A method and device for detecting a component of an overhead contact system, a terminal and a storage medium

Abstract

The application provides a catenary component detection method and device, a terminal and a storage medium. The method comprises the following steps: obtaining a low-illumination image and a normal-illumination image obtained by shooting the same position of a high-speed railway catenary; inputting the low-illumination image and the normal-illumination image into a trained intermediate domain migration model, migrating the low-illumination image and the normal-illumination image to an intermediate domain respectively through the intermediate domain migration model, and obtaining a first natural image and a second natural image; inputting the first natural image and the second natural image into a trained target detection model, and obtaining a catenary component detection result corresponding to each natural image through processing. The application can effectively realize the distribution migration and alignment of different image domain data by migrating the low-illumination image and the normal-illumination image to the intermediate domain to obtain corresponding natural images, thereby improving the catenary component detection precision.

Description

Technical Field

[0001] This invention relates to the field of computer vision technology, and in particular to a method, apparatus, terminal and storage medium for detecting overhead contact line components. Background Technology

[0002] The overhead contact line component condition detection and monitoring system is a crucial technical guarantee for the safe operation of high-speed railways. Because the overhead contact line inspection vehicle needs to operate at night to avoid conflicts with daytime trains, the image acquisition system is inevitably affected by uneven illumination. This leads to a decrease in the quality of overhead contact line images under low-light conditions, thus affecting the accuracy of component identification.

[0003] Currently, the basic solution for detecting overhead contact line images in low-light conditions is to enhance image quality by increasing image brightness. However, this method cannot effectively address the data distribution differences between images with normal illumination and those with low illumination, resulting in low accuracy in identifying overhead contact line components.

[0004] Therefore, existing technologies have shortcomings and need to be improved and developed. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a method, device, terminal and storage medium for detecting overhead contact line components, in order to address the above-mentioned deficiencies of the prior art and solve the problem of low identification accuracy of overhead contact line components in the prior art.

[0006] The technical solution adopted by this invention to solve the technical problem is as follows:

[0007] In a first aspect, embodiments of the present invention provide a method for detecting overhead contact line components, the method comprising:

[0008] Acquire low-light and normal-light images of the same location on the overhead contact line of a high-speed railway;

[0009] The low-light image and the normal-light image are input into a trained intermediate domain transfer model. The intermediate domain transfer model transfers the low-light image and the normal-light image to the intermediate domain, respectively, to obtain the first natural image and the second natural image.

[0010] The first natural image and the second natural image are input into the trained target detection model, and the detection results of the catenary components corresponding to each natural image are obtained through processing.

[0011] In one embodiment, the intermediate domain transfer model includes an input module, a first image inversion module, an image decomposition module, a media transmission image refinement module, a second image inversion module, and an output module; the step of inputting the low-light image and the normal-light image into the trained intermediate domain transfer model, and transferring the low-light image and the normal-light image to the intermediate domain by the intermediate domain transfer model to obtain the first natural image and the second natural image, includes:

[0012] The low-light image and the normal-light image are input into the input module, and then the low-light image and the normal-light image are transmitted to the first image inversion module via the input module.

[0013] The first image inversion module converts the low-light image and the normal-light image into corresponding fog images to obtain a first fog image and a second fog image, and then transmits the first fog image and the second fog image to the image decomposition module;

[0014] The image decomposition module processes the first fog image and the second fog image to obtain a first medium transmission image corresponding to the first fog image and a second medium transmission image corresponding to the second fog image.

[0015] The first medium-transmitted image and the second medium-transmitted image are input into the medium-transmitted image refinement module. The medium-transmitted image refinement module performs data correction to obtain the first natural image to be reversed and the second natural image to be reversed.

[0016] The first natural image to be inverted and the second natural image to be inverted are transmitted to the second image inversion module. The second image inversion module performs inversion processing on the images and outputs the first natural image and the second natural image.

[0017] The first natural image and the second natural image are output through the output module.

[0018] In one embodiment, the step of converting the low-light image and the normal-light image into corresponding fog images via the first image inversion module to obtain a first fog image and a second fog image includes:

[0019] The first image inversion module normalizes the low-light image and the normal-light image respectively to obtain a first normalized image and a second normalized image.

[0020] The first normalized image and the second normalized image are inverted to obtain the first fog image and the second fog image.

[0021] In one embodiment, the image decomposition module includes a multi-representation module, a multi-scale module, a multi-scale transmission constraint module, and a nonlinear regression module; the step of processing the first fog image and the second fog image through the image decomposition module to obtain a first medium transmission image corresponding to the first fog image and a second medium transmission image corresponding to the second fog image includes:

[0022] The first fog image and the second fog image are received by the multi-representation module. The input image is processed by extracting features and reducing dimensions using several convolutional layers and max pooling layers in the module. The extracted features are then concatenated to obtain a first feature set and a second feature set.

[0023] The first feature set and the second feature set are transmitted to the multi-scale module. The multi-scale module extracts multi-scale features of the first feature set and the second feature set using a preset number of scale kernels to obtain the first multi-scale feature and the second multi-scale feature. The first multi-scale feature and the second multi-scale feature are then transmitted to the multi-scale transmission constraint module.

[0024] The multi-scale transmission constraint module performs multi-scale pooling, convolutional feature extraction, and upsampling operations on the first multi-scale feature and the second multi-scale feature respectively to obtain a first feature map and a second feature map, and then transmits the first feature map and the second feature map to the nonlinear regression module.

[0025] The nonlinear regression module performs a nonlinear transformation on the first feature map and the second feature map to obtain a first medium transmission image corresponding to the first fog image and a second medium transmission image corresponding to the second fog image.

[0026] In one embodiment, the media transmission image refinement module includes a mean filtering module and a local linearization module; the step of inputting the first media transmission image and the second media transmission image into the media transmission image refinement module, and performing data correction by the media transmission image refinement module to obtain the first natural image to be reversed and the second natural image to be reversed includes:

[0027] The first medium transmission image and the second medium transmission image are input into the mean filtering module, and dilated convolution is performed by the mean filtering module to obtain the third medium transmission image and the fourth medium transmission image.

[0028] The third medium-transmitted image and the fourth medium-transmitted image are transmitted to the local linear module. The local linear module processes the third medium-transmitted image and the fourth medium-transmitted image to obtain the first natural image to be reversed and the second natural image to be reversed.

[0029] In one implementation, the training steps of the intermediate domain transfer model include:

[0030] Acquire training set data, which includes multiple sets of images of the overhead contact system of a high-speed railway. Each set of images contains low-light training images and normal-light training images at the same location.

[0031] Initialize the intermediate domain to be trained and train the transfer model;

[0032] Obtain a preset first loss function, input all the low-light training images and the normal-light training images into the intermediate domain training transfer model in batches, and use the gradient descent method to adjust the parameters in the image decomposition module of the intermediate domain training transfer model to minimize the first loss function.

[0033] After the image decomposition module completes the first preset round of training, the first target medium transmission training image and the second target medium transmission training image corresponding to each image set are obtained.

[0034] Obtain a preset second loss function, input all the first target medium transmission training images and the second target medium transmission training images into the medium transmission image refinement module of the intermediate domain training transfer model in batches, and use gradient descent to adjust the parameters of the medium transmission image refinement module to minimize the second loss function.

[0035] After the media transmission image refinement module completes the second preset round of training, a trained intermediate domain transfer model is obtained.

[0036] In one implementation, the step of inputting the first natural image and the second natural image into a trained target detection model, and processing them to obtain the detection result of the contact wire component corresponding to each natural image, includes:

[0037] Input the first natural image and the second natural image into the trained target detection model;

[0038] The target detection model uses a multi-layer neural network structure to extract and analyze features from the first natural image and the second natural image, thereby obtaining the detection results of the first contact wire component corresponding to the first natural image and the detection results of the second contact wire component corresponding to the second natural image.

[0039] Secondly, embodiments of the present invention also provide a contact wire component testing device, comprising:

[0040] The image acquisition module is used to acquire low-light and normal-light images of the same location on the overhead contact line of a high-speed railway.

[0041] The natural image generation module is used to input the low-light image and the normal-light image into a trained intermediate domain transfer model, and then transfer the low-light image and the normal-light image to the intermediate domain through the intermediate domain transfer model to obtain the first natural image and the second natural image.

[0042] The result generation module is used to input the first natural image and the second natural image into the trained target detection model, and process them to obtain the detection result of the contact wire component corresponding to each natural image.

[0043] Thirdly, embodiments of the present invention also provide a terminal, the terminal comprising: a memory, a processor, and a contact wire component detection program stored in the memory and executable on the processor, wherein the contact wire component detection program, when executed by the processor, implements the steps of the contact wire component detection method as described above.

[0044] Fourthly, embodiments of the present invention also provide a computer-readable storage medium storing a contact wire component detection program, the contact wire component detection program being executable to implement the steps of the contact wire component detection method as described above.

[0045] The beneficial effects of this invention are as follows: This invention acquires low-light and normal-light images of the same location on a high-speed railway overhead contact line; inputs the low-light and normal-light images into a trained intermediate domain transfer model, which transfers the low-light and normal-light images to an intermediate domain, respectively, to obtain a first natural image and a second natural image; inputs the first and second natural images into a trained target detection model, and processes them to obtain the detection results of the overhead contact line components corresponding to each natural image. This invention, by transferring low-light and normal-light images to an intermediate domain to obtain corresponding natural images, can effectively achieve the distribution migration and alignment of data from different image domains, thereby improving the detection accuracy of overhead contact line components. Attached Figure Description

[0046] Figure 1 This is a flowchart of a preferred embodiment of the contact wire component testing method of the present invention.

[0047] Figure 2 This is a schematic diagram of the imaging system for observation in foggy weather in this invention.

[0048] Figure 3 This is a schematic diagram of image alignment in this invention. Figure 1 .

[0049] Figure 4 This is a schematic diagram of image alignment in this invention. Figure 2 .

[0050] Figure 5 This is a schematic diagram illustrating image alignment using an intermediate domain in this invention.

[0051] Figure 6 This is a schematic diagram of the intermediate domain migration model of the present invention.

[0052] Figure 7 This is a flowchart for generating natural images in this invention.

[0053] Figure 8 This is a schematic diagram of the normal illumination image of the overhead contact line in this invention.

[0054] Figure 9 This is a schematic diagram of a low-light image of the overhead contact line in this invention.

[0055] Figure 10 This is a schematic diagram of the normal illumination image of the contact network after intermediate domain migration in this invention.

[0056] Figure 11 This is a schematic diagram of the low-light image of the contact network after intermediate domain migration in this invention.

[0057] Figure 12 This is a schematic diagram of the test results of a contact wire component in this invention.

[0058] Figure 13 This is a schematic diagram of the process for inspecting overhead contact line components in this invention.

[0059] Figure 14 This is a schematic diagram of a preferred embodiment of the contact wire component inspection device of the present invention.

[0060] Figure 15 This is a block diagram of the terminal principle of the present invention. Detailed Implementation

[0061] To make the objectives, technical solutions, and advantages of this invention clearer and more explicit, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0062] The overhead contact line component condition detection and monitoring system is a crucial technical guarantee for the safe operation of high-speed railways. Because the overhead contact line inspection vehicle needs to operate at night to avoid conflicts with daytime trains, the image acquisition system is inevitably affected by uneven illumination. This leads to a decrease in the quality of overhead contact line images under low-light conditions, thus affecting the accuracy of component identification.

[0063] Currently, the basic solution for detecting overhead contact line images in low-light conditions is to enhance image quality by increasing image brightness. However, this method cannot effectively address the data distribution differences between images with normal illumination and those with low illumination, resulting in low accuracy in identifying overhead contact line components.

[0064] To address the aforementioned deficiencies in existing technologies, this invention provides a method, apparatus, terminal, and storage medium for detecting overhead contact line components, belonging to the field of computer vision technology. The method includes: acquiring low-light and normal-light images of the same location on a high-speed railway overhead contact line; inputting the low-light and normal-light images into a trained intermediate domain transfer model, which transfers the low-light and normal-light images to an intermediate domain, respectively, to obtain a first natural image and a second natural image; inputting the first and second natural images into a trained target detection model, and processing them to obtain the detection result of the overhead contact line component corresponding to each natural image. This invention, by transferring low-light and normal-light images to an intermediate domain to obtain corresponding natural images, effectively achieves the distribution migration and alignment of data from different image domains, thereby improving the detection accuracy of overhead contact line components.

[0065] Please see Figure 1 The contact wire component testing method described in this embodiment of the invention includes the following steps:

[0066] Step S100: Obtain low-light images and normal-light images of the same location of the high-speed railway catenary.

[0067] Specifically, to avoid affecting the normal operation of trains, this invention uses a dedicated integrated train inspection vehicle to image the same location on the high-speed railway overhead contact line at night, obtaining low-light images and normal-light images. The low-light image represents an image with insufficient supplementary lighting, while the normal-light image represents an image with sufficient supplementary lighting.

[0068] Please see Figure 1 The contact wire component testing method described in this embodiment of the invention further includes the following steps:

[0069] Step S200: Input the low-light image and the normal-light image into the trained intermediate domain transfer model. The intermediate domain transfer model transfers the low-light image and the normal-light image to the intermediate domain respectively to obtain the first natural image and the second natural image.

[0070] Specifically, in existing technologies, the basic solution for detecting overhead contact line images in low-light conditions is to enhance image quality by increasing image brightness. However, this simple brightness enhancement method cannot effectively address the data distribution differences between low-light and normal-light images. Furthermore, there are contrast-adjustment-based methods that enhance railway track images to improve damage detection accuracy, but these methods rely on human experience and fail to consider the distribution differences between low-light and normal-light images. The intermediate domain migration model of this invention can effectively migrate low-light and normal-light images to an intermediate domain, achieving data distribution migration and alignment across different image domains. Subsequently, the migrated first and second natural images are used for detecting overhead contact line components. With the data distribution aligned, the detection model can better focus on the characteristics of the overhead contact line components themselves for analysis and identification, avoiding misjudgments caused by data differences, thereby effectively improving detection accuracy.

[0071] In one implementation, the intermediate domain transfer model includes an input module, a first image inversion module, an image decomposition module, a media transmission image refinement module, a second image inversion module, and an output module; the step of inputting the low-light image and the normal-light image into the trained intermediate domain transfer model, and transferring the low-light image and the normal-light image to the intermediate domain by the intermediate domain transfer model to obtain the first natural image and the second natural image, includes:

[0072] The low-light image and the normal-light image are input into the input module, and then the low-light image and the normal-light image are transmitted to the first image inversion module via the input module.

[0073] The first image inversion module converts the low-light image and the normal-light image into corresponding fog images to obtain a first fog image and a second fog image, and then transmits the first fog image and the second fog image to the image decomposition module;

[0074] The image decomposition module processes the first fog image and the second fog image to obtain a first medium transmission image corresponding to the first fog image and a second medium transmission image corresponding to the second fog image.

[0075] The first medium-transmitted image and the second medium-transmitted image are input into the medium-transmitted image refinement module. The medium-transmitted image refinement module performs data correction to obtain the first natural image to be reversed and the second natural image to be reversed.

[0076] The first natural image to be inverted and the second natural image to be inverted are transmitted to the second image inversion module. The second image inversion module performs inversion processing on the images and outputs the first natural image and the second natural image.

[0077] The first natural image and the second natural image are output through the output module.

[0078] Specifically, the intermediate domain migration model of the present invention is constructed based on the atmospheric scattering model, dark channel theory and guided filter model.

[0079] Since photos taken at night, when inverted, are highly similar to fog images, this invention uses an atmospheric scattering model to defog fog images, thereby enhancing both low-light and normal-light images. This is beneficial for the inspection of overhead contact line components.

[0080] The imaging process of foggy weather images and the mathematical formulas corresponding to their key elements are shown in formula (1).

[0081] I(x)=R(x)t(x)+A(x)(1-t(x)) (1)

[0082] Where I(x) is the fog image, R(x) is the natural image, t(x) is the medium transfer function, and A(x) is the natural light coefficient.

[0083] like Figure 2 As shown, the degradation of imaging quality in foggy weather is mainly due to the following two aspects: First, in foggy weather, the reflected light of objects is absorbed and scattered by suspended particles in the atmosphere, which weakens the reflected light of objects and results in low illumination of the imaging results of the observation system. This process corresponds to the R(x)t(x) part of formula (1); Second, in foggy weather, the ambient illumination such as sunlight is scattered by the scattering medium in the atmosphere, forming background illumination, which leads to blurred imaging results of the observation system. This process corresponds to the A(x)(1-t(x)) part of formula (1).

[0084] Dark channel theory utilizes statistics to analyze a large number of outdoor images without haze, discovering that in each pixel's RGB color channels, one channel always has a very low brightness value, even close to zero, as shown in formula (2). The medium transfer function t(x) can be approximated by the dark channel: t(x) ∝ 1-R dark (x). Therefore, when decomposing an image, the dark channel theory can be used to estimate and solve the medium transfer function.

[0085]

[0086] Where c represents the RGB channels of the image, Ω r (x) represents the image region at point x with radius r, where R c (y) represents the image Ω r The image in region (x), R dark(x) in the image Ω r The image is composed of the lowest values ​​of the RGB three-way pixels in the (x) region.

[0087] The image decomposition module of this invention is constructed based on the atmospheric scattering model and dark channel theory, primarily used to decompose the media transmission image from foggy images. However, the initially decomposed media transmission image often lacks smoothness, resulting in an uneven natural image. Therefore, a media transmission image refinement module is introduced to finely adjust the media transmission image, making it smoother and more natural, thereby decomposing a smoother natural image. In this invention, the media transmission image refinement module is constructed based on a guided filter model.

[0088] Guided filtering assumes that a point on a function is linearly related to its neighborhood. Similarly, an image can be viewed as a two-dimensional function. Based on this, it is assumed that the output and input of the guided filter function have a linear relationship within a two-dimensional window, as shown below:

[0089]

[0090] q i =p i -n i (4)

[0091] in, and The parameters, I, are obtained by performing linear regression within the k-th local window of the image. i It is a guide image (using both the low-light image and the normal-light image itself), q i The output is the filtered reconstructed image, ω. k p is the k-th local window in the image. i The input image to be filtered is n. i It's noise, and 'i' is a subscript.

[0092] The next core task is to find the coefficients of the linear function, which is essentially a linear regression problem. Since we want to minimize the difference between the output value q of the fitted function and the true value p, we transform it into an optimization problem, that is, to minimize the following formula (5).

[0093]

[0094] Here, ε is the regularization coefficient, which prevents the model parameter a from becoming too large, thereby ensuring the stability and effectiveness of the model.

[0095] In existing technologies, low-light images can be processed in the following ways: First, empirical alignment of the source domain is not used, as illustrated in the diagram below. Figure 3As shown; secondly, reverse migration alignment is performed using the source domain, as illustrated in the diagram below. Figure 4 As shown in the figure, the source domain is the normal illumination image, and the target domain is the low illumination image. Neither approach effectively addresses the migration and alignment of data distribution between different image domains. This invention, however, utilizes an atmospheric scattering model, using an intermediate image domain as a bridge to migrate the low illumination and normal illumination images to the same region. This allows for rapid and efficient migration and alignment of data from different image domains, effectively solving the problem of low detection accuracy for contact wire components caused by inconsistent data distribution between low illumination and normal illumination images. A schematic diagram is shown below. Figure 5 As shown.

[0096] The intermediate domain migration model of this invention is constructed based on actual physical models (atmospheric scattering model and guided filter model), and the intermediate migration network model of this invention has higher interpretability and generalization. Furthermore, since the construction of the intermediate domain is based on a physical model with image enhancement capabilities, it can not only migrate the data distribution between low-light and normal-light images, but also improve the detection accuracy of contact wire components under low light conditions while simultaneously improving the detection accuracy of normal-light images, thereby achieving a comprehensive improvement in overall data detection accuracy. This fills a gap in contact wire detection, especially under low-light conditions, significantly improving detection accuracy.

[0097] In one implementation, the step of converting the low-light image and the normal-light image into corresponding fog images via the first image inversion module to obtain a first fog image and a second fog image includes:

[0098] The first image inversion module normalizes the low-light image and the normal-light image respectively to obtain a first normalized image and a second normalized image.

[0099] The first normalized image and the second normalized image are inverted to obtain the first fog image and the second fog image.

[0100] Specifically, considering that nighttime images, after being inverted, exhibit a high degree of similarity to foggy images, the low-light and normal-light images are first normalized to obtain a first normalized image and a second normalized image. Normalization standardizes image data, such as pixel values, to a specific range. After this step, the first and second normalized images are inverted. This inversion transforms them into images with characteristics similar to foggy images—the first and second fog images required by this invention. Subsequently, based on these fog images, this invention can extract natural images using specific processing methods, which can then be used in subsequent processes such as the inspection of overhead contact line components.

[0101] In one implementation, the image decomposition module includes a multi-representation module, a multi-scale module, a multi-scale transmission constraint module, and a nonlinear regression module; the step of processing the first fog image and the second fog image through the image decomposition module to obtain a first medium transmission image corresponding to the first fog image and a second medium transmission image corresponding to the second fog image includes:

[0102] The first fog image and the second fog image are received by the multi-representation module. The input image is processed by extracting features and reducing dimensions using several convolutional layers and max pooling layers in the module. The extracted features are then concatenated to obtain a first feature set and a second feature set.

[0103] The first feature set and the second feature set are transmitted to the multi-scale module. The multi-scale module extracts multi-scale features of the first feature set and the second feature set using a preset number of scale kernels to obtain the first multi-scale feature and the second multi-scale feature. The first multi-scale feature and the second multi-scale feature are then transmitted to the multi-scale transmission constraint module.

[0104] The multi-scale transmission constraint module performs multi-scale pooling, convolutional feature extraction, and upsampling operations on the first multi-scale feature and the second multi-scale feature respectively to obtain a first feature map and a second feature map, and then transmits the first feature map and the second feature map to the nonlinear regression module.

[0105] The nonlinear regression module performs a nonlinear transformation on the first feature map and the second feature map to obtain a first medium transmission image corresponding to the first fog image and a second medium transmission image corresponding to the second fog image.

[0106] Specifically, the structure of the intermediate domain migration network of the present invention is as follows: Figure 6 As shown. After receiving the first fog image and the second fog image, they are transmitted to the multi-representation module. The multi-representation module performs convolution operation on the fog image. In order to better extract features, the multi-representation module of this invention adopts a 4-channel convolutional feature extraction network to obtain the first feature set and the second feature set. In the multi-scale module, in order to solve the problem of image target scale invariance, three scale kernels are used to extract multi-scale features to obtain the first multi-scale feature and the second multi-scale feature. In the multi-scale transmission constraint module, in order to solve the problem of image target spatial invariance, this invention introduces multi-scale pooling, convolutional feature extraction and upsampling operations to ensure the spatial invariance and scale invariance of the target. In the nonlinear regression module, its output value is limited to between 0 and 1. This limitation helps to ensure that the gradient is more likely to converge, thereby improving the training efficiency and stability of the entire model. Based on this, this invention uses a double boundary constraint function to constrain the output, as shown in formula (6).

[0107] F4=min(t max ,max(t min ,W4*F3+B4)) (6)

[0108] Where W4 and B4 are the model parameters of the nonlinear regression module, F3 is the feature output from the multi-scale transport constraint module, and t min =0,t max =1. This setting allows the nonlinear regression module to operate accurately within the established framework, providing a high-quality data foundation for subsequent processing. After processing by the nonlinear regression module, it can output the first medium transmission image corresponding to the first fog image and the second medium transmission image corresponding to the second fog image.

[0109] In one implementation, the media transmission image refinement module includes a mean filtering module and a local linearization module; the step of inputting the first media transmission image and the second media transmission image into the media transmission image refinement module, and performing data correction by the media transmission image refinement module to obtain the first natural image to be reversed and the second natural image to be reversed includes:

[0110] The first medium transmission image and the second medium transmission image are input into the mean filtering module, and dilated convolution is performed by the mean filtering module to obtain the third medium transmission image and the fourth medium transmission image.

[0111] The third medium-transmitted image and the fourth medium-transmitted image are transmitted to the local linear module. The local linear module processes the third medium-transmitted image and the fourth medium-transmitted image to obtain the first natural image to be reversed and the second natural image to be reversed.

[0112] Specifically, after the media transmission image is decomposed, because it is still relatively coarse, the resulting natural image R... low (Natural image decomposed from low-light image) and R nor The natural image derived from the normal illumination image is not smooth enough and requires further optimization. Therefore, a media transfer function image refinement module based on guided filtering theory is used to filter the media transfer function image t(x) to obtain a more refined image. Thus, the decomposed natural image R low and R nor Smoother and more natural.

[0113] In the optimization process, firstly, the mean filtering module is used to perform grouped dilated convolution, which can effectively denoise the image while preserving the edge information of the image, making the image texture clearer and the features more prominent. Through this step, the third-medium transmission image and the fourth-medium transmission image can be obtained. Then, the local linear module is used to process them. Specifically, the local linear module includes three local linearizers. The first local linearizer is used to calculate various averages, correlations, variances, and covariances. Its calculation formula (7) is shown below.

[0114]

[0115] Among them, f mean Mean represents the operation of averaging the data. I The mean of the image transmitted via a third medium. p corr is the average value of the image transmitted via the fourth medium. I It is the correlation calculation of images transmitted through a third medium, corr Ip This involves the calculation of the relationship between images transmitted via third and fourth media, var I It is the variance calculation of images transmitted through a third medium, cov Ip It is the calculation of the covariance between the third-media image and the fourth-media image.

[0116] The second local linearizer is used to calculate the correlation coefficient in formula (3) to obtain the filtered image. The formula (8) used is shown below:

[0117]

[0118] Based on the relationship between the medium transfer function and the dark channel Typically, a value A is estimated as 0.1% of the brightest value in the dark channel. However, due to the unique characteristics of contact wire images—the images are taken at night, and nearly half of each image is sky—the value of A can be set to 1. Therefore, based on the obtained... With A, formula (1) can be transformed into formula (9):

[0119]

[0120] The third linearizer is based on formula (9) and The natural image R(x) to be inverted can be obtained.

[0121] The flowchart for natural image generation in this invention is as follows: Figure 7As shown, after image inversion, the medium-transmitted image is obtained through processing by the multi-representation module, multi-scale module, multi-scale transmission constraint module, and nonlinear regression module. Then, the image is optimized using the mean filtering module and three local linearizers, and after inversion, the final natural image (i.e., the intermediate domain-transferred image) is obtained. Understandably, since the first image inversion module inverts the original image into a fog image, the second image inversion module is set to restore the natural image to its natural state from the decomposed fog image.

[0122] In one implementation, the training steps of the intermediate domain transfer model include:

[0123] Acquire training set data, which includes multiple sets of images of the overhead contact system of a high-speed railway. Each set of images contains low-light training images and normal-light training images at the same location.

[0124] Initialize the intermediate domain to be trained and train the transfer model;

[0125] Obtain a preset first loss function, input all the low-light training images and the normal-light training images into the intermediate domain training transfer model in batches, and use the gradient descent method to adjust the parameters in the image decomposition module of the intermediate domain training transfer model to minimize the first loss function.

[0126] After the image decomposition module completes the first preset round of training, the first target medium transmission training image and the second target medium transmission training image corresponding to each image set are obtained.

[0127] Obtain a preset second loss function, input all the first target medium transmission training images and the second target medium transmission training images into the medium transmission image refinement module of the intermediate domain training transfer model in batches, and use gradient descent to adjust the parameters of the medium transmission image refinement module to minimize the second loss function.

[0128] After the media transmission image refinement module completes the second preset round of training, a trained intermediate domain transfer model is obtained.

[0129] Specifically, to ensure the precision of the final output image, this invention trains the intermediate domain transfer model. Before training, a training dataset is constructed. The training dataset contains multiple image sets of high-speed railway overhead contact lines, each containing low-light training images and normal-light training images of the same location. During training, the image decomposition module is trained first, followed by the media transmission image refinement module.

[0130] The first loss function corresponding to the image decomposition module is constructed based on the medium transfer function image t(x) estimated according to the dark channel hypothesis, and is expressed as:

[0131]

[0132] Among them, t i (x)=imerode(1-I i (x)), where imerode is the image erosion operation. Θ is the sum of the parameters of the image decomposition module. i is the index, I i t(x) is the i-th input sample image, N is the number of samples input for training each time, and L is the calculated loss. λ(Ii(x); Θ) represents the output of the image decomposition module. During training, the model parameters are optimized by minimizing the first loss function, i.e., minimizing the difference between the output of the image decomposition module and the dark channel theory estimate t(x).

[0133] The second loss function corresponding to the media transmission image refinement module is expressed as:

[0134] L G =argmin(E(a k ,b k )+MMD(F,X,Y)) (11)

[0135] in,

[0136] Specifically, argmin(E(a) k ,b k The corresponding formula is formula (5), where x∈X is the low-light image of the contact wire, y∈Y is the normal-light image of the contact wire, and f∈F is the Gaussian kernel function operation. n and m are the number of input samples for each corresponding low-light image and normal-light image. The natural image R obtained after the previous step is decomposed... low (Natural image decomposed from low-light image) and R nor The natural images decomposed from the normal illumination image cannot be guaranteed to have the same distribution. Therefore, formula (5) is further introduced into the MMD loss to obtain formula (11). By minimizing the distribution distance between the low illumination and normal illumination images of the contact network during training, the distribution of the low illumination and normal illumination images of the contact network can be further reduced in practical applications. This process is to migrate the low illumination and normal illumination images to the intermediate domain. By calculating the second objective loss function using the gradient descent algorithm, the parameters of the media transmission image refinement module can be optimized, thereby obtaining the correlation function a→A. map Substituting b→B into formula (3), we get the following formula (13):

[0137]

[0138] When used in subsequent formal applications, the image transmitted through the medium can be obtained using formula (13).

[0139] After both modules have been trained, a trained intermediate domain transfer network is obtained. This invention introduces MMD loss into the second loss function to make the illumination distribution of the low-light image closer to that of the normal-light image, thereby reducing illumination differences in the low-light image and making the distribution of the two images more consistent. This method can effectively improve the accuracy of subsequent contact network component inspection.

[0140] In this invention, several contact wire normal illumination images are as follows: Figure 8 As shown, several low-light images of overhead contact lines are as follows: Figure 9 As shown, after processing by the intermediate domain transfer model, the normal illumination image (i.e., the second natural image) after intermediate domain transfer is as follows: Figure 10 As shown, the low-light image after intermediate domain migration (i.e., the first natural image) is as follows: Figure 11 As shown, by using the intermediate domain transfer model, the distributions of the two images can be made more consistent.

[0141] Please see Figure 1 The contact wire component testing method described in this embodiment of the invention further includes the following steps:

[0142] Step S300: Input the first natural image and the second natural image into the trained target detection model, and process them to obtain the detection results of the contact wire components corresponding to each natural image.

[0143] Specifically, the target detection model of this invention is a Faster-RCNN network model. It is trained using gradient descent, enabling the deep learning network to more accurately locate and classify overhead contact line components. Please refer to [link / reference]. Figure 12 This is a schematic diagram illustrating the detection results of a contact wire component according to the present invention. Figure 12 In the diagram, 1 is the insulator base, 2 is the insulator, 3 is the support sleeve, 4 is the support sleeve tightening bolt, 5 is the rotating double lug, 6 is the positioner base, 7 is the guy wire positioning hook, 8 is the double sleeve connector, 9 is the catenary cable base, and 10 is the windproof guy wire ring. Note: Figure 12 This is merely a schematic diagram of a component inspection result and does not imply that the present invention can only support component inspection as described above.

[0144] In one implementation, the step of inputting the first natural image and the second natural image into a trained target detection model, and processing them to obtain the detection result of the contact wire component corresponding to each natural image, includes:

[0145] Input the first natural image and the second natural image into the trained target detection model;

[0146] The target detection model uses a multi-layer neural network structure to extract and analyze features from the first natural image and the second natural image, thereby obtaining the detection results of the first contact wire component corresponding to the first natural image and the detection results of the second contact wire component corresponding to the second natural image.

[0147] Specifically, by utilizing the multi-layer neural network structure in the Faster-RCNN network model to process natural images, the detection results of catenary components corresponding to each natural image can be effectively obtained. The training dataset for the Faster-RCNN network model was obtained by labeling component samples from the collected images using the image annotation tool LabelMe.

[0148] In summary, this invention acquires low-light and normal-light images of the same location on a high-speed railway overhead contact line; inputs these images into a trained intermediate domain transfer model, which transfers them to an intermediate domain to obtain a first natural image and a second natural image; and inputs these images into a trained target detection model, which then processes them to obtain the detection results of the overhead contact line components corresponding to each natural image. The process can be described as follows: Figure 13 As shown, this invention obtains the corresponding natural image by migrating low-light and normal-light images to an intermediate domain, which can effectively realize the distribution migration and alignment of data from different image domains, thereby improving the detection accuracy of catenary components.

[0149] In one embodiment, such as Figure 14 As shown, based on the above-described method for detecting overhead contact line components, the present invention also provides a corresponding device for detecting overhead contact line components, comprising:

[0150] The image acquisition module 100 is used to acquire low-light images and normal-light images obtained by shooting the same location of the high-speed railway catenary.

[0151] The natural image generation module 200 is used to input the low-light image and the normal-light image into a trained intermediate domain transfer model, and then transfer the low-light image and the normal-light image to the intermediate domain through the intermediate domain transfer model to obtain the first natural image and the second natural image.

[0152] The result generation module 300 is used to input the first natural image and the second natural image into the trained target detection model, and process them to obtain the detection result of the contact wire component corresponding to each natural image.

[0153] In one embodiment, the intermediate domain transfer model includes an input module, a first image inversion module, an image decomposition module, a media transfer image refinement module, a second image inversion module, and an output module; the natural image generation module includes:

[0154] The first image input unit is used to input the low-light image and the normal-light image into the input module, and then transmit the low-light image and the normal-light image to the first image inversion module via the input module;

[0155] The first image inversion unit is used to convert the low-light image and the normal-light image into corresponding fog images via the first image inversion module to obtain a first fog image and a second fog image, and then transmit the first fog image and the second fog image to the image decomposition module.

[0156] An image decomposition unit is used to process the first fog image and the second fog image through the image decomposition module to obtain a first medium transmission image corresponding to the first fog image and a second medium transmission image corresponding to the second fog image;

[0157] The image optimization unit is used to input the first medium-transmitted image and the second medium-transmitted image into the medium-transmitted image refinement module, and perform data correction by the medium-transmitted image refinement module to obtain the first natural image to be reversed and the second natural image to be reversed.

[0158] The second image inversion unit is used to transmit the first natural image to be inverted and the second natural image to be inverted to the second image inversion module, and the second image inversion module performs inversion processing on the images to output the first natural image and the second natural image.

[0159] The first image output unit is used to output the first natural image and the second natural image through the output module.

[0160] In one embodiment, the device further includes:

[0161] The normalization unit is used to normalize the low-light image and the normal-light image respectively by the first image inversion module to obtain a first normalized image and a second normalized image.

[0162] The inversion subunit is used to invert the first normalized image and the second normalized image respectively to obtain the first fog image and the second fog image.

[0163] In one embodiment, the image decomposition module includes a multi-representation module, a multi-scale module, a multi-scale transport constraint module, and a nonlinear regression module; the device further includes:

[0164] The feature extraction unit is used to receive the first fog image and the second fog image through the multi-representation module, use several convolutional layers and max pooling layers in the module to perform feature extraction and dimensionality reduction on the input image, and concatenate the extracted features to obtain the first feature set and the second feature set.

[0165] A multi-scale feature extraction unit is used to transmit the first feature set and the second feature set to the multi-scale module, and the multi-scale module extracts multi-scale features of the first feature set and the second feature set respectively using a preset number of scale kernels to obtain the first multi-scale feature and the second multi-scale feature, and transmits the first multi-scale feature and the second multi-scale feature to the multi-scale transmission constraint module.

[0166] The feature map generation unit is used to perform multi-scale pooling, convolutional feature extraction and upsampling operations on the first multi-scale feature and the second multi-scale feature in sequence through the multi-scale transmission constraint module to obtain the first feature map and the second feature map, and then transmit the first feature map and the second feature map to the nonlinear regression module.

[0167] The nonlinear transformation unit is used to perform nonlinear transformation on the first feature map and the second feature map through the nonlinear regression module to obtain a first medium transmission image corresponding to the first fog image and a second medium transmission image corresponding to the second fog image.

[0168] In one embodiment, the media transmission image refinement module includes a mean filtering module and a local linearity module, and the device further includes:

[0169] The dilated convolution unit is used to input the first medium transmission image and the second medium transmission image into the mean filtering module, and perform dilated convolution operation through the mean filtering module to obtain the third medium transmission image and the fourth medium transmission image.

[0170] The natural image generation unit is used to transmit the third medium-transmitted image and the fourth medium-transmitted image to the local linear module, and the local linear module processes the third medium-transmitted image and the fourth medium-transmitted image to obtain a first natural image to be reversed and a second natural image to be reversed.

[0171] In one embodiment, the device further includes:

[0172] The training set acquisition unit is used to acquire training set data, which includes multiple image sets of high-speed railway catenary, each image set containing low-light training images and normal-light training images at the same location.

[0173] The initialization unit is used to initialize the intermediate domain training transfer model to be trained.

[0174] The first training unit is used to obtain a preset first loss function, input all the low-light training images and the normal-light training images into the intermediate domain training transfer model in batches, and use the gradient descent method to adjust the parameters in the image decomposition module of the intermediate domain training transfer model to minimize the first loss function.

[0175] The training image generation unit is used to obtain a first target medium transmission training image and a second target medium transmission training image corresponding to each image set after the image decomposition module has completed a first preset round of training.

[0176] The second training unit is used to obtain a preset second loss function, input all the first target medium transmission training images and the second target medium transmission training images into the medium transmission image refinement module of the intermediate domain training transfer model in batches, and use the gradient descent method to adjust the parameters of the medium transmission image refinement module to minimize the second loss function.

[0177] The training completion unit is used to obtain a trained intermediate domain transfer model after the second preset round of training of the media transmission image refinement module is completed.

[0178] In one embodiment, the result generation module includes:

[0179] The second image input unit is used to input the first natural image and the second natural image into the trained target detection model;

[0180] The result output unit is used to extract and analyze features from the first natural image and the second natural image through the multi-layer neural network structure in the target detection model, so as to obtain the detection result of the first contact wire component corresponding to the first natural image and the detection result of the second contact wire component corresponding to the second natural image.

[0181] Based on the above embodiments, the present invention also provides a terminal, the principle block diagram of which can be as follows: Figure 15As shown. The aforementioned terminal includes a processor, memory, network interface, and display screen connected via a device bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores operating devices and contact wire component detection programs. The internal memory provides an environment for the operation of the operating devices and contact wire component detection programs stored in the non-volatile storage medium. The network interface is used for communication with external terminals via a network connection. When the contact wire component detection program is executed by the processor, it implements the steps of any of the aforementioned contact wire component detection methods. The display screen of the terminal can be a liquid crystal display (LCD) or an electronic ink display.

[0182] Those skilled in the art will understand that Figure 15 The block diagram shown is only a partial structural diagram related to the present invention and does not constitute a limitation on the terminal to which the present invention is applied. A specific terminal may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0183] In one embodiment, a terminal is provided, the terminal including a memory, a processor, and a catenary component detection program stored in the memory and executable on the processor. When the catenary component detection program is executed by the processor, it implements the steps of any catenary component detection method provided in the embodiments of the present invention.

[0184] This invention also provides a computer-readable storage medium storing a catenary component detection program. When the catenary component detection program is executed by a processor, it implements the steps of any catenary component detection method provided in this invention.

[0185] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0186] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the above device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments 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. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this invention. The specific working process of the units and modules in the above device can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0187] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0188] 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.

[0189] In the embodiments provided by this invention, it should be understood that the disclosed apparatus / terminal devices and methods can be implemented in other ways. For example, the apparatus / terminal device embodiments described above are merely illustrative. For instance, the division of modules or units described above is only a logical functional division, and in actual implementation, it can be divided in other ways. For example, multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed.

[0190] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not mean that the essence of the corresponding technical solutions deviates from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for inspecting overhead contact line components, characterized in that, The method includes: Acquire low-light and normal-light images of the same location on the overhead contact line of a high-speed railway; The low-light image and the normal-light image are input into a trained intermediate domain transfer model. The intermediate domain transfer model transfers the low-light image and the normal-light image to the intermediate domain, respectively, to obtain the first natural image and the second natural image. The first natural image and the second natural image are input into the trained target detection model, and the detection results of the contact wire components corresponding to each natural image are obtained through processing. The intermediate domain transfer model includes an input module, a first image inversion module, an image decomposition module, a media transmission image refinement module, a second image inversion module, and an output module. The process of inputting the low-light image and the normal-light image into the trained intermediate domain transfer model, and then transferring the low-light image and the normal-light image to the intermediate domain by the intermediate domain transfer model to obtain the first natural image and the second natural image, includes: The low-light image and the normal-light image are input into the input module, and then the low-light image and the normal-light image are transmitted to the first image inversion module via the input module. The first image inversion module converts the low-light image and the normal-light image into corresponding fog images to obtain a first fog image and a second fog image, and then transmits the first fog image and the second fog image to the image decomposition module; The image decomposition module processes the first fog image and the second fog image to obtain a first medium transmission image corresponding to the first fog image and a second medium transmission image corresponding to the second fog image. The first medium-transmitted image and the second medium-transmitted image are input into the medium-transmitted image refinement module. The medium-transmitted image refinement module performs data correction to obtain the first natural image to be reversed and the second natural image to be reversed. The first natural image to be inverted and the second natural image to be inverted are transmitted to the second image inversion module. The second image inversion module performs inversion processing on the images and outputs the first natural image and the second natural image. The first natural image and the second natural image are output through the output module.

2. The contact wire component testing method according to claim 1, characterized in that, The process of converting the low-light image and the normal-light image into corresponding fog images via the first image inversion module to obtain a first fog image and a second fog image includes: The first image inversion module normalizes the low-light image and the normal-light image respectively to obtain a first normalized image and a second normalized image. The first normalized image and the second normalized image are inverted to obtain the first fog image and the second fog image.

3. The contact wire component inspection method according to claim 1, characterized in that, The image decomposition module includes a multi-representation module, a multi-scale module, a multi-scale transmission constraint module, and a nonlinear regression module; the image decomposition module processes the first fog image and the second fog image to obtain a first medium transmission image corresponding to the first fog image and a second medium transmission image corresponding to the second fog image, including: The first fog image and the second fog image are received by the multi-representation module. The input image is processed by extracting features and reducing dimensions using several convolutional layers and max pooling layers in the module. The extracted features are then concatenated to obtain a first feature set and a second feature set. The first feature set and the second feature set are transmitted to the multi-scale module, and then... The multi-scale module uses a preset number of scale kernels to extract multi-scale features from the first feature set and the second feature set respectively, to obtain the first multi-scale feature and the second multi-scale feature, and then transmits the first multi-scale feature and the second multi-scale feature to the multi-scale transmission constraint module. The multi-scale transmission constraint module performs multi-scale pooling, convolutional feature extraction, and upsampling operations on the first multi-scale feature and the second multi-scale feature respectively to obtain a first feature map and a second feature map, and then transmits the first feature map and the second feature map to the nonlinear regression module. The nonlinear regression module performs a nonlinear transformation on the first feature map and the second feature map to obtain a first medium transmission image corresponding to the first fog image and a second medium transmission image corresponding to the second fog image.

4. The method for detecting overhead contact line components according to claim 1, characterized in that, The media transmission image refinement module includes a mean filtering module and a local linearization module; the process of inputting the first media transmission image and the second media transmission image into the media transmission image refinement module, and performing data correction by the media transmission image refinement module to obtain the first natural image to be reversed and the second natural image to be reversed includes: The first medium transmission image and the second medium transmission image are input into the mean filtering module, and dilated convolution is performed by the mean filtering module to obtain the third medium transmission image and the fourth medium transmission image. The third medium-transmitted image and the fourth medium-transmitted image are transmitted to the local linear module. The local linear module processes the third medium-transmitted image and the fourth medium-transmitted image to obtain the first natural image to be reversed and the second natural image to be reversed.

5. The method for detecting overhead contact line components according to claim 1, characterized in that, The training steps of the intermediate domain transfer model include: Acquire training set data, which includes multiple sets of images of the overhead contact system of a high-speed railway. Each set of images contains low-light training images and normal-light training images at the same location. Initialize the intermediate domain to be trained and train the transfer model; Obtain a preset first loss function, input all the low-light training images and the normal-light training images into the intermediate domain training transfer model in batches, and use the gradient descent method to adjust the parameters in the image decomposition module of the intermediate domain training transfer model to minimize the first loss function. After the image decomposition module completes the first preset round of training, the first target medium transmission training image and the second target medium transmission training image corresponding to each image set are obtained. Obtain a preset second loss function, input all the first target medium transmission training images and the second target medium transmission training images into the medium transmission image refinement module of the intermediate domain training transfer model in batches, and use gradient descent to adjust the parameters of the medium transmission image refinement module to minimize the second loss function. After the media transmission image refinement module completes the second preset round of training, a trained intermediate domain transfer model is obtained.

6. The method for detecting overhead contact line components according to claim 1, characterized in that, The step of inputting the first natural image and the second natural image into the trained target detection model, and processing them to obtain the detection result of the contact wire component corresponding to each natural image, includes: Input the first natural image and the second natural image into the trained target detection model; The target detection model uses a multi-layer neural network structure to extract and analyze features from the first natural image and the second natural image, thereby obtaining the detection results of the first contact wire component corresponding to the first natural image and the detection results of the second contact wire component corresponding to the second natural image.

7. A contact wire component testing device, characterized in that, include: The image acquisition module is used to acquire low-light and normal-light images of the same location on the overhead contact line of a high-speed railway. The natural image generation module is used to input the low-light image and the normal-light image into a trained intermediate domain transfer model, and then transfer the low-light image and the normal-light image to the intermediate domain through the intermediate domain transfer model to obtain the first natural image and the second natural image. The intermediate domain transfer model includes an input module, a first image inversion module, an image decomposition module, a media transmission image refinement module, a second image inversion module, and an output module. The process of inputting the low-light image and the normal-light image into the trained intermediate domain transfer model, and then transferring the low-light image and the normal-light image to the intermediate domain by the intermediate domain transfer model to obtain the first natural image and the second natural image, includes: The low-light image and the normal-light image are input into the input module, and then the low-light image and the normal-light image are transmitted to the first image inversion module via the input module. The first image inversion module converts the low-light image and the normal-light image into corresponding fog images to obtain a first fog image and a second fog image, and then transmits the first fog image and the second fog image to the image decomposition module; The image decomposition module processes the first fog image and the second fog image to obtain a first medium transmission image corresponding to the first fog image and a second medium transmission image corresponding to the second fog image. The first medium-transmitted image and the second medium-transmitted image are input into the medium-transmitted image refinement module. The medium-transmitted image refinement module performs data correction to obtain the first natural image to be reversed and the second natural image to be reversed. The first natural image to be inverted and the second natural image to be inverted are transmitted to the second image inversion module. The second image inversion module performs inversion processing on the images and outputs the first natural image and the second natural image. The first natural image and the second natural image are output via the output module; The result generation module is used to input the first natural image and the second natural image into the trained target detection model, and process them to obtain the detection result of the contact wire component corresponding to each natural image.

8. A terminal, characterized in that, The terminal includes a memory, a processor, and a catenary component detection program stored in the memory and executable on the processor. When the catenary component detection program is executed by the processor, it implements the steps of the catenary component detection method as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a catenary component detection program, which, when executed by a processor, implements the steps of the catenary component detection method as described in any one of claims 1-6.

Images