A track fastener detection method and system based on deep learning
By using a deep learning-based method for detecting track fasteners, the problems of long time consumption and high cost associated with traditional manual inspections have been solved, achieving safe and efficient fastener status detection.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- BEIJING MASS TRANSIT RAILWAY OPERATION CORPORATION LIMITED
- Filing Date
- 2023-09-26
- Publication Date
- 2026-07-21
AI Technical Summary
Traditional track fastener inspection methods rely on manual inspection, which is time-consuming, costly, and poses safety risks, making it difficult to achieve accurate and rapid inspection.
A deep learning-based track fastener detection method is adopted. By acquiring image training samples for preprocessing and annotation, feature vectors are extracted, a detection model is established, damage values are calculated, and the fastener status is determined by comparing them with real-time images.
It achieves safer, more efficient, and more accurate track fastener inspection than manual inspection, improving the safety and efficiency of the inspection.
Smart Images
Figure CN117455839B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of track fastener inspection technology, and more specifically, relates to a track fastener inspection method and system based on deep learning. Background Technology
[0002] Since the beginning of the 21st century, with the continuous expansion of rail transit, ensuring the safe and stable operation of trains and conducting track inspections has become a prerequisite and foundation for railway and urban rail development. Track fasteners play a crucial role in fixing the rails, maintaining track gauge, and preventing longitudinal and lateral movement of the rails relative to the sleepers. When fasteners shift, partially break, or become missing on the track, they can easily cause instability in the train during operation, leading to accidents and causing significant losses to people's lives and property.
[0003] Therefore, accurate and rapid inspection of track fasteners is of great significance and value in order to ensure the safe operation of rail transit trains. Traditional fastener inspection usually adopts manual inspection, which is time-consuming, costly, and poses safety risks. Summary of the Invention
[0004] To address the above technical issues, this invention proposes a deep learning-based method for detecting track fasteners, comprising:
[0005] Acquire image training samples of track fastener images, preprocess the image training samples to generate preprocessed image training samples, and label the image training samples according to the track fastener status, which includes: fastener missing status, fastener damaged status, and normal fastener status.
[0006] The feature vectors of the images in the image training samples are extracted, a track fastener detection model is set, the feature vectors are input into the track fastener detection model, track fastener targets are detected and the damage values of the track fasteners in the image training samples are calculated, and the damage values of the track fasteners in the image training samples are respectively associated with the fastener missing state, the fastener damaged state and the normal fastener state;
[0007] A real-time image of the current track fastener is acquired. Based on the track fastener detection model, the track fastener target is detected and the damage value of the track fastener in the real-time image is calculated. The damage value of the track fastener in the image training sample is compared with the damage value of the track fastener to determine the current track fastener status.
[0008] Furthermore, the track fastener detection model includes:
[0009]
[0010] Where D is the damage value of the track fastener, n is the number of feature vectors in the image, and w i f represents the weight of the i-th feature vector of the image. i Let v be the i-th feature vector of the image, m be the number of gradients of the feature vectors in the image, and v be the number of gradients of the feature vectors in the image. j g represents the weight of the gradient change of the j-th feature vector in the image. j Let be the gradient function of the j-th feature vector in the image. Let I be the first derivative of image I in the x-direction. Let λ be the first derivative of image I in the y-direction, λ be the regularization parameter, and H be the texture evaluation function of the image.
[0011] Furthermore, the change function g of the gradient of the j-th feature vector in the image j include:
[0012]
[0013] Where K is the number of dimensions that map the gradient to the feature space, and w jk h represents the weights of the gradient of the j-th eigenvector mapped to the k-th dimension. k Let be the mapping function for the k-th dimension.
[0014] Furthermore, the texture evaluation function H of the image includes:
[0015]
[0016] Where N is the number of pixels in the image, and i′ is pixel i′.
[0017] Furthermore, the mapping function h of the k-th dimension k include:
[0018]
[0019] Among them, a k b is the first mapping factor in the k-th dimension. k It is the second mapping factor of the k-th dimension.
[0020] This invention also proposes a deep learning-based track fastener detection system, comprising:
[0021] The sample processing module is used to acquire image training samples of track fastener images, preprocess the image training samples to generate preprocessed image training samples, and label the image training samples according to the track fastener status, which includes: fastener missing status, fastener damaged status, and normal fastener status.
[0022] The model module is set up to extract feature vectors from the images in the image training samples. The track fastener detection model is set up, and the feature vectors are input into the track fastener detection model to detect track fastener targets and calculate the damage values of the track fasteners in the image training samples. The damage values of the track fasteners in the image training samples are then associated with the fastener missing state, the fastener damaged state, and the normal fastener state, respectively.
[0023] The detection module is used to acquire real-time images of the current track fasteners, detect track fastener targets according to the track fastener detection model, calculate the damage value of the track fasteners in the real-time images, and compare it with the damage value of the track fasteners in the image training samples to determine the current track fastener status.
[0024] Furthermore, the track fastener detection model includes:
[0025]
[0026] Where D is the damage value of the track fastener, n is the number of feature vectors in the image, and w i Let f_i be the weight of the i-th feature vector of the image, fi be the i-th feature vector of the image, m be the number of gradients of the feature vectors in the image, and v_i be the weight of the i-th feature vector of the image. j g represents the weight of the gradient change of the j-th feature vector in the image. j Let be the gradient function of the j-th feature vector in the image. Let I be the first derivative of image I in the x-direction. Let λ be the first derivative of image I in the y-direction, λ be the regularization parameter, and H be the texture evaluation function of the image.
[0027] Furthermore, the change function gj of the gradient of the j-th feature vector in the image includes:
[0028]
[0029] Where K is the number of dimensions that map the gradient to the feature space, and w jk h represents the weights of the gradient of the j-th eigenvector mapped to the k-th dimension. k Let be the mapping function for the k-th dimension.
[0030] Furthermore, the texture evaluation function H of the image includes:
[0031]
[0032] Where N is the number of pixels in the image, and i′ is pixel i′.
[0033] Furthermore, the mapping function h of the k-th dimension k include:
[0034]
[0035] Among them, a k b is the first mapping factor in the k-th dimension. k It is the second mapping factor of the k-th dimension.
[0036] Compared with the prior art, the above-described technical solutions conceived in this invention have the following beneficial effects:
[0037] This invention acquires image training samples of track fastener images, preprocesses these samples to generate preprocessed image training samples, and labels them according to track fastener states, including: missing fastener state, damaged fastener state, and normal fastener state. Feature vectors are extracted from the images in the training samples, a track fastener detection model is set up, and the feature vectors are input into the model to calculate the damage value of the track fasteners in the training samples. These damage values are then correlated with the missing, damaged, and normal fastener states, respectively. A real-time image of the current track fastener is acquired, and the damage value of the track fastener in the real-time image is calculated using the detection model. This value is compared with the damage value of the track fastener in the training samples to determine the current track fastener state. This invention, through the above technical solution, is safer, more efficient, and more accurate than manual inspection. Attached Figure Description
[0038] Figure 1 This is a flowchart of Embodiment 1 of the present invention;
[0039] Figure 2 This is a structural diagram of the system of Embodiment 2 of the present invention;
[0040] Figure 3 This is a schematic diagram of the R3det process of the present invention;
[0041] Figure 4 This is a schematic diagram of the feature refining module of the present invention;
[0042] Figure 5 This is an example diagram of abnormal status types of fasteners according to the present invention. Detailed Implementation
[0043] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.
[0044] The method provided by this invention can be implemented in a terminal environment that may include one or more of the following components: a processor, a storage medium, and a display screen. The storage medium stores at least one instruction, which is loaded and executed by the processor to implement the method described in the following embodiments.
[0045] A processor may include one or more processing cores. The processor uses various interfaces and lines to connect various parts of the terminal, and performs various functions and processes data by running or executing instructions, programs, code sets or instruction sets stored in the storage medium, and by calling data stored in the storage medium.
[0046] Storage media can include random access memory (RAM) or read-only memory (ROM). Storage media can be used to store instructions, programs, code, code sets, or instructions.
[0047] The display screen is used to show the user interface of each application.
[0048] In the formula of this invention, all subscripts are only used to distinguish parameters and have no actual meaning.
[0049] In addition, those skilled in the art will understand that the structure of the terminal described above does not constitute a limitation on the terminal. The terminal may include more or fewer components, or combine certain components, or have different component arrangements. For example, the terminal may also include radio frequency circuits, input units, sensors, audio circuits, power supplies, and other components, which will not be described in detail here.
[0050] Example 1
[0051] like Figure 1 As shown, this embodiment of the invention provides a deep learning-based method for detecting track fasteners, including:
[0052] Step 101: Obtain image training samples of track fastener images (this step utilizes a line scan camera of the track inspection system installed at the bottom of the operating train to acquire training samples, where each image contains a unique ID and shooting time, etc.). Preprocess the image training samples to increase the number of defect samples, generating preprocessed image training samples. Label the image training samples according to the type of fastener abnormality, and label them according to the track fastener status, such as... Figure 5 As shown, the track fastener states include: fastener missing state, fastener damaged state, and normal fastener state. The labeled track fastener dataset is fed into a single-level detector refined based on rotating target features. This algorithm consists of a backbone network and two task subnetworks. Figure 3As shown, the backbone network employs an FPN (Feature Pyramid Network) to generate rich multi-scale convolutional feature pyramids based on a feedforward ResNet (residual network) architecture. On this backbone, the algorithm adds two task subnetworks: the first performs convolutional object classification on the backbone network's output; the second performs convolutional bounding box regression. The network design is very simple, which allows for faster operation while maintaining detection accuracy. Meanwhile, as... Figure 4 As shown, the Feature Refinement (FRM) module uses feature interpolation to obtain the position information of the refined bounding boxes and reconstruct the feature map, achieving feature alignment. FRM can also reduce the number of refined bounding boxes after the first stage, thereby accelerating the model's recognition speed.
[0053] Step 102: Extract the feature vector of the image in the image training sample, set up the track fastener detection model, input the feature vector into the track fastener detection model, detect the track fastener target and calculate the damage value of the track fastener in the image training sample, and correspond the damage value of the track fastener in the image training sample to the fastener missing state, the fastener damaged state and the normal fastener state respectively;
[0054] Specifically, the track fastener detection model includes:
[0055]
[0056] Where D is the damage value of the track fastener, n is the number of feature vectors in the image, and w i f represents the weight of the o-th feature vector of the image. i Let v be the i-th feature vector of the image, m be the number of gradients of the feature vectors in the image, and v be the number of gradients of the feature vectors in the image. j g represents the weight of the gradient change of the j-th feature vector in the image. j Let be the gradient function of the j-th feature vector in the image. Let I be the first derivative of image I in the x-direction. Let λ be the first derivative of image I in the y-direction, λ be the regularization parameter, and H be the texture evaluation function of the image.
[0057] Specifically, the gradient change function g of the j-th feature vector in the image j include:
[0058]
[0059] Where K is the number of dimensions that map the gradient to the feature space, and w jk h represents the weights of the gradient of the j-th eigenvector mapped to the k-th dimension. k Let be the mapping function for the k-th dimension.
[0060] Specifically, the texture evaluation function H of the image includes:
[0061]
[0062] Where N is the number of pixels in the image, and i′ is pixel i′.
[0063] Specifically, the mapping function hk of the k-th dimension includes:
[0064]
[0065] Among them, a k b is the first mapping factor in the k-th dimension. k It is the second mapping factor of the k-th dimension.
[0066] Step 103: Obtain a real-time image of the current track fastener. Based on the track fastener detection model, detect the track fastener target and calculate the damage value of the track fastener in the real-time image. Compare the damage value of the track fastener in the image training sample to determine the current track fastener status. Locate the problem fastener based on the image ID and the recorded shooting time, and output the location of the problem fastener to facilitate subsequent inspection and repair by staff.
[0067] Example 2
[0068] like Figure 2 As shown, this embodiment of the invention also provides a deep learning-based track fastener detection system, comprising:
[0069] The sample processing module is used to acquire image training samples of track fastener images (this step utilizes a line scan camera of the track inspection system installed at the bottom of the operating train to acquire training samples, where each image contains a unique ID and shooting time, etc.), preprocesses the image training samples, increases the number of defect samples, generates preprocessed image training samples, and labels the image training samples according to the type of fastener abnormality, labeling them according to the track fastener status, such as... Figure 5 As shown, the track fastener status includes: fastener missing status, fastener damaged status, and normal fastener status. The labeled high-speed rail track fastener dataset is fed into a single-level detector refined based on rotating target features. This algorithm consists of a backbone network and two task subnets. (See figure) Figure 3As shown, the backbone network employs an FPN (Feature Pyramid Network) to generate rich multi-scale convolutional feature pyramids based on a feedforward ResNet (residual network) architecture. On this backbone, the algorithm adds two task subnetworks: the first performs convolutional object classification on the backbone network's output; the second performs convolutional bounding box regression. The network design is very simple, which allows for faster operation while maintaining detection accuracy. Meanwhile, as... Figure 4 As shown, the Feature Refinement (FRM) module uses feature interpolation to obtain the positional information of the refined bounding boxes and reconstruct the feature map, achieving feature alignment. FRM can also reduce the number of refined bounding boxes after the first stage, thereby speeding up the model.
[0070] The model module is set up to extract feature vectors from the images in the image training samples. The track fastener detection model is set up, the feature vectors are input into the track fastener detection model, the damage value of the track fasteners in the image training samples is calculated, and the damage value of the track fasteners in the image training samples is respectively associated with the fastener missing state, the fastener damaged state, and the normal fastener state.
[0071] Specifically, the track fastener detection model includes:
[0072]
[0073] Where D is the damage value of the track fastener, n is the number of feature vectors in the image, and w i f represents the weight of the i-th feature vector of the image. i Let v be the i-th feature vector of the image, m be the number of gradients of the feature vectors in the image, and v be the number of gradients of the feature vectors in the image. j g represents the weight of the gradient change of the j-th feature vector in the image. j Let be the gradient function of the j-th feature vector in the image. Let I be the first derivative of image I in the x-direction. Let λ be the first derivative of image I in the y-direction, λ be the regularization parameter, and H be the texture evaluation function of the image.
[0074] Specifically, the gradient change function g of the j-th feature vector in the image j include:
[0075]
[0076] Where K is the number of dimensions that map the gradient to the feature space, and w jk h represents the weights of the gradient of the j-th eigenvector mapped to the k-th dimension. k Let be the mapping function for the k-th dimension.
[0077] Specifically, the texture evaluation function H of the image includes:
[0078]
[0079] Where N is the number of pixels in the image, and i′ is pixel i′.
[0080] Specifically, the mapping function h of the k-th dimension k include:
[0081]
[0082] Among them, a k b is the first mapping factor in the k-th dimension. k It is the second mapping factor of the k-th dimension.
[0083] The detection module is used to acquire real-time images of the current track fasteners, and to detect the fastener targets in real time according to the track fastener detection model. It also calculates the damage value of the track fasteners in the real-time images and compares it with the damage value of the track fasteners in the image training samples to determine the current status of the track fasteners. Based on the image ID and the recorded shooting time, it locates the problematic fasteners and outputs their location to facilitate subsequent inspection and repair by staff.
[0084] Example 3
[0085] This invention also proposes a storage medium storing multiple instructions for implementing the deep learning-based track fastener detection method.
[0086] Optionally, in this embodiment, the storage medium may be located in any computer terminal in a group of computer terminals in a computer network, or in any mobile terminal in a group of mobile terminals.
[0087] Optionally, in this embodiment, the storage medium is configured to store program code for performing the following steps: Step 101, acquiring image training samples of track fastener images (wherein, this step utilizes a line scan camera of the track inspection system installed at the bottom of the operating train to acquire training samples, wherein each image contains a unique ID and shooting time, etc.), preprocessing the image training samples, increasing the number of defect samples, generating preprocessed image training samples, and labeling the image training samples according to the type of fastener abnormality state, labeling according to the track fastener state, such as... Figure 5 As shown, the track fastener states include: fastener missing state, fastener damaged state, and normal fastener state. The labeled track fastener dataset is fed into a single-level detector refined based on rotating target features. This algorithm consists of a backbone network and two task subnetworks. Figure 3As shown, the backbone network employs an FPN (Feature Pyramid Network) to generate rich multi-scale convolutional feature pyramids based on a feedforward ResNet (residual network) architecture. On this backbone, the algorithm adds two task subnetworks: the first performs convolutional object classification on the backbone network's output; the second performs convolutional bounding box regression. The network design is very simple, which allows for faster operation while maintaining detection accuracy. Meanwhile, as... Figure 4 As shown, the Feature Refinement (FRM) module uses feature interpolation to obtain the positional information of the refined bounding boxes and reconstruct the feature map, achieving feature alignment. FRM can also reduce the number of refined bounding boxes after the first stage, thereby speeding up the model.
[0088] Step 102: Extract the feature vector of the image in the image training sample, set up the track fastener detection model, input the feature vector into the track fastener detection model, calculate the damage value of the track fastener in the image training sample, and correspond the damage value of the track fastener in the image training sample to the fastener missing state, the fastener damaged state, and the normal fastener state, respectively.
[0089] Specifically, the track fastener detection model includes:
[0090]
[0091] Where D is the damage value of the track fastener, n is the number of feature vectors in the image, and w i f represents the weight of the i-th feature vector of the image. i Let v be the i-th feature vector of the image, m be the number of gradients of the feature vectors in the image, and v be the number of gradients of the feature vectors in the image. j g represents the weight of the gradient change of the j-th feature vector in the image. j Let be the gradient function of the j-th feature vector in the image. Let I be the first derivative of image I in the x-direction. Let λ be the first derivative of image I in the y-direction, λ be the regularization parameter, and H be the texture evaluation function of the image.
[0092] Specifically, the gradient change function g of the j-th feature vector in the image j include:
[0093]
[0094] Where K is the number of dimensions that map the gradient to the feature space, and w jk h represents the weights of the gradient of the j-th eigenvector mapped to the k-th dimension. k Let be the mapping function for the k-th dimension.
[0095] Specifically, the texture evaluation function H of the image includes:
[0096]
[0097] Where N is the number of pixels in the image, and i′ is pixel i′.
[0098] Specifically, the mapping function h of the k-th dimension k include:
[0099]
[0100] Among them, a k b is the first mapping factor in the k-th dimension. k It is the second mapping factor of the k-th dimension.
[0101] Step 103: Obtain a real-time image of the current track fastener, and detect the track fastener target according to the track fastener detection model and calculate the damage value of the track fastener in the real-time image. Compare the damage value of the track fastener in the image training sample to determine the current track fastener status. Locate the problem fastener according to the image ID and the recorded shooting time, and output the location of the problem fastener to facilitate subsequent inspection and repair by the staff.
[0102] Example 4
[0103] This invention also proposes an electronic device, including a processor and a storage medium connected to the processor. The storage medium stores multiple instructions, which can be loaded and executed by the processor to enable the processor to execute a deep learning-based track fastener detection method.
[0104] Specifically, the electronic device in this embodiment can be a computer terminal, which may include one or more processors and a storage medium.
[0105] The storage medium can be used to store software programs and modules, such as the deep learning-based track fastener detection method in this embodiment of the invention. The corresponding program instructions / modules allow the processor to execute various functional applications and data processing by running the software programs and modules stored in the storage medium, thus realizing the aforementioned deep learning-based track fastener detection method. The storage medium may include high-speed random access storage media, and may also include non-volatile storage media, such as one or more magnetic storage systems, flash memory, or other non-volatile solid-state storage media. In some instances, the storage medium may further include storage media remotely configured relative to the processor, which can be connected to the terminal via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.
[0106] The processor can call the information and application stored in the storage medium through the transmission system to execute the following steps: Step 101, acquire image training samples of track fastener images (wherein, this step utilizes the line scan camera of the track inspection system installed at the bottom of the operating train to acquire training samples, where each image contains a unique ID and shooting time, etc.), preprocess the image training samples, increase the number of defect samples, generate preprocessed image training samples, and label the image training samples according to the type of fastener abnormality state, labeling them according to the track fastener state, such as... Figure 5 As shown, the track fastener status includes: fastener missing status, fastener damaged status, and normal fastener status. The labeled high-speed rail track fastener dataset is fed into a single-level detector refined based on rotating target features. This algorithm consists of a backbone network and two task subnets. (See figure) Figure 3 As shown, the backbone network employs an FPN (Feature Pyramid Network) to generate rich multi-scale convolutional feature pyramids based on a feedforward ResNet (residual network) architecture. On this backbone, the algorithm adds two task subnetworks: the first performs convolutional object classification on the backbone network's output; the second performs convolutional bounding box regression. The network design is very simple, which allows for faster operation while maintaining detection accuracy. Meanwhile, as... Figure 4 As shown, the Feature Refinement (FRM) module uses feature interpolation to obtain the positional information of the refined bounding boxes and reconstruct the feature map, achieving feature alignment. FRM can also reduce the number of refined bounding boxes after the first stage, thereby speeding up the model.
[0107] Step 102: Extract the feature vector of the image in the image training sample, set up the track fastener detection model, input the feature vector into the track fastener detection model, detect the track fastener target and calculate the damage value of the track fastener in the image training sample, and correspond the damage value of the track fastener in the image training sample to the fastener missing state, the fastener damaged state and the normal fastener state respectively;
[0108] Specifically, the track fastener detection model includes:
[0109]
[0110] Where D is the damage value of the track fastener, n is the number of feature vectors in the image, wi is the weight of the i-th feature vector in the image, and f i Let v be the i-th feature vector of the image, m be the number of gradients of the feature vectors in the image, and v be the number of gradients of the feature vectors in the image. j g represents the weight of the gradient change of the j-th feature vector in the image. j Let be the gradient function of the j-th feature vector in the image. Let I be the first derivative of image I in the x-direction. Let λ be the first derivative of image I in the y-direction, λ be the regularization parameter, and H be the texture evaluation function of the image.
[0111] Specifically, the gradient change function g of the j-th feature vector in the image j include:
[0112]
[0113] Where K is the number of dimensions that map the gradient to the feature space, and w jk h represents the weights of the gradient of the j-th eigenvector mapped to the k-th dimension. k Let be the mapping function for the k-th dimension.
[0114] Specifically, the texture evaluation function H of the image includes:
[0115]
[0116] Where N is the number of pixels in the image, and i′ is pixel i′.
[0117] Specifically, the mapping function h of the k-th dimension k include:
[0118]
[0119] Among them, a k b is the first mapping factor in the k-th dimension. k It is the second mapping factor of the k-th dimension.
[0120] Step 103: Obtain a real-time image of the current track fastener, and detect the track fastener target according to the track fastener detection model and calculate the damage value of the track fastener in the real-time image. Compare the damage value of the track fastener in the image training sample to determine the current track fastener status. Locate the problem fastener according to the image ID and the recorded shooting time, and output the location of the problem fastener to facilitate subsequent inspection and repair by the staff.
[0121] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.
[0122] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.
[0123] In the several embodiments provided by this invention, it should be understood that the disclosed technical content can be implemented in other ways. The system embodiments described above are merely illustrative; for example, 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 units or modules, and may be electrical or other forms.
[0124] 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.
[0125] Furthermore, 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. The integrated unit can be implemented in hardware or as a software functional unit.
[0126] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part 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 the present invention. The aforementioned storage medium includes: USB flash drives, read-only storage media (ROM), random access storage media (RAM), portable hard drives, magnetic disks, optical disks, and other media capable of storing program code.
[0127] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.
Claims
1. A method for detecting track fasteners based on deep learning, characterized in that, include: Acquire image training samples of track fastener images, preprocess the image training samples to generate preprocessed image training samples, and label the image training samples according to the track fastener status, which includes: fastener missing status, fastener damaged status, and normal fastener status. The feature vectors of the images in the image training samples are extracted, a track fastener detection model is set, the feature vectors are input into the track fastener detection model, track fastener targets are detected and the damage values of the track fasteners in the image training samples are calculated, and the damage values of the track fasteners in the image training samples are respectively associated with the fastener missing state, the fastener damaged state and the normal fastener state; The track fastener detection model includes: in, This represents the damage value of the track fastener. The number of feature vectors in the image. For the image's first The weights of each feature vector. For the image's first 1 eigenvector This represents the number of gradients in the feature vectors of the image. For the first in the image The weights of the gradient changes of each eigenvector For the first in the image The function of the gradient of each eigenvector For image exist First derivative in the direction, For image exist First derivative in the direction, For regularization parameters, This is a texture evaluation function for the image; The image in the first The function of the gradient of each eigenvector include: in, The number of dimensions to which gradients are mapped to the feature space. For the first The gradient of the i-th eigenvector is mapped to the i-th eigenvector gradient. Weights of each dimension For the first A mapping function for each dimension; The texture evaluation function of the image include: in, The number of pixels in the image. For pixels ; The first Mapping function of each dimension include: in, For the first The first mapping factor of each dimension For the first The second mapping factor in each dimension; A real-time image of the current track fastener is acquired, and the track fastener target is detected according to the track fastener detection model. The damage value of the track fastener in the real-time image is calculated and compared with the damage value of the track fastener in the image training sample to determine the current track fastener status.
2. A deep learning-based track fastener detection system, characterized in that, include: The sample processing module is used to acquire image training samples of track fastener images, preprocess the image training samples to generate preprocessed image training samples, and label the image training samples according to the track fastener status, which includes: fastener missing status, fastener damaged status, and normal fastener status. The model module is set up to extract feature vectors from the images in the image training samples. The track fastener detection model is set up, and the feature vectors are input into the track fastener detection model to detect track fastener targets and calculate the damage values of the track fasteners in the image training samples. The damage values of the track fasteners in the image training samples are then associated with the fastener missing state, the fastener damaged state, and the normal fastener state, respectively. The track fastener detection model includes: in, This represents the damage value of the track fastener. The number of feature vectors in the image. For the image's first The weights of each feature vector. For the image's first 1 eigenvector This represents the number of gradients in the feature vectors of the image. For the first in the image The weights of the gradient changes of each eigenvector For the first in the image The function of the gradient of each eigenvector For image exist First derivative in the direction, For image exist First derivative in the direction, For regularization parameters, This is a texture evaluation function for the image; The image in the first The function of the gradient of each eigenvector include: in, The number of dimensions to which gradients are mapped to the feature space. For the first The gradient of the i-th eigenvector is mapped to the i-th eigenvector gradient. Weights of each dimension For the first A mapping function for each dimension; The texture evaluation function of the image include: in, The number of pixels in the image. For pixels ; The first Mapping function of each dimension include: in, For the first The first mapping factor of each dimension For the first The second mapping factor in each dimension; The detection module is used to acquire real-time images of the current track fasteners, detect track fastener targets according to the track fastener detection model, calculate the damage value of the track fasteners in the real-time images, and compare it with the damage value of the track fasteners in the image training samples to determine the current track fastener status.