A fastener model detection and identification method and system for a ballastless track

Abstract

The application provides a fastener model detection and identification method and system for a ballastless track, and the method comprises the following steps: presetting a coding rule, replacing a character model identifier of a fastener according to the coding rule, obtaining a corresponding geometric coding identifier, and pressing the geometric coding identifier on a corresponding identification area of the fastener; collecting an image of the identification area of the fastener, pre-processing the image, and obtaining a standard image set; training the standard image set based on a neural network model, obtaining an image recognition model, and performing feature recognition on a standard image to be identified according to the image recognition model; and outputting a model identifier of the corresponding fastener according to an identification result and the coding rule. The method converts fastener type and thickness value information of the model identifier of the fastener into the corresponding geometric coding identifier through the preset coding rule, facilitates subsequent image recognition, effectively guarantees the effectiveness of the identification, and improves the efficiency and accuracy of model training through pre-processing of the image, so that the required fastener model can be accurately and quickly found.

Description

Technical Field

[0001] This invention relates to the field of ballastless track fastener model identification technology, and in particular to a method and system for detecting and identifying ballastless track fastener models. Background Technology

[0002] Ballastless track, also known as ballastless roadbed, uses an integral track slab. With high-strength fasteners and fixing facilities, it firmly fixes the track to a foundation such as concrete or steel sleepers, achieving high track stability and load-bearing capacity. This fastener system not only significantly reduces track movement and deformation, but also improves the safety of railway operation.

[0003] A method for detecting track fasteners based on computer image and vision recognition, disclosed in CN105574872B, utilizes common computer vision models to design a suitable hardware configuration for the system. It calculates parameters to meet the required accuracy of the image acquisition system, selects appropriate components, and builds a hardware platform for the track detection system. For the image data acquired by the hardware system, digital image processing technology is used for analysis and processing. Key technologies such as track image target area extraction, rail surface defect identification and classification, and fastener feature recognition are emphasized and developed. Finally, based on the processing results, various defects in track components are identified and reported to railway staff for maintenance reference.

[0004] However, the model numbers of existing ballastless track fastener systems are identified by text. Because the text is small, it is easily obscured by dust, rust, etc., making it difficult to identify the model numbers of already installed gaskets and blocks. Furthermore, the image size and processing quality are poor during image processing, resulting in inaccurate or even impossible image acquisition and recognition in subsequent image acquisition. Consequently, it is impossible to accurately and quickly find the required fastener model, thus reducing the efficiency of maintenance work. Summary of the Invention

[0005] In view of this, the present invention proposes a method and system for detecting and identifying fastener models for ballastless tracks. By using preset coding rules, the fastener type and thickness numerical information of the fastener model identifier are converted into corresponding geometric code identifiers, which facilitates subsequent image recognition and effectively ensures the effectiveness of recognition. Furthermore, by preprocessing the images, the efficiency and accuracy of model training can be improved, thereby enabling the required fastener model to be found accurately and quickly.

[0006] The technical solution of the present invention is implemented as follows: In a first aspect, the present invention provides a method for detecting and identifying the fastener model of ballastless track, comprising the following steps:

[0007] S1, preset coding rules, replace the text model mark of the fastener according to the coding rules, obtain the corresponding geometric code mark, and press it on the corresponding mark area of ​​the fastener;

[0008] S2, acquire images of the fastener marking area, preprocess the images, and obtain a standard image set;

[0009] S3, based on the neural network model, trains the standard image set to obtain the image recognition model, and performs feature recognition on the standard image to be recognized according to the image recognition model;

[0010] S4, based on the identification results and coding rules, outputs the model identifier of the corresponding fastener.

[0011] Based on the above technical solution, preferably, the preset coding rule in step S1 replaces the text model identifier of the fastener to obtain the corresponding geometric code identifier. The fastener model of the ballastless track includes thickness numerical information. The geometric code identifier includes circular, triangular, rectangular, pentagram, rhombus, and pentagonal shapes. The geometric code identifier is associated with the thickness numerical information: a circle represents a value of 5, a triangle represents a value of 3, a rectangle represents a value of 2, a pentagram represents a value of 1, a rhombus represents a value of 0.5, and a pentagon represents a value of 10. Multiple corresponding shapes are obtained from the geometric code identifier shapes based on the thickness numerical information, and the sum of the corresponding values ​​of these multiple shapes equals the thickness numerical information. The geometric code identifier shapes corresponding to the text model identifier of the fastener are combined sequentially to obtain the geometric code identifier corresponding to the fastener model identifier.

[0012] Based on the above technical solutions, preferably, in step S1... Divide the marking area of ​​the fastener into Multiple sub-marking areas of equal size are pressed into the center of each sub-marking area according to the geometric code markings corresponding to the fastener model markings.

[0013] Based on the above technical solutions, preferably, step S2 includes the following sub-steps:

[0014] S21, the camera takes multiple images of each fastener marking area to obtain image data under different ambient lighting and dirt conditions, and the image of the fastener marking area is located at the center of the image.

[0015] S22, the Canny edge detection algorithm is used to segment and extract the fastener identification area of ​​each image data, and then the image is cropped according to the five sub-identification areas to obtain the sub-region images;

[0016] S23, construct a dynamic adjustment model based on the network model, and perform image enhancement processing on each sub-region image according to the dynamic adjustment model to obtain a standard block image;

[0017] S24, the image feature information of each standard block image under different ambient lighting and dirt conditions is fused, and the fused feature data in the same fastener identification area is spliced ​​together, and corresponding identification labels are set to obtain a standard image set.

[0018] Based on the above technical solutions, preferably, step S23, which involves constructing a dynamically adjusted model based on a network model, includes the following sub-steps:

[0019] A database is constructed that stores historical images of each sub-region under different ambient lighting conditions;

[0020] Historical images of each sub-region under different ambient lighting conditions are acquired. These images are then processed to obtain grayscale images. The grayscale images are further divided into multiple overlapping sub-blocks. The size of each sub-region's historical image is set to W×H, the size of each sub-block to S×S, and the overlap pixel value to 0. The number of sub-blocks and the starting position information of each sub-block are calculated.

[0021] The number of sub-blocks in the horizontal direction is expressed as:

[0022] N x =W / (SO);

[0023] The number of sub-blocks in the vertical direction is expressed as:

[0024] N y =H / (SO);

[0025] The starting position information of each sub-block is expressed as follows:

[0026] P(x, y)=(x·(SO), y·(SO));

[0027] In the formula, (x, y) is the position index of the sub-block, and the value of x ranges from 0 to N. x -1, y ranges from 0 to N y -1;

[0028] Based on the starting position information and size of each sub-block, images of each sub-block are extracted from the historical images of the sub-region, and the histogram and cumulative distribution of the histogram of each sub-block are calculated.

[0029] The histogram expressions for each sub-block are as follows:

[0030]

[0031] In the formula, H i Let B be the histogram of the i-th sub-block. i Let g be the i-th sub-block, g be the gray value, and δ be the Dirac function;

[0032] The cumulative distribution expression of a histogram is:

[0033]

[0034] In the formula, H i (j) represents the i-th sub-block B i The number of pixels with gray value j in the grayscale histogram;

[0035] Set a contrast limit value, and crop the histogram of each sub-block image according to the contrast limit value. The expression is:

[0036]

[0037] In the formula, clip is the contrast limit value;

[0038] The cropped histogram is redistributed, and the total number of pixels is kept constant according to the linear mapping relationship to obtain a sub-block equalized image;

[0039] The equalized images of each sub-block are evaluated based on visual quality assessment, and the size and contrast limits of the sub-blocks are calibrated to generate a training set of sub-block images.

[0040] A dynamic adjustment model is constructed based on a CNN neural network model. Data from the sub-block image training set is input into the CNN neural network model for iterative training to obtain the trained dynamic adjustment model, which is used to output the prediction results of sub-block size and contrast limit values.

[0041] Based on the above technical solutions, preferably, step S23, which involves performing image enhancement processing on each sub-region image to obtain a standard block image, includes the following sub-steps:

[0042] Each sub-region image is input into the dynamic adjustment model. Based on the prediction results of the sub-block size and contrast limit value output by the dynamic adjustment model, the corresponding sub-region image is divided and the image is enhanced according to the prediction results of the output sub-block size and contrast limit value to obtain the corresponding sub-block equalized image.

[0043] All sub-block equalized images are stitched together according to the positions of the corresponding sub-region images, and bilinear interpolation is used to smooth the pixels in the overlapping areas to obtain the standard block image. The bilinear interpolation expression is:

[0044]

[0045] In the formula, I smoothing (x,y) represents the smoothed pixel value, I(B) i (x,y)) and I(B i+1(x,y) represent the pixel values ​​of the corresponding pixel points in the equalized images of two adjacent sub-blocks.

[0046] Based on the above technical solutions, preferably, in step S4, the corresponding fastener model identifier is output according to the recognition result and coding rules. Specifically, the corresponding geometric code identifier in the image is obtained from the coding rules based on the recognized geometric code identifier. The corresponding values ​​are accumulated to obtain and output the thickness value information of the corresponding fastener model.

[0047] Secondly, the present invention also provides a fastener model detection and identification system for ballastless tracks, implemented using the aforementioned fastener model detection and identification method for ballastless tracks, the system comprising:

[0048] The setting module is used to preset the coding rules, replace the text model mark of the fastener according to the coding rules, obtain the corresponding geometric code mark, and press it on the corresponding mark area of ​​the fastener;

[0049] The acquisition module is used to acquire images of the fastener marking area, preprocess the images, and obtain a standard image set;

[0050] The recognition module is used to train a standard image set based on a neural network model to obtain an image recognition model, and to perform feature recognition on the standard image to be recognized based on the image recognition model.

[0051] The output module is used to output the model identifier of the corresponding fastener based on the recognition result and coding rules.

[0052] Thirdly, the present invention also provides an electronic device, including at least one processor, at least one memory, a communication interface, and a bus; wherein the processor, memory, and communication interface communicate with each other through the bus; the memory stores a program for detecting and identifying the fastener type of ballastless track, which can be executed by the processor, and the program for detecting and identifying the fastener type of ballastless track is configured to implement the fastener type detection and identification method for ballastless track as described above.

[0053] Fourthly, the present invention also provides a computer-readable storage medium storing a method program for detecting and identifying the fastener model of a ballastless track, wherein when the method program is executed, it implements the method for detecting and identifying the fastener model of a ballastless track as described above.

[0054] The fastener model detection and identification method and system for ballastless tracks of the present invention have the following advantages over the prior art:

[0055] (1) By using preset coding rules, the thickness value information of the fastener model identification is converted into the corresponding geometric code identification, and the corresponding geometric code identification is pressed into each sub-identification area, avoiding the identification error caused by wear or blurring of text or number identification, which facilitates subsequent image recognition and effectively ensures the effectiveness of recognition. By preprocessing the image and constructing a standard image set, the efficiency and accuracy of model training are improved. By combining the neural network model for feature recognition, the accuracy of information is improved. The required fastener model can be found accurately and quickly, so as to facilitate replacement or repair work, improve maintenance efficiency, and reduce maintenance costs.

[0056] (2) By collecting image data under different lighting and dirt conditions multiple times, the Canny edge detection algorithm is used for segmentation and extraction. A dynamic adjustment model is built based on the network model for image enhancement processing. Finally, the image feature information under different lighting conditions and different dirt conditions is fused and stitched into a standard image set, which can obtain a clearer and higher contrast standard image set, providing a more reliable data foundation for subsequent image recognition. By fusing image feature information under different lighting conditions, the accuracy of image recognition is improved and the recognition error caused by changes in lighting is reduced.

[0057] (3) By constructing a dynamic adjustment model to dynamically adjust the size and contrast limit of sub-blocks, the local contrast of the image is improved. The parameters of image enhancement processing can be adaptively adjusted according to different image features to obtain a clearer and higher contrast standard block image. By overlapping sub-blocks and edge smoothing, the visible seams generated during splicing are greatly reduced, the image quality is enhanced, and the accuracy of image recognition is improved. Attached Figure Description

[0058] To more clearly illustrate the technical solutions in the embodiments of the present invention 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 some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0059] Figure 1 This is a flowchart of the fastener model detection and identification method for ballastless track according to the present invention;

[0060] Figure 2 This is a schematic diagram of the hardware operating environment of the device involved in the embodiments of the present invention;

[0061] Figure 3 This is a schematic diagram of the functional modules of the fastener model detection and identification system for ballastless track of the present invention. Detailed Implementation

[0062] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0063] like Figure 1 As shown, the present invention provides a method for detecting and identifying the fastener model of a ballastless track, comprising the following steps:

[0064] S1, preset coding rules, replace the text model mark of the fastener according to the coding rules, obtain the corresponding geometric code mark, and press it on the corresponding mark area of ​​the fastener.

[0065] In step S1, the preset coding rule replaces the textual model identifier of the fastener to obtain the corresponding geometric code identifier. The fastener model of the ballastless track includes thickness numerical information. The geometric code identifier includes circular, triangular, rectangular, pentagram, rhombus, and pentagonal shapes. The geometric code identifier is associated with the thickness numerical information: a circle represents a value of 5, a triangle represents a value of 3, a rectangle represents a value of 2, a pentagram represents a value of 1, a rhombus represents a value of 0.5, and a pentagon represents a value of 10. Multiple corresponding shapes are obtained from the geometric code identifier shapes based on the thickness numerical information, and the sum of the corresponding values ​​of these multiple shapes equals the thickness numerical information. The geometric code identifier shapes corresponding to the textual model identifier of the fastener are combined in sequence to obtain the geometric code identifier corresponding to the fastener model identifier.

[0066] It's important to note that the correspondence between fastener types and geometric codes is intuitive. Each fastener type corresponds to a unique geometric shape. For geometric shapes with advantages, larger shapes and more prominent features are easier to identify, even if rusted or covered in dust, facilitating quick fastener model identification. However, for thickness information encoding, common thickness values ​​(such as 0.5, 1, 2, 3, 5, 10, etc.) are first matched with specific geometric codes. When the thickness value is not one of these common values, multiple geometric codes need to be combined to achieve the desired value. For example, if the thickness value is 7.5, it can be decomposed into 5 (circle) + 2 (rectangle) + 0.5 (rhombus), and these geometric shapes are then combined sequentially.

[0067] In this embodiment, the thickness numerical information of the fastener model identifier is converted into a corresponding geometric code identifier through a preset coding rule. Specifically, the thickness numerical information of each fastener model is assigned one or more specific geometric shapes. These geometric shapes are combined in numerical order and pressed onto the identification area of ​​the fastener. The thickness numerical information of the fastener model is represented by the geometric code identifier. Since the geometric code identifier is easy to identify and distinguish, it facilitates subsequent image recognition and effectively ensures the effectiveness of the recognition. At the same time, the geometric code identifier system avoids recognition errors that may be caused by wear or blurring of text or numerical identifiers, improving the accuracy of information. It can accurately and quickly find the required fastener model, thereby facilitating replacement or repair work; it can not only improve maintenance efficiency but also reduce maintenance costs.

[0068] In step S1, the marking area of ​​the fastener is divided into multiple sub-marking areas of equal size, and the geometric code markings corresponding to the model marking of the fastener are pressed into the center of each sub-marking area.

[0069] It should be noted that by dividing the fastener markings into multiple sub-areas and pressing geometric coding markings at the center of each area, the information display becomes clearer and more intuitive, improving readability.

[0070] Because it has more geometric features than text and a larger marking area, the markings remain clearly visible even after installation, even when covered by rust or dust. Therefore, when collecting data from the target area, it is important to capture not only images under different lighting conditions but also images showing signs of rust and dust. Due to its distinct geometric features and large geometric area, it is easier to identify, whether through image learning or manual judgment.

[0071] In addition, during the identification of shims in the fastener model detection of ballastless track, multiple shims of different sizes overlap, making identification impossible. Therefore, in this embodiment, the marking areas of the overlapping shims are staggered to ensure that the marking area of ​​each shim can be identified, thus ensuring the accuracy of identification.

[0072] S2: Acquire images of the fastener marking area, preprocess the images, and obtain a standard image set.

[0073] Step S2 includes the following sub-steps:

[0074] S21, the camera takes multiple images of each fastener marking area to obtain image data under different ambient lighting and dirt conditions, and the image of the fastener marking area is located at the center of the image.

[0075] It should be noted that a high-resolution, low-noise camera was selected, and a fixed shooting angle and distance were set to ensure that the size and proportion of each captured image remained consistent. To simulate various lighting conditions that may be encountered in actual use, multiple images were captured at different times, such as morning, noon, and evening, and under different light intensities. At the same time, professional lighting equipment was used to simulate extreme lighting conditions, such as strong light and low light environments. During the acquisition process, it was ensured that the image of the fastener marking area was always located in the center of the image to reduce errors caused by image offset.

[0076] S22, the Canny edge detection algorithm is used to segment and extract the fastener identification area of ​​each image data, and then the image is cropped according to the five sub-identification areas to obtain the sub-region images.

[0077] It should be noted that the Canny edge detection algorithm is used. This algorithm can accurately detect edge information in the image, thereby accurately segmenting the fastener identification area. After segmenting the fastener identification area, it is cropped according to five preset sub-regions to obtain the image data of each sub-region, and then associated according to the sequence number.

[0078] S23, construct a dynamic adjustment model based on the network model, and perform image enhancement processing on each sub-region image according to the dynamic adjustment model to obtain a standard block image;

[0079] In this embodiment, a dynamic adjustment model is constructed based on the output of the network model. This model can automatically adjust the parameters of image enhancement processing according to the characteristics of each sub-region image to obtain a clearer and higher contrast standard block image.

[0080] Specifically, step S23 includes the following sub-steps:

[0081] A database is constructed that stores historical images of each sub-region under different ambient lighting conditions;

[0082] Historical images of each sub-region under different ambient lighting conditions are acquired. These images are then processed to obtain grayscale images. The grayscale images are further divided into multiple overlapping sub-blocks. The size of each sub-region's historical image is set to W×H, the size of each sub-block to S×S, and the overlap pixel value to 0. The number of sub-blocks and the starting position information of each sub-block are calculated.

[0083] The number of sub-blocks in the horizontal direction is expressed as:

[0084] N x =W / (SO);

[0085] The number of sub-blocks in the vertical direction is expressed as:

[0086] N y =H / (SO);

[0087] The starting position information of each sub-block is expressed as follows:

[0088] P(x, y)=(x·(SO), y·(SO));

[0089] In the formula, (x, y) is the position index of the sub-block, and the value of x ranges from 0 to N. x -1, y ranges from 0 to N y -1;

[0090] Based on the starting position information and size of each sub-block, images of each sub-block are extracted from the historical images of the sub-region, and the histogram and cumulative distribution of the histogram of each sub-block are calculated.

[0091] The histogram expressions for each sub-block are as follows:

[0092]

[0093] In the formula, H i Let B be the histogram of the i-th sub-block. i Let g be the i-th sub-block, g be the gray value, and δ be the Dirac function;

[0094] The cumulative distribution expression of a histogram is:

[0095]

[0096] In the formula, H i (j) represents the i-th sub-block B i The number of pixels with gray value j in the grayscale histogram;

[0097] Set a contrast limit value, and crop the histogram of each sub-block image according to the contrast limit value. The expression is:

[0098]

[0099] In the formula, clip is the contrast limit value;

[0100] The cropped histogram is redistributed, and the total number of pixels is kept constant according to the linear mapping relationship to obtain a sub-block equalized image;

[0101] The equalized images of each sub-block are evaluated based on visual quality assessment, and the size and contrast limits of the sub-blocks are calibrated to generate a training set of sub-block images.

[0102] A dynamic adjustment model is constructed based on a CNN neural network model. Data from the sub-block image training set is input into the CNN neural network model for iterative training to obtain the trained dynamic adjustment model, which is used to output the prediction results of sub-block size and contrast limit values.

[0103] It should be noted that by constructing a dynamic adjustment model to dynamically adjust the size and contrast limits of sub-blocks, the local contrast of the image is improved without introducing excessive noise. Furthermore, the parameters of the image enhancement processing can be adaptively adjusted according to different image features to obtain clearer and higher contrast standard block images.

[0104] The process involves evaluating the equalization of each sub-block image based on visual quality assessment. Specifically, the visual quality assessment includes sharpness evaluation, contrast evaluation, and noise evaluation metrics. The Laplacian operator is used to evaluate the sharpness of the equalization sub-block image, resulting in a sharpness evaluation metric. A sharpness threshold is set, and images with sharpness exceeding the threshold are selected. The Michelson contrast algorithm is used to calculate the contrast of the equalization sub-block image, and a contrast threshold is set, selecting images with moderate contrast. A noise evaluation algorithm is used to detect the noise level in the equalization sub-block image, and a noise threshold is set, removing images with excessive noise. A weighted calculation is performed based on the sharpness evaluation, contrast evaluation, and noise evaluation metrics to obtain a comprehensive score. A preset calibration threshold is then used to calibrate the sub-blocks with comprehensive scores exceeding the calibration threshold, specifying the size and contrast limit parameters.

[0105] In addition, step S23, which describes performing image enhancement processing on each sub-region image to obtain a standard block image, includes the following sub-steps:

[0106] Each sub-region image is input into the dynamic adjustment model. Based on the prediction results of the sub-block size and contrast limit value output by the dynamic adjustment model, the corresponding sub-region image is divided and the image is enhanced according to the prediction results of the output sub-block size and contrast limit value to obtain the corresponding sub-block equalized image.

[0107] All sub-block equalized images are stitched together according to the positions of the corresponding sub-region images, and bilinear interpolation is used to smooth the pixels in the overlapping areas to obtain the standard block image. The bilinear interpolation expression is:

[0108]

[0109] In the formula, I smoothing (x,y) represents the smoothed pixel value, I(B) i (x,y)) and I(B i+1 (x,y) represent the pixel values ​​of the corresponding pixel points in the equalized images of two adjacent sub-blocks.

[0110] It should be noted that the images of each sub-region are input into the pre-trained dynamic adjustment model. Based on the predictive ability of the dynamic adjustment model, the model outputs the predicted results of the sub-block size and contrast limit value corresponding to each sub-region image. These prediction results will guide the subsequent image segmentation and enhancement processing, ensuring that each sub-region image can receive the enhancement processing most suitable for its features. According to the predicted results of the sub-block size and contrast limit value output by the dynamic adjustment model, the corresponding sub-region images are segmented and enhanced to obtain a series of enhanced sub-block equalized images. All sub-block equalized images are stitched together according to the position of their corresponding sub-region images in the original image. Since there is overlap between sub-blocks, it is necessary to smooth the pixels in the overlapping areas. Through overlapping sub-blocks and edge smoothing, the visible seams generated during stitching are greatly reduced, the image quality is enhanced, and thus the accuracy of image recognition is improved.

[0111] S24, the image feature information of each standard block image under different ambient lighting and dirt conditions is fused, and the fused feature data in the same fastener identification area is spliced ​​together, and corresponding identification labels are set to obtain a standard image set.

[0112] In this embodiment, image data under different ambient lighting and dirt conditions are collected multiple times. The Canny edge detection algorithm is used for segmentation and extraction, and a dynamic adjustment model is built based on a network model for image enhancement. Finally, the image feature information under different lighting conditions is fused and stitched into a standard image set. Through multiple acquisitions and image enhancement processing, a clearer and higher contrast standard image set can be obtained, providing a more reliable data foundation for subsequent image recognition. By fusing image feature information under different lighting conditions, the accuracy of image recognition is improved and the recognition error caused by changes in lighting is reduced.

[0113] S3, based on the neural network model, trains the standard image set to obtain the image recognition model, and performs feature recognition on the standard image to be recognized according to the image recognition model;

[0114] It should be noted that an image recognition model is obtained by training a standard image set based on a convolutional neural network. The convolutional neural network includes an input layer, a convolutional layer, a pooling layer, and a fully connected layer. The hyperparameters of the model are determined, and the model is trained using a training set. The model parameters are updated through the backpropagation algorithm. The trained image recognition model is loaded into memory for feature recognition. The standard images to be recognized are preprocessed in the same way as during training. The model is used to extract features from the preprocessed images to identify key information in the images and output the corresponding geometric code identifiers in the images.

[0115] S4, based on the identification results and coding rules, outputs the model identifier of the corresponding fastener.

[0116] In step S4, the model identifier of the corresponding fastener is output according to the recognition result and the coding rule. The corresponding geometric code identifier in the image is obtained from the coding rule according to the recognized geometric code identifier. The corresponding values ​​are accumulated to obtain and output the thickness value information of the corresponding fastener model.

[0117] In this embodiment, a pre-defined encoding rule is used to replace the fastener model identifier with the corresponding geometric code identifier, and these geometric code identifiers are pressed onto the corresponding identification area of ​​the fastener. By acquiring images of the fastener identification area and performing preprocessing, a standard image set is constructed. The standard image set is trained using a neural network model to obtain an image recognition model capable of recognizing fastener features. When recognizing a new fastener image, the trained image recognition model is used to perform feature recognition on the image, and the corresponding fastener model identifier is output according to the recognition result and the pre-defined encoding rule. By combining the neural network model with the pre-defined encoding rule, fastener model identifier recognition is achieved quickly and accurately. By preprocessing the images and constructing the standard image set, the efficiency and accuracy of model training are improved. Using the neural network model for feature recognition not only improves the recognition accuracy but also enhances the robustness and adaptability of the system. Furthermore, by replacing the model identifier with the geometric code identifier through the pre-defined encoding rule, the recognition process is simplified, the error rate is reduced, and the recognition accuracy is further improved.

[0118] like Figure 3 As shown, in a second aspect, the present invention also provides a fastener model detection and identification system for ballastless tracks, implemented using the aforementioned fastener model detection and identification method for ballastless tracks. The system includes:

[0119] The setting module is used to preset the coding rules, replace the text model mark of the fastener according to the coding rules, obtain the corresponding geometric code mark, and press it on the corresponding mark area of ​​the fastener;

[0120] The acquisition module is used to acquire images of the fastener marking area, preprocess the images, and obtain a standard image set;

[0121] The recognition module is used to train a standard image set based on a neural network model to obtain an image recognition model, and to perform feature recognition on the standard image to be recognized based on the image recognition model.

[0122] The output module is used to output the model identifier of the corresponding fastener based on the recognition result and coding rules.

[0123] It should be noted that this system is a system corresponding to the above-mentioned marine engineering equipment weighing and measurement method. All implementation methods in the above method embodiments are applicable to the embodiments of this system and can achieve the same technical effect.

[0124] Those skilled in the art will 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, or a combination of computer software and electronic hardware. 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 implementations should not be considered beyond the scope of this invention.

[0125] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the system and modules described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0126] like Figure 2 As shown in the embodiments provided by this invention, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection between devices or units, and may be electrical, mechanical, or other forms.

[0127] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0128] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0129] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks.

[0130] Furthermore, it should be noted that in the system and method of the present invention, it is obvious that the components or steps can be decomposed and / or recombined. These decompositions and / or recombinations should be considered equivalent solutions of the present invention. Moreover, the steps performing the above series of processes can naturally be executed in the order described, but are not necessarily required to be executed in chronological order; some steps can be executed in parallel or independently of each other. Those skilled in the art will understand that all or any step or component of the method and apparatus of the present invention can be implemented in any computing device (including processors, storage media, etc.) or network of computing devices, in hardware, firmware, software, or a combination thereof. This is something that those skilled in the art can achieve by using their basic programming skills after reading the description of the present invention.

[0131] Therefore, the object of the present invention can also be achieved by running a program or a set of programs on any computing system. The computing system can be a known general-purpose system. Therefore, the object of the present invention can also be achieved simply by providing a program product containing program code implementing the method or apparatus. That is, such a program product also constitutes the present invention, and the storage medium storing such a program product also constitutes the present invention. Obviously, the storage medium can be any known storage medium or any storage medium developed in the future. It should also be noted that in the apparatus and method of the present invention, it is obvious that the components or steps can be decomposed and / or recombined. These decompositions and / or recombinations should be considered equivalent to the present invention. Furthermore, the steps performing the above series of processes can naturally be performed in the order described, but are not necessarily required to be performed in chronological order. Some steps can be performed in parallel or independently of each other.

[0132] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for detecting and identifying the fastener model of a ballastless track, characterized in that, Includes the following steps: S1, preset coding rules, replace the text model mark of the fastener according to the coding rules, obtain the corresponding geometric code mark, and press it on the corresponding mark area of ​​the fastener; The fastener models of the ballastless track all include thickness numerical information. The geometric code identifiers include circular, triangular, rectangular, pentagram, rhombus, and pentagonal shapes. The geometric code identifiers are associated with the thickness numerical information: a circle represents a value of 5, a triangle represents a value of 3, a rectangle represents a value of 2, a pentagram represents a value of 1, a rhombus represents a value of 0.5, and a pentagon represents a value of 10. Based on the thickness numerical information, multiple corresponding shapes are obtained from the geometric code identifier shapes, and the sum of the corresponding values ​​of multiple corresponding shapes equals the thickness numerical information. The geometric code identifier shapes corresponding to the text model identifier of the fastener are combined in sequence to obtain the geometric code identifier corresponding to the model identifier of the fastener. In step S1, the marking area of ​​the fastener is divided into multiple sub-marking areas of equal size, and the geometric code markings corresponding to the model marking of the fastener are pressed into the center of each sub-marking area according to the order of the markings. S2, acquire images of the fastener marking area, preprocess the images, and obtain a standard image set; Step S2 includes the following sub-steps: S21, the camera takes multiple images of each fastener marking area to obtain image data under different ambient lighting and dirt conditions, and the image of the fastener marking area is located at the center of the image. S22, the Canny edge detection algorithm is used to segment and extract the fastener identification area of ​​each image data, and then the image is cropped according to the five sub-identification areas to obtain the sub-region images; S23, construct a dynamic adjustment model based on the network model, and perform image enhancement processing on each sub-region image according to the dynamic adjustment model to obtain a standard block image; The process of building a dynamically adjustable model based on a network model includes the following sub-steps: A database is constructed that stores historical images of each sub-region under different ambient lighting conditions; Historical images of each sub-region under different ambient lighting conditions are acquired. These images are then processed to obtain grayscale images. The grayscale images are further divided into multiple overlapping sub-blocks. The size of each sub-region's historical image is set to W×H, the size of each sub-block to S×S, and the overlap pixel value to 0. The number of sub-blocks and the starting position information of each sub-block are calculated. The number of sub-blocks in the horizontal direction is expressed as: N x = W / ( S- O); The number of sub-blocks in the vertical direction is expressed as: N y = H / ( S- O); The starting position information of each sub-block is expressed as follows: P( x , y )=( x ·(S−O),y·(S−O)); In the formula, ( x , y () is the position index of the sub-block. x The value range is 0 to N x -1, y The value range is 0 to N y -1; Based on the starting position information and size of each sub-block, images of each sub-block are extracted from the historical images of the sub-region, and the histogram and cumulative distribution of the histogram of each sub-block are calculated. The histogram expressions for each sub-block are as follows: ； In the formula, H i For the first i Histogram of each sub-block B i For the first i Each sub-block, where g is the grayscale value. δ It is the Dirac function; The cumulative distribution expression of a histogram is: ； In the formula, H i ( j ) indicates the first i Sub-block B i The grayscale value in the grayscale histogram is j The number of pixels; Set a contrast limit value, and crop the histogram of each sub-block image according to the contrast limit value. The expression is: ； In the formula, clip This is the contrast limit value; The cropped histogram is redistributed, and the total number of pixels is kept constant according to the linear mapping relationship to obtain a sub-block equalized image; The equalized images of each sub-block are evaluated based on visual quality assessment, and the size and contrast limits of the sub-blocks are calibrated to generate a training set of sub-block images. A dynamic adjustment model is constructed based on a CNN neural network model. Data from the training set of sub-block images is input into the CNN neural network model for iterative training to obtain the trained dynamic adjustment model, which is used to output the prediction results of sub-block size and contrast limit values. Image enhancement processing is performed on each sub-region image to obtain a standard block image, including the following sub-steps: Each sub-region image is input into the dynamic adjustment model. Based on the prediction results of the sub-block size and contrast limit value output by the dynamic adjustment model, the corresponding sub-region image is divided and the image is enhanced according to the prediction results of the output sub-block size and contrast limit value to obtain the corresponding sub-block equalized image. All sub-block equalized images are stitched together according to the positions of the corresponding sub-region images, and bilinear interpolation is used to smooth the pixels in the overlapping areas to obtain the standard block image. The bilinear interpolation expression is: ； In the formula, I smoothing ( x , y () represents the smoothed pixel value. I ( B i ( x , y ))and I ( B i+1 ( x , y )) represent the pixel values ​​of the corresponding pixels in the equalized images of two adjacent sub-blocks; S24, fuse the image feature information of each standard block image under different ambient lighting and dirt conditions, stitch together the fused feature data in the same fastener identification area, and set corresponding identification labels to obtain a standard image set; S3, based on the neural network model, trains the standard image set to obtain the image recognition model, and performs feature recognition on the standard image to be recognized according to the image recognition model; S4, based on the identification results and coding rules, outputs the model identifier of the corresponding fastener.

2. The method for detecting and identifying the fastener model of ballastless track as described in claim 1, characterized in that: In step S4, the model identifier of the corresponding fastener is output according to the recognition result and the coding rule. The corresponding geometric code identifier in the image is obtained from the coding rule according to the recognized geometric code identifier. The corresponding values ​​are accumulated to obtain and output the thickness value information of the corresponding fastener model.

3. A fastener model detection and identification system for ballastless track, implemented using a fastener model detection and identification method for ballastless track as described in any one of claims 1 to 2, characterized in that, The system includes: The setting module is used to preset the coding rules, replace the text model mark of the fastener according to the coding rules, obtain the corresponding geometric code mark, and press it on the corresponding mark area of ​​the fastener; The acquisition module is used to acquire images of the fastener marking area, preprocess the images, and obtain a standard image set; The recognition module is used to train a standard image set based on a neural network model to obtain an image recognition model, and to perform feature recognition on the standard image to be recognized based on the image recognition model. The output module is used to output the model identifier of the corresponding fastener based on the recognition result and coding rules.

4. An electronic device, characterized in that, It includes at least one processor, at least one memory, a communication interface, and a bus; wherein the processor, memory, and communication interface communicate with each other through the bus; the memory stores a program for detecting and identifying the fastener type of ballastless track, which can be executed by the processor, and the program for detecting and identifying the fastener type of ballastless track is configured to implement the fastener type detection and identification method for ballastless track as claimed in any one of claims 1 to 2.

5. A computer-readable storage medium, characterized in that, The storage medium stores a method program for detecting and identifying the fastener model of a ballastless track. When the method program is executed, it implements the method for detecting and identifying the fastener model of a ballastless track as described in any one of claims 1 to 2.

Images