and ultrasonic detection data; sending the first information to a preset de-noising model for de-noising processing to obtain de-noised first information; sending the de-noised first information to a preset feature extraction module for feature extraction to obtain feature data of the turnout steel rail light band; and sending the feature data of the turnout steel rail light band to a preset target recognition model for target recognition, and performing classification and anomaly judgment on the light band based on a result obtained by target recognition to obtain a final recognition result of the light band. According to the invention, the precision and robustness of light band identification are improved, and real-time and intelligent monitoring and early warning can be realized in a complex environment.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent detection and image recognition, and more specifically, to a method, system, device, and storage medium for identifying turnout rail light bands. Background Art

[0002] In the field of high-speed railways, the contact state of turnout rails directly affects the smoothness and safety of train operation, and the turnout rail light band is an important indicator for measuring the wheel-rail contact state. By analyzing the integrity, shape, and dynamic changes of the light band, the wear condition, contact uniformity, and potential fault risks of the rail can be judged. At present, the detection of the light band mainly relies on manual inspections and traditional image processing methods. The manual inspection method relies on maintenance personnel to visually inspect or use handheld devices for detection, which has a large degree of subjectivity. The detection accuracy is affected by factors such as environmental lighting and personnel experience. At the same time, the cycle is long and the efficiency is low, making it difficult to meet the real-time monitoring requirements of high-speed rail operation. For traditional image processing methods, such as those based on edge detection and threshold segmentation, although light band recognition can be achieved under specific environments, it is easily interfered under complex working conditions, resulting in large detection errors. In addition, existing methods are mostly based on single-modal data and cannot effectively combine light band images, geometric shape data, and ultrasonic detection data, making it difficult to accurately distinguish normal wear, abnormal wear, and contact abnormalities. At the same time, traditional methods lack the ability to model the change of the light band over time and cannot predict the evolution trend of the light band, making it easy for potential abnormalities to go undetected in a timely manner.

[0003] Therefore, the existing technology has problems such as low detection efficiency, insufficient recognition accuracy, low data utilization rate, and inability to analyze the dynamic changes of the light band in light band detection, making it difficult to meet the safety monitoring requirements of high-speed rail turnout rails. Summary of the Invention

[0004] The purpose of the present invention is to provide a method, system, device, and storage medium for identifying turnout rail light bands to improve the above problems. To achieve the above purpose, the technical solutions adopted by the present invention are as follows:

[0005] In a first aspect, the present application provides a method for identifying a turnout rail light band, including:

[0006] Obtaining first information, where the first information includes turnout rail light band image data, turnout rail geometric shape data, and ultrasonic detection data;

[0007] Sending the first information to a preset denoising model for denoising processing to obtain denoised first information;

[0008] Sending the denoised first information to a preset feature extraction module for feature extraction to obtain feature data of the turnout rail light band;

[0009] Send the characteristic data of the turnout rail light band to a preset target recognition model for target recognition, and classify and perform anomaly judgment on the light band based on the result obtained from the target recognition to obtain the final recognition result of the light band.

[0010] In a second aspect, the present application also provides an identification system for a turnout rail light band, including:

[0011] An acquisition unit, configured to acquire first information, where the first information includes turnout rail light band image data, turnout rail geometric shape data, and ultrasonic detection data;

[0012] A denoising unit, configured to send the first information to a preset denoising model for denoising processing to obtain the denoised first information;

[0013] An extraction unit, configured to send the denoised first information to a preset feature extraction module for feature extraction to obtain the characteristic data of the turnout rail light band;

[0014] An identification unit, configured to send the characteristic data of the turnout rail light band to a preset target recognition model for target recognition, and classify and perform anomaly judgment on the light band based on the result obtained from the target recognition to obtain the final recognition result of the light band.

[0015] In a third aspect, the present application also provides an identification device for a turnout rail light band, including:

[0016] A memory, configured to store a computer program;

[0017] A processor, configured to implement the steps of the identification method for the turnout rail light band when executing the computer program.

[0018] In a fourth aspect, the present application also provides a storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the above-mentioned identification method based on the turnout rail light band are implemented.

[0019] The beneficial effects of the present invention are:

[0020] By integrating the light band image, geometric shape data, and ultrasonic detection data of the turnout rail, this method can obtain the contact state information of the rail from multiple dimensions, providing a more comprehensive and accurate analysis basis. In the data preprocessing stage, advanced algorithms such as wavelet transform and multi-resolution reconstruction are used to remove noise and enhance data features, effectively improving the data quality and signal-to-noise ratio. The feature extraction module extracts key features from the light band image through techniques such as adaptive edge detection, Hough transform, and gradient direction analysis, enhancing the accurate recognition ability of the light band area. The target recognition module combines YOLOv8 and autoencoders with time series analysis, which can not only accurately classify the integrity of the light band but also predict the dynamic change trend of the light band based on historical data, and detect potential anomalies in real time. Compared with traditional methods, the combination of multi-modal data fusion and deep learning models in this application improves the accuracy and robustness of light band recognition, and can achieve real-time and intelligent monitoring and early warning in complex environments.

[0021] Other features and advantages of the present invention will be described in the subsequent specification, and part of them will become obvious from the specification, or can be understood by implementing the embodiments of the present invention. The objectives and other advantages of the present invention can be realized and obtained by the structures specifically pointed out in the written specification, claims, and drawings. Brief Description of the Drawings

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the embodiments. It should be understood that the following drawings only show some embodiments of the present invention, so they should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can be obtained based on these drawings without creative efforts.

[0023] Figure 1 It is a schematic flow chart of the method for identifying the light band of the turnout rail described in the embodiments of the present invention;

[0024] Figure 2 It is a schematic structural diagram of the system for identifying the light band of the turnout rail described in the embodiments of the present invention;

[0025] Figure 3 It is a schematic structural diagram of the device for identifying the light band of the turnout rail described in the embodiments of the present invention.

[0026] In the figure: 701, acquisition unit; 702, denoising unit; 703, extraction unit; 704, recognition unit; 7021, first denoising sub-unit; 7022, second denoising sub-unit; 7023, third denoising sub-unit; 7031, first extraction sub-unit; 7032, second extraction sub-unit; 7033, third extraction sub-unit; 7041, first recognition sub-unit; 7042, second recognition sub-unit; 7043, third recognition sub-unit; 7044, fourth recognition sub-unit; 7045, fifth recognition sub-unit; 800, marking receiving device; 801, processor; 802, memory; 803, multimedia component; 804, input / output (I / O) interface; 805, communication component. Detailed implementation

[0027] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and illustrated herein generally may be arranged and designed in a variety of different configurations. Therefore, the detailed description of the embodiments of the present invention provided herein is not intended to limit the scope of the claimed invention, but is merely representative of selected embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0028] It should be noted that like reference numerals and letters denote like items in the following drawings. Therefore, once an item is defined in one drawing, it does not require further definition and explanation in subsequent drawings. At the same time, in the description of the present invention, the terms "first", "second", etc. are only used for descriptive distinction and cannot be construed as indicating or implying relative importance.

[0029] Embodiment 1:

[0030] This embodiment provides a method for recognizing the light band of turnout rails.

[0031] See Figure 1 , the figure shows that this method includes steps S1, S2, S3 and S4.

[0032] Step S1, acquire first information, where the first information includes turnout rail light band image data, turnout rail geometric shape data, and ultrasonic detection data;

[0033] It is understandable that in this step, the turnout rail light band image data is acquired through a high-speed camera or a laser imaging device, which can intuitively reflect the contact state of the turnout rail surface, especially the light band morphology, and display surface defects such as track wear and cracks. In the high-speed rail track system, the integrity of the light band directly affects the smoothness and safety of train operation. Therefore, the accurate acquisition of light band images is of crucial importance.

[0034] The turnout rail geometric shape data is obtained through laser scanning or high-precision measurement devices, recording structural information such as the geometric dimensions, angles, and curvature of the track. These data help to understand the impact of track shape changes on the rail contact state. Especially in complex turnout areas, geometric shape data is very important for identifying potential problems. The ultrasonic detection data is acquired through ultrasonic sensors. These data can detect the internal structure of the turnout rail and reveal hidden defects such as cracks, air bubbles, and delaminations. Compared with image data, ultrasonic data provides deep structural information and plays an irreplaceable role in accurately evaluating the health state of the turnout rail. By comprehensively collecting these three types of data, it can provide multi-modal input for subsequent data processing and analysis, and provide comprehensive support for rail state monitoring and anomaly detection.

[0035] Step S2: Send the first information to a preset denoising model for denoising processing to obtain the denoised first information;

[0036] It is understandable that in this step, after the first information is processed by the denoising model, its signal-to-noise ratio is significantly improved, and the noise interference is greatly reduced, thus providing a clean and reliable data input for subsequent feature extraction and target recognition. This step effectively removes noise and retains the useful information in the original data, thereby significantly improving the overall detection accuracy and the robustness of the system. In this step, step S2 includes step S21, step S22, and step S23.

[0037] Step S21: Linearly interpolate the data in the first information according to the corresponding acquisition time, and align the linearly interpolated data in terms of time to obtain the time-aligned data;

[0038] It can be understood that in this step, a linear interpolation method is used to align the data of different sensors in time. Linear interpolation is a method of estimating unknown data points through the linear relationship between known data points. In this step, for the time series data of each sensor, if there is a time interval between data points, linear interpolation will fill in the missing values according to the values of the known data points and the time interval. Specifically, for the data of a certain sensor, assuming that there is an interval between the timestamps T1 and T2 collected, within this interval, by calculating the slope between the data points of T1 and T2, all intermediate values between T1 and T2 are interpolated, so as to ensure the continuity of the data in time. After the interpolation is completed, the data of all sensors will have a unified time step, which can ensure the synchronization of different types of data on the time axis and avoid data distortion or errors in subsequent analysis caused by time misalignment.

[0039] Step S22: Use the turnout rail light band image data in the first information as the reference points with the light band edge pixel points, and align the geometric shape data and ultrasonic detection data in the first information with the reference points respectively to obtain the aligned spatial data;

[0040] It can be understood that during the spatial alignment process, the light band edge pixel points in the turnout rail light band image data are used as reference points. The light band edge is the significant boundary between the turnout rail surface and its surrounding environment in the image, and has strong positioning characteristics. Therefore, selecting the edge pixel points of the light band as reference points can align with other data more accurately. The edge information of the light band is extracted through the Canny edge detection algorithm to obtain a set of coordinates of the edge pixel points.

[0041] Next, for other data in the first information, such as the geometric shape data and ultrasonic detection data of the turnout rail, they need to be aligned with the above-extracted light band edge pixel points through perspective transformation respectively. The geometric shape data of the turnout rail usually includes information such as the curvature, width, and inclination angle of the rail. The ultrasonic detection data usually contains the detection results of the rail surface (such as thickness, crack position, etc.), and these data also need to be aligned with the light band edge through coordinate mapping.

[0042] Step S23: Take the aligned time data and spatial data as multi-modal data, and perform wavelet transform and multi-resolution reconstruction on the multi-modal data. Among them, perform wavelet transform on the geometric shape data and ultrasonic data of the turnout rail in the multi-modal data, and perform multi-resolution reconstruction on the light band image data of the turnout rail in the multi-modal data to obtain the denoised multi-modal data.

[0043] It can be understood that in this step, wavelet transform is used to denoise the geometric shape data and ultrasonic data of the turnout rail. Wavelet transform can effectively extract the low-frequency information of the signal while eliminating high-frequency noise, so it is particularly suitable for processing signals with noise. In the geometric shape data of the rail, the noise may come from environmental interference or sensor errors, while the ultrasonic detection data may contain noise caused by surface irregularities or abnormal contacts. Through wavelet transform, these noises can be removed, making the remaining signal clearer and facilitating subsequent analysis. Among them, wavelet transform decomposes the geometric shape data and ultrasonic detection data of the turnout rail into low-frequency and high-frequency parts through multi-scale decomposition. The low-frequency part mainly contains the smooth trend of the signal, while the high-frequency part contains noise components. By applying threshold processing, the high-frequency noise is removed, and the effective low-frequency information is retained, so as to achieve the effect of signal denoising. After the wavelet transform is completed, the processed data is reconstructed using the inverse wavelet transform to obtain the geometric shape data and ultrasonic data after noise reduction, improving the clarity and reliability of the data

[0044] For the light band image data of the turnout rail, multi-resolution reconstruction technology is adopted. This technology uses image information at different resolutions and enhances the details of the image through layer-by-layer image reconstruction, so that important features in the image (such as light band edges and regional changes) are presented more precisely. While denoising, multi-resolution reconstruction can retain the key structure of the image, avoid feature loss caused by over-smoothing, and ensure the accuracy of subsequent feature extraction and target recognition tasks. The multi-resolution reconstruction method decomposes the image or signal into multiple scales at different resolutions, so as to analyze the signal at different resolutions. In the process of image reconstruction, the low-resolution image is first used as the basis, and the details of higher resolutions are gradually fused

[0045] Step S3: Send the denoised first information to a preset feature extraction module for feature extraction to obtain the feature data of the turnout rail light band;

[0046] It can be understood that in this step, the feature extraction module can accurately locate the specific area of the light band and quickly extract multi-dimensional light band features, providing basic data for subsequent light band recognition. In this step, step S3 includes step S31, step S32 and step S33

[0047] Step S31: Extract the light band area of the turnout rail in the denoised first information. Among them, perform edge detection on the light band area in the denoised first information, identify the boundary area of the light band, and use the Hough line transform to fit the light band edge curve. Extract the key area in the light band through the region projection analysis method to obtain the key image information of the light band area. The key area is the region of interest of the light band

[0048] It can be understood that in this step, the light band region in the denoised first information is processed by an edge detection algorithm. In this process, the algorithm detects the edges of the light band based on the brightness changes in the image. And the threshold of edge detection is adjusted based on the signal intensity detected by the edge detection algorithm, so that the accuracy of edge detection is more stable under different lighting conditions. Then, the Hough line transform is used to fit the edges of the light band to obtain the accurate edge curve of the light band. The process of the Hough line transform includes performing edge detection on the image, extracting the edge points of the light band, and then representing these edge points as a set of parameters in the polar coordinate system. Next, using a voting mechanism, multiple edge points vote under the same polar coordinates, thus forming an accumulation matrix in the polar coordinate space, representing the straight lines in the image. By identifying the peaks in the accumulation matrix, the edge straight lines of the light band can be accurately extracted from the image. This method has good anti-noise ability and can effectively identify the edges of the light band. Even if there are missing or irregular light band edges, the Hough transform can still provide reliable straight line extraction. The region projection analysis method includes performing region segmentation on the image, identifying the light band region, and separating it from other parts. Then, a projection operation is performed on the light band region, and the pixel values of this region are accumulated along a certain direction (such as the horizontal or vertical direction) to obtain a projection map. By analyzing the peak positions and change trends of the projection map, the geometric features of the light band, such as the width and thickness of the light band, can be extracted. Finally, by analyzing the projection results, the most representative region in the light band is determined as the region of interest.

[0049] Step S32: Extract the light band features from the cropped image of the light band region. Among them, the image information of the light band region is converted to the HSV color space to obtain the brightness feature of the light band and the converted image information.

[0050] It can be understood that in this step, the image of the light band region is converted from the RGB color space to the HSV color space, and the brightness component is extracted. This component can effectively reflect the intensity changes of the light band, thereby identifying the presence and morphological features of the light band. This process not only helps to eliminate the influence of the external environment on the image, but also makes the light band part in the image more distinct from the background. The converted image information can retain the color and detail features of the light band, providing more accurate data for subsequent feature extraction and analysis.

[0051] Step S33: Calculate the gradient direction distribution of the light band through the converted image information and the HOG feature extraction method, and use the gradient direction distribution of the light band as the light band texture feature.

[0052] It can be understood that in this step, the image is first divided into multiple small block regions, and the gradient direction and magnitude of the pixels are calculated within each block. Then, the gradient directions within each block are quantized according to a predetermined angular range to form a direction histogram. These histograms are combined to form a set of representative features that describe the texture patterns of the light band in each local region. Since the light band usually has specific texture features, such as uniform brightness changes or edge lines, texture features with discriminability can be effectively extracted from the image by this method.

[0053] Step S4: Send the feature data of the turnout rail light band to a preset target recognition model for target recognition, and classify and perform anomaly judgment on the light band based on the result obtained from the target recognition to obtain the final recognition result of the light band.

[0054] It can be understood that this step can effectively improve the accuracy of light band recognition, and at the same time enhance the anomaly detection ability by combining information in the time dimension to ensure real-time monitoring and intelligent analysis of the light band status. In this step, step S4 includes step S41, step S42, step S43, step S44, and step S45.

[0055] Step S41: Perform target detection on the feature data of the turnout rail light band based on a preset YOLOv8 target detection model. Among them, the YOLOv8 target detection model is trained with the feature data of the turnout rail light band, and by optimizing the size of the anchor boxes in the YOLOv8 target detection model, the boundary box coordinates of the optimized light band region are identified.

[0056] It can be understood that the size of the anchor boxes of the YOLOv8 model is optimized in this step. Since the morphology of the light band region in the image has strong regularity, the typical boundary box sizes of the light band can be statistically analyzed in combination with historical data, and the distribution of the light band boundary boxes is clustered and analyzed using the K-means clustering algorithm to obtain the optimal anchor box size distribution. Then, during the training process of YOLOv8, these optimized anchor box sizes are replaced with the default anchor boxes, enabling the model to more accurately fit the actual morphology of the light band region during detection, thereby reducing the deviation of the target boundary box and improving the accuracy of light band positioning. In addition, in order to further improve the detection effect of the model, data augmentation strategies such as random cropping, scale transformation, and contrast adjustment are adopted during the training stage to enhance the robustness of the model to light band features in complex environments.

[0057] Step S42: Perform integrity classification based on the boundary box coordinates of the optimized light band region. Among them, by analyzing the difference between the coordinates of two adjacent boundary boxes and judging the category of the light band through a preset category threshold corresponding to each difference, the classification result of the light band is obtained. The categories of the light band include complete light band, intermittent light band, and missing light band.

[0058] It can be understood that in this step, for each pair of adjacent bounding boxes, the coordinate difference in the light band extension direction is calculated, and this difference reflects the integrity of the light band region in this direction. If the distance between adjacent bounding boxes is within a certain threshold range, the light band in this region is considered complete; if the distance exceeds the preset threshold but still maintains a certain regularity, it is judged as an intermittent light band; if the distance far exceeds the set threshold, or there is no effective light band region for multiple consecutive bounding boxes, it is determined that the light band in this region is missing. This threshold-based method can quickly distinguish different types of light band states and adapt to the changes in light band distribution under different track environments.

[0059] Step S43: Perform dynamic change analysis of the light band based on the classification result of the light band. Specifically, by constructing a time-series sliding window model, inputting the feature data and classification results of the light band at multiple preset historical moments, enabling the preset bidirectional long short-term memory network to learn the state evolution of the light band at different time points, and predicting the trend of the light band features, to obtain the light band feature data and classification results at different time points;

[0060] It can be understood that in this step, by constructing a time-series sliding window model, the dynamic changes of the light band at different time points are analyzed to judge the evolution trend of the light band state and predict the possible future changes in the light band features. First, after obtaining the classification result of the light band, a sliding window is constructed in chronological order, and the feature data and classification results of the light band at multiple adjacent moments are used as input data, enabling the model to learn the law of the light band changing over time. The size of the sliding window is set according to the time scale of the light band change to ensure that it can capture both short-term fluctuations and reflect long-term trends. During the data input process, a bidirectional long short-term memory network is used for training. This network can not only consider the forward changes of the time series but also combine the reverse information, so as to more comprehensively learn the dynamic evolution mode of the light band features. Specifically, the forward calculation unit of the network is responsible for analyzing the changes of the light band features from the past to the present, while the reverse calculation unit traces back the historical data to identify the key factors that may lead to the current state. This bidirectional learning method helps to improve the accuracy of prediction, enabling the model to not only detect the current state of the light band but also predict the upcoming changes.

[0061] Step S44: Send the light band feature data and classification results at different time points to an autoencoder for detection. Specifically, based on the autoencoder, the preset normal light band feature data is mapped to a low-dimensional feature space, and the data in the mapped low-dimensional feature space and the preset normal light band classification results are marked correspondingly to obtain the data information after learning;

[0062] It is understandable that the autoencoder in this step consists of an encoder and a decoder. First, the encoder receives the input optical band feature data and maps it to a low-dimensional feature space. During this process, the network extracts the main features of the data while removing redundant information, so that the low-dimensional representation can retain the key features of the optical band state to the greatest extent. Since the input data contains the optical band features at different time points, the encoder needs to learn the pattern of the optical band evolving over time in the normal state to ensure that the mapped data can accurately describe the dynamic change features of the optical band.

[0063] Step S45: Reconstruct the data information after learning based on the auto-decoder to obtain the loss function of the autoencoder, and send the optical band feature data and optical band classification results at different time points to the autoencoder after constructing the loss function for anomaly judgment to obtain the anomaly judgment result of the optical band.

[0064] It is understandable that in this step, first, the learned low-dimensional feature data is input into the auto-decoder, and the auto-decoder attempts to restore these low-dimensional representations to the original optical band feature data. In an ideal situation, if the input data conforms to the normal optical band pattern learned during the training of the autoencoder, the reconstructed data should be highly consistent with the original input, and the reconstruction error is small. However, when the input data contains abnormal optical band features, since this data has not appeared or has a large deviation in distribution during the training process, the auto-decoder cannot effectively restore its true features, resulting in a significant increase in the reconstruction error.

[0065] Next, by calculating the difference between the input data and the reconstructed data, the loss function of the autoencoder is constructed. The value of the loss function can be used as a metric for the degree of anomaly, usually calculated in the form of mean square error or reconstruction residual. If the optical band feature data and classification result at a certain moment generate a large loss value during the reconstruction process, it indicates that the data may be in an abnormal state. Finally, the optical band feature data and classification results at all time points are all judged for anomalies by calculating the loss function. If the loss value exceeds the preset threshold, it is marked as an abnormal optical band state; otherwise, the optical band is considered to be in a normal state.

[0066] Among them, the loss function is as follows:

[0067]

[0068] Among them, L represents the final loss value, n represents the total number of samples, X i represents the i-th original optical band feature data, X' i represents the i-th reconstructed optical band feature data, ||X i - X' i || 2 represents the Euclidean distance of the i-th sample.

[0069] Example 2:

[0070] As Figure 2 shown, this embodiment provides a recognition system for the light band of turnout rails. Refer to Figure 2 The system includes an acquisition unit 701, a denoising unit 702, an extraction unit 703, and a recognition unit 704.

[0071] The acquisition unit 701 is used to acquire first information, where the first information includes turnout rail light band image data, turnout rail geometric shape data, and ultrasonic detection data;

[0072] The denoising unit 702 is used to send the first information to a preset denoising model for denoising processing to obtain denoised first information;

[0073] Among them, the denoising unit 702 includes a first denoising sub-unit 7021, a second denoising sub-unit 7022, and a third denoising sub-unit 7023.

[0074] The first denoising sub-unit 7021 is used to linearly interpolate the data in the first information according to the corresponding acquisition time, and align the linearly interpolated data in terms of time to obtain time-aligned data;

[0075] The second denoising sub-unit 7022 is used to use the pixel points at the edge of the light band of the turnout rail light band image data in the first information as reference points, and align the geometric shape data and ultrasonic detection data in the first information with the reference points respectively to obtain spatially aligned data;

[0076] The third denoising sub-unit 7023 is used to use the time-aligned data and spatial data as multimodal data, and perform wavelet transform and multi-resolution reconstruction on the multimodal data. Among them, wavelet transform is performed on the geometric shape data and ultrasonic data of the turnout rail in the multimodal data, and multi-resolution reconstruction is performed on the light band image data of the turnout rail in the multimodal data to obtain denoised multimodal data.

[0077] The extraction unit 703 is used to send the denoised first information to a preset feature extraction module for feature extraction to obtain feature data of the turnout rail light band;

[0078] Among them, the extraction unit 703 includes a first extraction sub-unit 7031, a second extraction sub-unit 7032, and a third extraction sub-unit 7033.

[0079] The first extraction subunit 7031 is used to extract the turnout rail light band area in the denoised first information based on an edge detection algorithm. Specifically, edge detection is performed on the light band area in the denoised first information to identify the boundary area of the light band, and the Hough line transform is used to fit the light band edge curve. The key area in the light band is extracted through the region projection analysis method to obtain the key image information of the light band area, and the key area is the region of interest of the light band.

[0080] The second extraction subunit 7032 is used to extract the light band features from the cropped image of the light band area. Specifically, the image information of the light band area is converted to the HSV color space to obtain the brightness feature of the light band and the converted image information.

[0081] The third extraction subunit 7033 is used to calculate the gradient direction distribution of the light band through the converted image information and the HOG feature extraction method, and use the gradient direction distribution of the light band as the light band texture feature.

[0082] The recognition unit 704 is used to send the feature data of the turnout rail light band to a preset target recognition model for target recognition, and classify and perform anomaly judgment on the light band based on the result of the target recognition to obtain the final recognition result of the light band.

[0083] Among them, the recognition unit 704 includes a first recognition subunit 7041, a second recognition subunit 7042, a third recognition subunit 7043, a fourth recognition subunit 7044, and a fifth recognition subunit 7045.

[0084] The first recognition subunit 7041 is used to perform target detection on the feature data of the turnout rail light band based on a preset YOLOv8 target detection model. Specifically, the YOLOv8 target detection model is trained with the feature data of the turnout rail light band, and the size of the anchor box in the YOLOv8 target detection model is optimized to identify the boundary box coordinates of the optimized light band area.

[0085] The second recognition subunit 7042 is used to perform integrity classification based on the boundary box coordinates of the optimized light band area. Specifically, the difference between adjacent two boundary box coordinates is analyzed, and the category of the light band is judged through a preset category threshold corresponding to each difference to obtain the classification result of the light band. The categories of the light band include a complete light band, an intermittent light band, and a missing light band.

[0086] The third recognition subunit 7043 is configured to perform optical band dynamic change analysis based on the classification result of the optical band. Specifically, by constructing a time-series sliding window model, inputting the feature data and optical band classification results of the optical band at multiple preset historical moments, enabling the preset bidirectional long short-term memory network to learn the state evolution of the optical band at different time points, and predicting the trend of the optical band features, so as to obtain the optical band feature data and optical band classification results at different time points.

[0087] The fourth recognition subunit 7044 is configured to send the optical band feature data and optical band classification results at different time points to an autoencoder for detection. Specifically, based on the autoencoder, the preset normal optical band feature data is mapped to a low-dimensional feature space, and the mapped low-dimensional feature space data and the preset normal optical band classification results are correspondingly marked to obtain the data information after learning is completed.

[0088] The fifth recognition subunit 7045 is configured to reconstruct the data information after learning is completed based on the decoder to obtain the loss function of the autoencoder, and send the optical band feature data and optical band classification results at different time points to the autoencoder after constructing the loss function for anomaly judgment to obtain the anomaly judgment result of the optical band.

[0089] It should be noted that regarding the system in the above embodiments, the specific manners in which each module performs operations have been described in detail in the embodiments related to the method, and will not be elaborated herein.

[0090] Embodiment 3:

[0091] Corresponding to the above method embodiment, an identification device for turnout rail optical bands is further provided in this embodiment. The identification device for turnout rail optical bands described below can be correspondingly referred to the identification method for turnout rail optical bands described above.

[0092] Figure 3 It is a block diagram of an identification device 800 for turnout rail optical bands shown according to an exemplary embodiment. As Figure 3 shown, the identification device 800 for turnout rail optical bands may include: a processor 801, a memory 802. The identification device 800 for turnout rail optical bands may further include one or more of a multimedia component 803, an I / O interface 804, and a communication component 805.

[0093] Among them, the processor 801 is used to control the overall operation of the turnout rail light band recognition device 800 to complete all or part of the steps in the above-mentioned turnout rail light band recognition method. The memory 802 is used to store various types of data to support the operation of the turnout rail light band recognition device 800. These data may include, for example, instructions for any application or method operating on the turnout rail light band recognition device 800, as well as application-related data, such as contact data, sent and received messages, pictures, audio, video, and so on. The memory 802 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk. The multimedia component 803 may include a screen and an audio component. The screen can be, for example, a touch screen, and the audio component is used to output and / or input audio signals. For example, the audio component may include a microphone for receiving external audio signals. The received audio signal can be further stored in the memory 802 or sent through the communication component 805. The audio component also includes at least one speaker for outputting audio signals. The I / O interface 804 provides an interface between the processor 801 and other interface modules, and the above-mentioned other interface modules can be a keyboard, a mouse, buttons, etc. These buttons can be virtual buttons or physical buttons. The communication component 805 is used for wired or wireless communication between the turnout rail light band recognition device 800 and other devices. Wireless communication, such as Wi-Fi, Bluetooth, near field communication (NFC), 2G, 3G or 4G, or a combination of one or more of them. Accordingly, the communication component 805 may include: a Wi-Fi module, a Bluetooth module, an NFC module.

[0094] In an exemplary embodiment, the recognition device 800 for the switch rail light band can be implemented by one or more application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components, and is used to execute the above-mentioned method for recognizing the switch rail light band.

[0095] In another exemplary embodiment, a computer storage medium including program instructions is further provided. When the program instructions are executed by a processor, the steps of the above-mentioned method for recognizing the switch rail light band are implemented. For example, the computer storage medium can be the above-mentioned memory 802 including program instructions, and the above-mentioned program instructions can be executed by the processor 801 of the recognition device 800 for the switch rail light band to complete the above-mentioned method for recognizing the switch rail light band.

[0096] Embodiment 4:

[0097] Corresponding to the above method embodiment, a storage medium is further provided in this embodiment. A storage medium described below can be correspondingly referred to with a method for recognizing a switch rail light band described above.

[0098] A storage medium stores a computer program. When the computer program is executed by a processor, the steps of the method for recognizing the switch rail light band in the above method embodiment are implemented.

[0099] Specifically, the storage medium can be various storage media such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disc that can store program codes.

[0100] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

[0101] As described above, it is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A method for identifying a turnout rail light band, characterized in that: include: Acquire first information, wherein the first information includes turnout rail light band image data, turnout rail geometry data, and ultrasonic detection data; Sending the first information to a preset denoising model for denoising to obtain denoised first information; The denoised first information is sent to a preset feature extraction module for feature extraction to obtain feature data of the turnout rail light band; The characteristic data of the turnout rail light band is sent to a preset target recognition model for target recognition, and the light band is classified and abnormality judged based on the result obtained by the target recognition to obtain the final recognition result of the light band.

2. The method for identifying the light band of a turnout rail according to claim 1, characterized in that , sending the first information to a preset denoising model for denoising processing to obtain the denoised first information, including: Performing linear interpolation on the data in the first information according to the corresponding acquisition time therein, and performing time alignment on the linearly interpolated data to obtain aligned time data; Using the edge pixel points of the light band of the turnout rail image data in the first information as reference points, aligning the geometric data and ultrasonic detection data in the first information with the reference points respectively, to obtain aligned spatial data; The aligned time data and space data are taken as multimodal data, and the multimodal data is subjected to wavelet transform and multi-resolution reconstruction, wherein the geometric data and ultrasonic data of the turnout rail in the multimodal data are subjected to wavelet transform, and the light band image data of the turnout rail in the multimodal data is subjected to multi-resolution reconstruction to obtain the multimodal data after noise reduction.

3. The method for identifying the light band of a turnout rail according to claim 1, characterized in that , sending the denoised first information to a preset feature extraction module for feature extraction, and obtaining feature data of the turnout rail light band, including: The turnout rail light band region in the first information after denoising is extracted based on an edge detection algorithm, wherein edge detection is performed on the light band region in the first information after denoising, the boundary region of the light band is identified, and the light band edge curve is obtained by fitting using Hough straight line transform, and the key region in the light band is extracted by regional projection analysis method to obtain key image information of the light band region, wherein the key region is the region of interest of the light band; Extracting light band features from the cropped image of the light band area, wherein the image information of the light band area is converted into an HSV color space to obtain brightness features of the light band and converted image information; The gradient direction distribution of the light band is calculated by using the converted image information and the HOG feature extraction method, and the gradient direction distribution of the light band is used as the light band texture feature.

4. The method for identifying the light band of a turnout rail according to claim 1, characterized in that , sending the characteristic data of the turnout rail light strip to a preset target recognition model for target recognition, and classifying and judging abnormalities of the light strip based on the result of target recognition, and obtaining the final recognition result of the light strip, including: Performing target detection on the feature data of the turnout rail light band based on a preset YOLOv8 target detection model, wherein the YOLOv8 target detection model is trained by the feature data of the turnout rail light band, and the coordinates of the bounding box of the optimized light band area are identified by optimizing the size of the anchor box in the YOLOv8 target detection model; Integrity classification is performed based on the bounding box coordinates of the optimized light band area, wherein the classification result of the light band is obtained by analyzing the difference between two adjacent bounding box coordinates and judging the category of the light band by a preset category threshold corresponding to each difference, and the categories of the light band include complete light band, intermittent light band and missing light band; Based on the classification results of the light bands, a dynamic change analysis of the light bands is performed, wherein a time series sliding window model is constructed, characteristic data of the light bands at multiple preset historical moments and the classification results of the light bands are input, so that a preset bidirectional long short-term memory network learns the state evolution of the light bands at different time points, and a trend prediction is performed on the light band characteristics, thereby obtaining the characteristic data of the light bands and the classification results of the light bands at different time points; The light band feature data and light band classification results at different time points are sent to the autoencoder for detection, wherein the preset normal light band feature data is mapped to a low-dimensional feature space based on the autoencoder, and the mapped low-dimensional feature space data and the preset normal light band classification results are marked correspondingly to obtain data information after learning is completed; Based on the autodecoder, the data information after learning is reconstructed to obtain the loss function of the autoencoder, and the light band feature data and light band classification results at different time points are sent to the autoencoder after the loss function is constructed for abnormality judgment to obtain the abnormality judgment result of the light band.

5. A turnout rail light strip recognition system, characterized in that: include: An acquisition unit, used for acquiring first information, wherein the first information includes turnout rail light band image data, turnout rail geometry data and ultrasonic detection data; A denoising unit, configured to send the first information to a preset denoising model for denoising to obtain denoised first information; An extraction unit, used for sending the denoised first information to a preset feature extraction module for feature extraction to obtain feature data of the turnout rail light band; The recognition unit is used to send the characteristic data of the turnout rail light band to a preset target recognition model for target recognition, and classify and judge the abnormality of the light band based on the result obtained by the target recognition to obtain the final recognition result of the light band.

6. The turnout rail light strip recognition system according to claim 5, characterized in that: The denoising unit comprises: A first denoising subunit, configured to perform linear interpolation on the data in the first information according to the corresponding acquisition time therein, and perform time alignment on the data obtained after the linear interpolation to obtain aligned time data; A second denoising subunit is used to use the light band edge pixel point of the turnout rail light band image data in the first information as a reference point, and align the geometric data and ultrasonic detection data in the first information with the reference point to obtain aligned spatial data; The third denoising subunit is used to use the aligned time data and space data as multimodal data, and perform wavelet transform and multi-resolution reconstruction on the multimodal data, wherein the geometric data and ultrasonic data of the turnout rail in the multimodal data are subjected to wavelet transform, and the light band image data of the turnout rail in the multimodal data are subjected to multi-resolution reconstruction to obtain the multimodal data after noise reduction.

7. The turnout rail light strip recognition system according to claim 5, characterized in that: The extraction unit comprises: A first extraction subunit is used to extract the turnout rail light band area in the first information after denoising based on an edge detection algorithm, wherein edge detection is performed on the light band area in the first information after denoising to identify the boundary area of ​​the light band, and a light band edge curve is obtained by fitting using Hough straight line transformation, and a key area in the light band is extracted by a regional projection analysis method to obtain key image information of the light band area, wherein the key area is a region of interest of the light band; The second extraction subunit is used to extract light band features from the cropped image of the light band area, wherein the image information of the light band area is converted into an HSV color space to obtain brightness features of the light band and converted image information; The third extraction subunit is used to calculate the gradient direction distribution of the light band by using the converted image information and the HOG feature extraction method, and use the gradient direction distribution of the light band as the light band texture feature.

8. The turnout rail light strip recognition system according to claim 5, characterized in that: The identification unit comprises: A first identification subunit is used to perform target detection on the feature data of the turnout rail light band based on a preset YOLOv8 target detection model, wherein the YOLOv8 target detection model is trained by the feature data of the turnout rail light band, and the coordinates of the bounding box of the optimized light band area are identified by optimizing the size of the anchor box in the YOLOv8 target detection model; A second identification subunit is used to perform integrity classification based on the bounding box coordinates of the optimized light band area, wherein the classification result of the light band is obtained by analyzing the difference between two adjacent bounding box coordinates and judging the category of the light band by a preset category threshold corresponding to each difference, and the categories of the light band include complete light band, intermittent light band and missing light band; A third identification subunit is used to analyze the dynamic changes of the light band based on the classification results of the light band, wherein a time series sliding window model is constructed, and characteristic data of the light band at multiple preset historical moments and the classification results of the light band are input, so that a preset bidirectional long short-term memory network learns the state evolution of the light band at different time points, and performs trend prediction on the light band characteristics, thereby obtaining the characteristic data of the light band and the classification results of the light band at different time points; The fourth identification subunit is used to send the light band feature data and the light band classification results at different time points to the autoencoder for detection, wherein the preset normal light band feature data is mapped to a low-dimensional feature space based on the autoencoder, and the mapped low-dimensional feature space data and the preset normal light band classification results are marked correspondingly to obtain data information after learning is completed; The fifth identification subunit is used to reconstruct the data information after learning based on the autodecoder to obtain the loss function of the autoencoder, and send the light band feature data and light band classification results at different time points to the autoencoder after constructing the loss function for abnormality judgment to obtain the abnormality judgment result of the light band.

9. A device for identifying a turnout rail light band, characterized in that: include: Memory for storing computer programs; A processor, used to implement the steps of the method for identifying the turnout rail light strip as claimed in any one of claims 1 to 4 when executing the computer program.

10. A storage medium, characterized in that: The storage medium stores a computer program, which, when executed by a processor, implements the steps of the method for identifying a turnout rail light strip as claimed in any one of claims 1 to 4.

Images