Rail sleeper chipping disease detection system based on PCA technology

Abstract

The invention discloses a track sleeper chipping disease detection system based on a PCA technology, and relates to the technical field of chipping disease detection, and the system comprises the steps: collecting an original sleeper image through an image collection module as a to-be-detected image; the sleeper chipping image analysis module extracts a data dictionary by collecting a normal sleeper image set, generates a data model through PCA algorithm dimensionality reduction, constructs a reference image in combination with projection transformation of a to-be-detected image, preprocesses the to-be-detected image, and makes a difference between the to-be-detected image and the reference image to obtain a comparison image. And judging a falling block disease region through threshold segmentation and connected region area calculation, and marking an output result. According to the method, high-efficiency and high-precision recognition of sleeper falling block disease detection is realized, and the method is effectively adaptive to scenes with complex sleeper surface textures and redundant high-dimensional data.

Description

Technical Field

[0001] This invention relates to the field of track blockage detection technology, and more particularly to a track blockage detection system based on PCA technology. Background Technology

[0002] As the core load-bearing component of railway tracks, railway sleepers directly bear the heavy load of trains and tracks. Their structural integrity (especially preventing sags) is crucial to ensuring safe railway transportation. With the rapid expansion of urban rail transit and railway networks, the need for sleeper maintenance and inspection is becoming increasingly urgent.

[0003] Currently, sleeper block detection still primarily relies on manual inspection. While some solutions incorporate image acquisition devices such as lasers and industrial cameras, manual judgment of the results is still required after acquisition. Manual inspection is not only limited by environmental factors such as weather and lighting, but also suffers from low efficiency, strong subjectivity, high rates of missed and false detections, and high labor intensity. It is difficult to adapt to the needs of large-scale, high-frequency track maintenance, and cannot guarantee the consistency and reliability of the inspection. In addition, although machine vision technology has been widely used in track inspection, the complex texture of sleeper surfaces makes them susceptible to interference from dust, water stains, and minor scratches after deployment. The surface condition changes frequently, and sleeper images are high-dimensional data containing a large amount of redundant information, which also directly affects the accuracy of block defect identification. For these reasons, conventional visual inspection methods cannot accurately extract core features and cannot be directly applied to sleeper block detection.

[0004] In summary, existing testing technologies are insufficient to meet the requirements of automation, high precision, and high efficiency in sleeper testing, and there is an urgent need to provide corresponding technical solutions. Summary of the Invention

[0005] In view of this, this application provides a track sleeper blockage detection system based on PCA technology to address the shortcomings of existing technologies.

[0006] The first aspect of this application provides a track sleeper deterioration detection system based on PCA technology, comprising: The image acquisition module, fixed to the track defect detection device, is used to acquire original images of the sleepers as images of the sleepers to be inspected. The track defect detection device includes an electric train, a data acquisition module, a system control module, and a power supply module. The electric train provides the driving vehicle. The data acquisition module is a computing platform used to acquire imaging data from the image acquisition module and transmit it to the sleeper block drop image analysis module. The system control module is an embedded computing platform that calculates the real-time speed of the electric train, outputs corresponding pulse signals to the image acquisition module based on the real-time speed, and controls the image acquisition module to acquire images at a set frequency. The power supply module is a battery pack or an external power supply interface used to power all modules. The sleeper block falling image analysis module is used to realize the construction of reference image, processing of sleeper image to be detected, image comparison and execution of defect judgment algorithm; The benchmark image construction algorithm includes: acquiring a set of normal sleeper images and extracting a data dictionary from the set of normal sleeper images using a preset method; performing data dimensionality reduction and data model generation on the data dictionary using the PCA algorithm; and performing projection transformation based on the sleeper image to be detected to obtain a benchmark image that is correlated with the sleeper image to be detected. The image processing algorithm for the sleeper to be detected includes: preprocessing the image of the sleeper to be detected; The image comparison algorithm includes: performing an absolute difference operation between the preprocessed image of the sleeper to be detected and the reference image to obtain a comparison image; The disease determination algorithm includes: performing threshold segmentation on the comparison image, calculating the area of ​​connected regions in the segmented image, determining regions with areas exceeding a set threshold as diseased areas, and labeling the corresponding regions and outputting the disease detection results.

[0007] In one possible implementation of the first aspect, the preset method is the SC encoding method.

[0008] In one possible implementation of the first aspect, performing data dimensionality reduction and data model generation on the data dictionary using the PCA algorithm includes: The sample data of the normal sleeper image set is decentered to eliminate global offset interference; Based on the decentralized sample data, the covariance matrix describing the linear correlation of pixel dimensions is calculated; The covariance matrix is ​​decomposed into eigenvalues ​​to obtain the corresponding set of eigenvalues ​​and set of eigenvectors. Based on the set of eigenvalues, the eigenvalues ​​are sorted from largest to smallest, and the eigenvalues ​​in the sorted set of eigenvalues ​​are filtered; in the filtered set of eigenvalues, the K largest eigenvalues ​​and the K corresponding eigenvectors are selected, and a new eigenvector matrix is ​​formed by using all the selected eigenvectors as row vectors. The original image is reconstructed based on the new feature vector matrix to obtain the reference image.

[0009] In one possible implementation of the first aspect, filtering the feature values ​​in the sorted feature value set includes: The feature values ​​are selected based on the cumulative contribution rate, which is calculated using the following formula: The cumulative contribution rate of the eigenvalues. These are the sorted feature values. The principal component dimensions after filtering. This refers to the original data dimension.

[0010] In one possible implementation of the first aspect, the selection of feature values ​​based on the cumulative contribution rate includes: A preset contribution rate threshold is used, and the corresponding feature value is retained only when the cumulative contribution rate is not less than the preset contribution rate threshold.

[0011] In one possible implementation of the first aspect, preprocessing the sleeper image to be detected includes: applying a Gaussian filter to the sleeper image to be detected.

[0012] In one possible implementation of the first aspect, the comparison image is obtained by performing an absolute difference operation between the preprocessed image of the sleeper to be detected and the reference image, including: To compare the images, The image is of the sleeper to be inspected. This is the baseline image.

[0013] In one possible implementation of the first aspect, thresholding the comparison image includes: For the binarized image after thresholding, in coordinates Pixel value at; To compare images in coordinates Pixel value at; This is the preset pixel grayscale threshold; when When this occurs, it is determined to be a normal area; when At that time, it was determined to be an area suspected of being affected by chipping.

[0014] In one possible implementation of the first aspect, calculating the area of ​​connected regions in the segmented image and determining regions with areas exceeding a set threshold as patchy disease areas includes: The area of ​​connected regions is calculated for suspected fragmented defect areas after segmentation, and the pixel area of ​​each connected region is counted; suspected fragmented defect areas with pixel areas exceeding a set threshold are identified as fragmented defect areas.

[0015] In one possible implementation of the first aspect, the annotation of the corresponding area and the output of the disease detection results include: After completing the defect analysis, the system outputs a feedback result image with the outline of the missing block area, and simultaneously outputs the corresponding detection result of 0; and an image of a normal sleeper without markings, and simultaneously outputs the corresponding detection result of 1. A test result of 0 indicates that the sleepers in the corresponding area have a blockage defect; a test result of 1 indicates that the sleepers in the corresponding area do not have a blockage defect.

[0016] The beneficial effects are as follows: This invention discloses a track sleeper blockage detection system based on PCA technology. The system uses an image acquisition module to acquire original sleeper images as the images to be detected. A sleeper blockage image analysis module extracts a data dictionary from a set of normal sleeper images, generates a data model through PCA algorithm dimensionality reduction, and constructs a reference image by combining the projection transformation of the image to be detected. After preprocessing, the image to be detected is subtracted from the reference image to obtain a comparison image. Then, threshold segmentation and connected region area calculation are used to determine the blockage defect area and label the output results. This invention achieves full automation of the sleeper blockage detection process, effectively adapting to scenarios with complex sleeper surface textures and high-dimensional data redundancy. It reduces the subjectivity and environmental limitations caused by manual intervention, improving the accuracy, efficiency, and reliability of detection, and providing precise data support for track maintenance. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of a track sleeper debriding detection system based on PCA technology provided in an embodiment of this application; Figure 2 This is a comparative schematic diagram of the image to be detected and the PCA reconstructed image provided in the embodiments of this application; Figure 3 This is a schematic diagram showing the image to be detected after marking the sleeper blockage defect area in the embodiment of this application. Detailed Implementation

[0019] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0020] In this application, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0021] Example In existing technologies, manual inspection is not only limited by environmental factors such as weather and lighting, but also suffers from low efficiency, strong subjectivity, high rates of missed and false detections, and high labor intensity. It is difficult to adapt to the large-scale, high-frequency track maintenance needs, and cannot guarantee the consistency and reliability of inspections. Furthermore, although machine vision technology has been widely used in track inspection, the complex texture of sleeper surfaces makes them susceptible to interference from dust, water stains, and minor scratches after deployment. The surface condition changes frequently, and sleeper images are high-dimensional data containing a large amount of redundant information, which directly affects the accuracy of identifying defects such as ballast stones.

[0022] Therefore, this application provides a track sleeper deterioration detection system based on PCA technology, such as... Figure 1 As shown, it includes: The image acquisition module, fixed to the track defect detection device, is used to acquire original images of the sleepers as images of the sleepers to be inspected. The sleeper block falling image analysis module is used to realize the construction of reference image, processing of sleeper image to be detected, image comparison and execution of defect judgment algorithm; The benchmark image construction algorithm includes: acquiring a set of normal sleeper images and extracting a data dictionary from the set of normal sleeper images using a preset method; performing data dimensionality reduction and data model generation on the data dictionary using the PCA algorithm; and performing projection transformation based on the sleeper image to be detected to obtain a benchmark image that is correlated with the sleeper image to be detected. The image processing algorithm for the sleeper to be detected includes: preprocessing the image of the sleeper to be detected; The image comparison algorithm includes: performing an absolute difference operation between the preprocessed image of the sleeper to be detected and the reference image to obtain a comparison image; The disease determination algorithm includes: performing threshold segmentation on the comparison image, calculating the area of ​​connected regions in the segmented image, determining regions with areas exceeding a set threshold as diseased areas, and labeling the corresponding regions and outputting the disease detection results.

[0023] This embodiment provides a track sleeper blockage detection system based on PCA technology, including an image acquisition module and a sleeper blockage image analysis module. The working principle of the system is explained by analyzing each module, as follows: Image acquisition module: Fixed to the track defect detection device, it collects original images of the sleepers, including two-dimensional texture images and three-dimensional depth images, as images of the sleepers to be detected.

[0024] Sleeper breakage image analysis module: This is used to implement the algorithms for constructing a reference image, processing images of sleepers to be inspected, image comparison, and defect determination. The specific execution of the algorithm is as follows: 1. Baseline image construction (e.g.) Figure 2 (Image on the right) Images of normal sleepers were collected, including those without missing pieces or damage, to form a data dictionary. The sample set; Extract the data dictionary and use a preset method (SC encoding or k-means clustering) to extract feature representation units from the normal sleeper image set to form data dictionary B (the dictionary is composed of multiple code words / atoms and is the basic expression of sleeper features). PCA baseline image construction involves using the PCA algorithm to reduce the dimensionality of the data dictionary and generate a data model, and then performing a projection transformation based on the sleeper image to be detected to obtain a baseline image that is correlated with the sleeper image to be detected.

[0025] In this embodiment, the execution process of the PCA algorithm is adapted to the scenario of reconstructing the benchmark image for detecting sleeper blockage defects. The core purpose is to reduce the dimensionality of high-dimensional sleeper image data, purify features, and remove noise, and finally reconstruct a normal sleeper benchmark image without defects or redundant interference, so as to provide accurate reference for subsequent sleeper blockage detection images.

[0026] Define a unified parameter, assuming that the sleeper image samples are high-dimensional data, the original data dimension is p, and the sample contains N normal sleeper image samples without missing blocks or defects; Indicates the first A high-dimensional pixel vector of a sleeper image sample This is a dataset of railway sleeper samples; Let be the covariance matrix of the sample set. Next, based on the defined unified parameters and core formulas, the execution steps of the PCA algorithm will be explained, including: Formula (1) is the formula for the sample mean, as follows: in, This represents the global mean vector of the sleeper sample dataset. The calculation logic is to sum the pixel vectors of N normal sleeper image samples in the sample set and then take the arithmetic mean; the vector dimension is... The dimension is consistent with that of a single sleeper image sample; this mean vector represents the average pixel brightness level of all normal sleeper images, reflects the global brightness characteristics of sleeper images, and is a common basic feature of sleeper images.

[0027] Formula (2) is the formula for the sample covariance matrix, as follows: in, The covariance matrix of the decentralized sleeper sample dataset has dimensions of . ; It is the first The decentralized sample is obtained by subtracting the mean vector from each sleeper sample; It is the transpose of the decentralized samples; the denominator is taken as... To ensure unbiased estimation and avoid computational bias when the sample size is small, the accuracy of the covariance matrix is ​​guaranteed. The covariance matrix is ​​the core of the PCA algorithm. Essentially, it describes the linear correlation between high-dimensional pixels in the sleeper image. The diagonal elements of the matrix represent the variance of a single pixel dimension; the larger the variance, the more significant the feature of that pixel dimension. The off-diagonal elements represent the covariance between two different pixel dimensions; a positive covariance indicates a positive correlation between the two pixel dimensions, a negative covariance indicates a negative correlation, and a covariance of 0 indicates no linear correlation between the two pixel dimensions.

[0028] Formula (3) is the linear transformation formula, as follows: in, This is a decentralized railway sleeper sample. Projected vectors to the new coordinate system The low-dimensional eigenvalues ​​on the surface are one-dimensional scalars; These are the unit basis vectors of the new coordinate system (with dimensions of ). ), It's the transpose of the basis vectors. Essentially, it maps the high-dimensional sleeper image pixel vectors onto a new one-dimensional coordinate axis through orthogonal linear projection, achieving dimensionality reduction from high to one dimension; the projected low-dimensional eigenvalues... This represents the sleeper sample in the basis vector. Features expressed in direction.

[0029] The matrix form of the linear transformation of formula (4) is as follows: Among them, the formula is the batch matrix expression of formula (3), which is suitable for engineering calculations with multiple samples and multiple dimensions; The low-dimensional feature matrix (dimension) of the sleeper sample set , ); The unit basis vector matrix (dimensions) of the new coordinate system It consists of k mutually orthogonal unit basis vectors; The decentralized sleeper sample dataset matrix (dimensions) ); This is the transpose of the basis vector matrix. Essentially, it projects N decentralized high-dimensional sleeper samples onto a low-dimensional feature space composed of k orthogonal basis vectors at once, achieving batch dimensionality reduction from a p-dimensional high-dimensional space to a k-dimensional low-dimensional space. After dimensionality reduction, each sleeper sample can be characterized by only k feature values.

[0030] Formula (5) is the variance formula for the new data, as follows: in, The low-dimensional eigenvalues ​​after projection in formula (3) The variance; It is the covariance matrix of formula (2). These are the projection basis vectors. It is the transpose of the basis vectors. This formula is proven mathematically: the variance of low-dimensional features is entirely determined by the projected basis vectors. Covariance Matrix The variance directly corresponds to the amount of effective information in the sleeper image carried by the low-dimensional features after projection; the larger the variance, the stronger the information contained in the basis vector. The more core sleeper features (outline, complete surface texture, shape) that can be retained in a direction, the less redundant noise there is; the smaller the variance, the more invalid noise information such as dust and scratches are contained in that direction.

[0031] Formula (6) is the formula for the covariance of the new data, as follows: Here, it represents two different low-dimensional features. , Covariance between , These are the two corresponding projection basis vectors. Let be the covariance matrix. The numerical value of covariance represents the information redundancy between two low-dimensional features. When the covariance is 0, there is no linear correlation between the two features and no information redundancy. When the covariance is not 0, there is information overlap between the two features, which will cause feature redundancy and increase the amount of computation.

[0032] Formula (7) is the formula for the constrained optimization problem, as follows: in, The objective is to find the optimal projection basis vectors. This allows the variance of the low-dimensional features after projection to reach its maximum value. Represents constraints, projected basis vectors The basis vectors must be unit vectors. If the basis vectors are not restricted to unit vectors, the variance can be increased infinitely by infinitely increasing the length of the basis vectors. In this case, maximizing the variance has no practical significance. By restricting them to unit vectors, the magnitude of the variance is determined only by the direction of the basis vectors, ensuring the validity and uniqueness of the optimization results.

[0033] The core execution steps of the PCA algorithm consist of five core steps, including: Step 1): Decenter the sample data of the normal sleeper image set to eliminate global offset interference; Step 2): Based on the decentralized sample data, calculate the covariance matrix describing the linear correlation of pixel dimensions; Step 3): Perform eigenvalue decomposition on the covariance matrix to obtain the corresponding set of eigenvalues ​​and set of eigenvectors; Step 4): Based on the set of eigenvalues, sort the eigenvalues ​​from largest to smallest, and filter the eigenvalues ​​in the sorted set of eigenvalues; in the filtered set of eigenvalues, select the K eigenvectors corresponding to the K largest eigenvalues, and form a new eigenvector matrix using all the selected eigenvectors as row vectors; Step 5): Reconstruct the original image based on the new feature vector matrix to obtain the reference image.

[0034] Step 1) involves decentralization, corresponding to formula (1). The mean vector of the sleeper samples is calculated based on formula (1). For each normal sleeper image sample in the sample set Perform decentralized computation to obtain decentralized samples: The final decentralized sleeper sample set is obtained as follows: This method completely eliminates global brightness shift interference in sleeper images caused by factors such as lighting, camera exposure, and shooting environment. This ensures that subsequent covariance matrix calculations and feature extraction focus solely on the core features of the sleeper, such as its texture, contour, and surface morphology, rather than irrelevant brightness differences.

[0035] In step 2), the covariance matrix is ​​calculated according to formula (2). The decentralized sleeper sample set obtained in step 1) is used... Substitute directly into formula (2) to complete the covariance matrix. The calculation yields the dimension as covariance matrix The aim is to quantify the linear correlation between high-dimensional pixels in sleeper images, distinguish strongly correlated pixels representing core sleeper features from weakly correlated pixels representing noise, and provide core data support for subsequent extraction of the optimal feature vector.

[0036] In step 3), the eigenvalues ​​and eigenvectors of the covariance matrix are calculated, corresponding to formulas (3) to (7). The covariance matrix obtained in step 2) Perform eigenvalue decomposition and solve for the results. All eigenvalues ​​and their corresponding eigenvectors, with the eigenvalues ​​and eigenvectors satisfying the following relationship: The goal is to identify eigenvalues ​​that characterize the magnitude of core features in a sleeper image, and eigenvectors that can capture the core features of the sleeper. The larger the eigenvalue, the more core features of the sleeper the corresponding eigenvector can retain, and the lower the noise.

[0037] Step 4) corresponds to formulas (4) and (5), and the operation is as follows: Sort all the feature values ​​obtained in step 3) in descending order to obtain the sorted feature values: The corresponding feature vectors are synchronously sorted as follows ; The K largest eigenvalues ​​are selected based on their cumulative contribution rate. The formula for the cumulative contribution rate is: The value range is 95%~99%; Select the K eigenvectors corresponding to the first K eigenvalues, use these K eigenvectors as row vectors, and combine them to construct a new eigenvector matrix. .

[0038] The aim is to eliminate redundant noise and retain core features. Feature vectors corresponding to small feature values ​​only carry noise information such as dust, water stains, and scratches in the sleeper image and are removed during screening. Feature vectors corresponding to large feature values ​​carry core features such as the sleeper's outline and complete surface texture and are retained after screening. Through this step, the high-dimensional sleeper image features are compressed into a K-dimensional low-dimensional feature space, achieving efficient dimensionality reduction while preserving core features to the maximum extent without information loss.

[0039] Step 5) The inverse operation of formula (4) and formula (1) is performed as follows: Preprocessing involves analyzing the images of the sleepers to be inspected. (High-dimensional pixel vector) Substituting the mean vector into formula (1) Decentralization is achieved: ; Low-dimensional projection involves transforming the decentralized sleeper image to be detected into a new feature vector matrix constructed in step 4). Substituting into the linear transformation logic of formula (4), the projection from high dimension to low dimension is completed, and the low-dimensional feature vector is obtained. This low-dimensional feature vector contains only the core features of a normal sleeper and has no redundant noise. Reverse reconstruction involves performing the inverse linear transformation on the low-dimensional feature vectors, mapping them back to the original high-dimensional space, and then using the mean vector from formula (1). After completing the mean restoration, the reconstructed sleeper reference image vector is obtained: ; Image restoration involves reconstructing high-dimensional pixel vectors. The image is then stretched back into a two-dimensional image matrix, ultimately yielding a normal sleeper reference image free of defects, noise, and redundant interference. .

[0040] The aim is to reconstruct the sleeper reference image through dimensionality reduction and feature extraction. It is the common and optimal feature benchmark for all normal sleepers, eliminating interference from dust, light, scratches, and other factors, as well as abnormal features of sleeper block defects. It is the only accurate reference benchmark for image comparison in subsequent sleeper block detection.

[0041] 2. Image processing of the sleeper to be detected (e.g.) Figure 2 (Left image) The core significance of performing Gaussian filtering on the acquired images of the sleepers to be inspected is to effectively suppress image noise and smooth image details while preserving the key texture features of the sleepers. This provides high-quality image data for subsequent LLC benchmark image comparison, threshold segmentation, and defect identification, avoiding detection errors caused by noise interference.

[0042] 3. Image comparison The preprocessed image of the sleeper to be detected is compared with the reference image by performing an absolute difference operation to obtain a comparison image, including: To compare the images, The image is of the sleeper to be inspected. This is the baseline image.

[0043] 4. Disease identification Binarization thresholding was performed on the comparison image using the following thresholding formula: Convert the grayscale difference image into a black-and-white binary image containing only pixel values ​​0 and 65535. Pixel value 0 indicates a normal sleeper area, while pixel value 65535 indicates a suspected defect area where the grayscale difference exceeds the threshold.

[0044] Binary image after threshold segmentation An eight-neighbor connected component labeling algorithm is used to traverse the entire image. All white pixels with a pixel value of 65535 in the binary image are labeled as connected regions. For each independent white connected region, the total number of pixels contained in it is counted. Combined with the preset resolution of the inspection imaging system, the counted number of pixels is converted into the actual physical area, and a set threshold is set. The actual physical area of ​​the calculated connected region is compared with the set threshold.

[0045] Connected regions whose actual physical area exceeds a set threshold are identified as sleeper chipping defects. The outline of these defects is marked on the original sleeper image to be inspected (e.g., ...). Figure 3 As shown in the figure, the defect detection result is 0; while the connected area whose actual physical area does not exceed the set threshold is determined to be without sleeper defects, and an unmarked normal sleeper image is output, and the detection result is 1.

[0046] In some embodiments, the preset method is the SC encoding method.

[0047] In some embodiments, performing data dimensionality reduction and data model generation on the data dictionary using the PCA algorithm includes: The sample data of the normal sleeper image set is decentered to eliminate global offset interference; Based on the decentralized sample data, the covariance matrix describing the linear correlation of pixel dimensions is calculated; The covariance matrix is ​​decomposed into eigenvalues ​​to obtain the corresponding set of eigenvalues ​​and set of eigenvectors. Based on the set of eigenvalues, the eigenvalues ​​are sorted from largest to smallest, and the eigenvalues ​​in the sorted set of eigenvalues ​​are filtered; in the filtered set of eigenvalues, the K largest eigenvalues ​​and the K corresponding eigenvectors are selected, and a new eigenvector matrix is ​​formed by using all the selected eigenvectors as row vectors. The original image is reconstructed based on the new feature vector matrix to obtain the reference image.

[0048] In some embodiments, filtering feature values ​​in the sorted feature value set includes: The feature values ​​are selected based on the cumulative contribution rate, which is calculated using the following formula: The cumulative contribution rate of the eigenvalues. These are the sorted feature values. The principal component dimensions after filtering. This refers to the original data dimension.

[0049] In some embodiments, the selection of feature values ​​based on the cumulative contribution rate includes: A preset contribution rate threshold is used, and the corresponding feature value is retained only when the cumulative contribution rate is not less than the preset contribution rate threshold.

[0050] In some embodiments, preprocessing the sleeper image to be detected includes: applying Gaussian filtering to the sleeper image to be detected.

[0051] In some embodiments, performing an absolute difference operation between the preprocessed sleeper image to be detected and the reference image to obtain a comparison image includes: To compare the images, The image is of the sleeper to be inspected. This is the baseline image.

[0052] In some embodiments, threshold segmentation of the comparison images includes: For the binarized image after thresholding, in coordinates Pixel value at; To compare images in coordinates Pixel value at; This is the preset pixel grayscale threshold; when When this occurs, it is determined to be a normal area; when At that time, it was determined to be an area suspected of being affected by chipping.

[0053] In some embodiments, calculating the area of ​​connected regions in the segmented image and identifying regions with areas exceeding a set threshold as patchy disease areas includes: The area of ​​connected regions is calculated for suspected fragmented defect areas after segmentation, and the pixel area of ​​each connected region is counted; suspected fragmented defect areas with pixel areas exceeding a set threshold are identified as fragmented defect areas.

[0054] In some embodiments, marking the corresponding area and outputting the disease detection results includes: After completing the defect analysis, the system outputs a feedback result image with the outline of the missing block area, and simultaneously outputs the corresponding detection result of 0; and an image of a normal sleeper without markings, and simultaneously outputs the corresponding detection result of 1. A test result of 0 indicates that the sleepers in the corresponding area have a blockage defect; a test result of 1 indicates that the sleepers in the corresponding area do not have a blockage defect.

[0055] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computing software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0056] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0057] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A track sleeper spalling defect detection system based on PCA technology, characterized in that, include: The image acquisition module, fixed to the track defect detection device, is used to acquire original images of the sleepers as images of the sleepers to be inspected. The sleeper block falling image analysis module is used to realize the construction of reference image, processing of sleeper image to be detected, image comparison and execution of defect judgment algorithm; The benchmark image construction algorithm includes: acquiring a set of normal sleeper images and extracting a data dictionary from the set of normal sleeper images using a preset method; performing data dimensionality reduction and data model generation on the data dictionary using the PCA algorithm; and performing projection transformation based on the sleeper image to be detected to obtain a benchmark image that is correlated with the sleeper image to be detected. The image processing algorithm for the sleeper to be detected includes: preprocessing the image of the sleeper to be detected; The image comparison algorithm includes: performing an absolute difference operation between the preprocessed image of the sleeper to be detected and the reference image to obtain a comparison image; The disease determination algorithm includes: performing threshold segmentation on the comparison image, calculating the area of ​​connected regions in the segmented image, determining regions with areas exceeding a set threshold as diseased areas, and labeling the corresponding regions and outputting the disease detection results.

2. The track sleeper deterioration detection system based on PCA technology according to claim 1, characterized in that, The preset method is the SC encoding method.

3. The track sleeper deterioration detection system based on PCA technology according to claim 1, characterized in that, The data dictionary is subjected to dimensionality reduction and data model generation using the PCA algorithm, including: The sample data of the normal sleeper image set is decentered to eliminate global offset interference; Based on the decentralized sample data, the covariance matrix describing the linear correlation of pixel dimensions is calculated; The covariance matrix is ​​decomposed into eigenvalues ​​to obtain the corresponding set of eigenvalues ​​and set of eigenvectors. Based on the set of eigenvalues, the eigenvalues ​​are sorted from largest to smallest, and the eigenvalues ​​in the sorted set of eigenvalues ​​are filtered; in the filtered set of eigenvalues, the K largest eigenvalues ​​and the K corresponding eigenvectors are selected, and a new eigenvector matrix is ​​formed by using all the selected eigenvectors as row vectors. The original image is reconstructed based on the new feature vector matrix to obtain the reference image.

4. The track sleeper deterioration detection system based on PCA technology according to claim 3, characterized in that, Filtering the feature values ​​in the sorted feature value set includes: The feature values ​​are selected based on the cumulative contribution rate, which is calculated using the following formula: The cumulative contribution rate of the eigenvalues. These are the sorted feature values. The principal component dimensions after filtering. This refers to the original data dimension.

5. The track sleeper deterioration detection system based on PCA technology according to claim 4, characterized in that, The selection of eigenvalues ​​based on cumulative contribution rate includes: A preset contribution rate threshold is used, and the corresponding feature value is retained only when the cumulative contribution rate is not less than the preset contribution rate threshold.

6. The track sleeper debriding detection system based on PCA technology according to claim 1, characterized in that, Preprocessing the image of the sleeper to be detected includes: applying Gaussian filtering to the image of the sleeper to be detected.

7. The track sleeper deterioration detection system based on PCA technology according to claim 1, characterized in that, The preprocessed image of the sleeper to be detected is compared with the reference image by performing an absolute difference operation to obtain a comparison image, including: To compare the images, The image is of the sleeper to be inspected. This is the baseline image.

8. The track sleeper deterioration detection system based on PCA technology according to claim 1, characterized in that, Thresholding segmentation of the compared images includes: For the binarized image after thresholding, in coordinates Pixel value at; To compare images in coordinates Pixel value at; This is the preset pixel grayscale threshold; when When this occurs, it is determined to be a normal area; when At that time, it was determined to be an area suspected of being affected by chipping.

9. A track sleeper deterioration detection system based on PCA technology according to claim 8, characterized in that, Calculate the area of ​​connected regions in the segmented image, and determine the regions with areas exceeding a set threshold as areas of missing blocks or defects, including: The area of ​​connected regions is calculated for suspected fragmented defect areas after segmentation, and the pixel area of ​​each connected region is counted; suspected fragmented defect areas with pixel areas exceeding a set threshold are identified as fragmented defect areas.

10. A track sleeper deterioration detection system based on PCA technology according to claim 1, characterized in that, The process of marking the corresponding areas and outputting the disease detection results includes: After completing the defect analysis, the system outputs a feedback result image with the outline of the missing block area, and simultaneously outputs the corresponding detection result of 0; and an image of a normal sleeper without markings, and simultaneously outputs the corresponding detection result of 1. A test result of 0 indicates that the sleepers in the corresponding area have a blockage defect; a test result of 1 indicates that the sleepers in the corresponding area do not have a blockage defect.

Images