Swin Transform-CNN (Convolutional Neural Network)-based static geometric parameter detection method for overhead line system

Abstract

The invention relates to a contact network static geometric parameter detection method based on a Swin Transform-CNN (Convolutional Neural Network), and aims to improve the precision and efficiency of contact network geometric parameter detection. The method comprises the following steps: firstly, acquiring static image data of the overhead line system through image acquisition equipment, and carrying out preprocessing, such as denoising, graying and normalization processing; then, the image data are input into a hybrid model based on Swin Transform and a convolutional neural network (CNN), the Swin Transform is used for extracting global geometric features of the overhead line system, and the CNN is responsible for extracting local detail features; by fusing the global and local features, the static geometric parameters of the overhead line system, including the height, inclination angle, tension, position offset and the like of the contact line, can be accurately detected. The method not only can adapt to complex environmental changes, but also can generate a visual detection report, provide real-time alarm and maintenance suggestions, and improve the maintenance efficiency and safety of the overhead line system.

Description

Technical Field

[0001] The invention belongs to the technical field of high-speed railway contact network, and in particular relates to a contact network static geometric parameter detection method based on Swin Transformer-CNN. Background Art

[0002] The overhead contact network is an important part of the railway electrification system, and its geometric parameters (such as height, tension, position offset, etc.) directly affect the safe operation of trains. Traditional methods for detecting the geometric parameters of the overhead contact network mainly rely on technologies such as manual inspection and laser measurement, but these methods have the problems of low efficiency, high cost and strong dependence on the environment. Especially in long-distance and high-density railway networks, manual inspection cannot monitor the geometric state of the entire overhead contact network system in real time and accurately, and laser measurement equipment is expensive and sensitive to weather conditions, which limits its application in harsh environments. With the development of computer vision technology, image processing technology based on deep learning provides a new idea for the automatic detection of the geometric parameters of the overhead contact network. Among them, convolutional neural network (CNN) performs well in image feature extraction, can automatically extract local features in images through convolution layers, and is widely used in target detection and image classification. However, CNN has limited ability to capture global features and is difficult to effectively process complex geometric structures such as overhead contact networks.

[0003] To solve this problem, Swin Transformer, as an emerging network architecture, models the global information of the image through the self-attention mechanism, which can effectively capture the global and local geometric features in complex scenes. Therefore, the hybrid model based on Swin Transformer and CNN can combine the advantages of both, while extracting the local geometric information of the contact network, retaining the global features, making the detection of the geometric parameters of the contact network more accurate and efficient.

[0004] The detection method using Swin Transformer-CNN can not only detect the geometric parameters of the contact network in real time and accurately in complex environments, but also provide real-time alarms and maintenance suggestions for abnormal situations through the generated visual reports, thereby improving the safety and reliability of the railway electrification system. Summary of the invention

[0005] The present invention proposes a contact network static geometric parameter detection method based on Swin Transformer-CNN, aiming to solve the problems of low efficiency, low precision and poor environmental adaptability existing in traditional contact network geometric parameter detection methods.

[0006] This method combines Swin Transformer with convolutional neural network (CNN) to make full use of the advantages of both, and realizes the accurate extraction of global and local geometric features of the contact network, thereby realizing efficient and accurate static geometric parameter detection. The method first collects the image data of the contact network through a camera device, and performs preprocessing, such as denoising, graying and normalization, to improve the image quality. Subsequently, the global features are extracted through the sliding window self-attention mechanism of the Swin Transformer module, and the local geometric detail features are extracted through the CNN module. After the combination of the two, the static geometric parameters of the contact network (such as contact line height, inclination angle and position offset, etc.) can be detected and calculated. This method can not only effectively cope with complex railway environments, but also has efficient and accurate geometric parameter detection capabilities, and can also generate visual detection reports, provide real-time monitoring and alarm functions, and greatly improve the maintenance and safety management efficiency of the contact network.

[0007] To achieve the above object, the technical solution adopted by the present invention is: a method for detecting static geometric parameters of a contact network based on Swin Transformer-CNN, comprising the steps of:

[0008] S100, collecting static image data of the contact network, wherein the image data includes geometric structure information of the contact network;

[0009] S200, preprocessing the collected image data, including image denoising, grayscale conversion and normalization, to improve the image processing effect;

[0010] S300, inputting the preprocessed image data into a hybrid model based on Swin Transformer and convolutional neural network (CNN), wherein the model includes a Swin Transformer module and a CNN module, wherein the Swin Transformer module is used to extract global geometric features, and the CNN module is used to extract local geometric features;

[0011] S400, Swin Transformer module extracts global features of the input image through the sliding window self-attention mechanism, using the formula:

[0012]

[0013] Among them, Q i , K i , V i are query, key and value matrices respectively, d k is the dimension of the attention head;

[0014] At S500, the CNN module extracts local features through convolution operations, which are defined as:

[0015] Z conv =σ(W*X+b)

[0016] Among them, W is the convolution kernel weight, X is the input image, b is the bias term, * represents the convolution operation, and σ is the activation function;

[0017] S600, combines global features and local features to obtain static geometric parameters of the contact network, including the height, inclination angle, tension, etc. of the contact network;

[0018] S700 outputs the test results and generates a visual report of the contact network geometric parameters.

[0019] Furthermore, the image preprocessing includes performing edge enhancement operation on the image and performing edge detection using the Sobel operator, which is defined as:

[0020]

[0021] Among them, I is the input image, G x and G y are the gradients in the horizontal and vertical directions respectively.

[0022] Furthermore, the Swin Transformer module uses a multi-head self-attention mechanism, where each head is calculated by the formula:

[0023] Z multi-head =Concat(head1,head2,...,head h )W O

[0024] Where h is the number of attention heads, W O is the output weight matrix.

[0025] Furthermore, the geometric parameter detection includes the detection of key points of the contact network and the calculation of geometric parameters, and the contact network geometric parameters include the height, position offset, inclination angle, etc. of the contact line;

[0026] Furthermore, the detection results are subjected to geometric parameter anomaly detection by generating a reconstruction error map, and the reconstruction error is defined as:

[0027]

[0028] Among them, x i and are the pixel values ​​of the input image and the reconstructed image respectively, and n is the total number of image pixels.

[0029] Furthermore, it includes real-time monitoring of the contact network based on the detection results, and triggering an alarm prompt or maintenance suggestion when the geometric parameters exceed the preset threshold.

[0030] The beneficial effects of adopting this technical solution are:

[0031] 1. Improve detection accuracy and comprehensiveness: By combining Swin Transformer with CNN, this method can simultaneously extract global and local geometric features of the contact network. Swin Transformer can capture global context information in the image, while CNN is good at processing local details, thus ensuring accurate detection of geometric parameters, which is especially suitable for scenes with complex geometric structures such as contact networks.

[0032] 2. Improve detection efficiency: Traditional manual detection and laser measurement methods are usually time-consuming and require complex equipment. This technical solution can significantly shorten the detection time through the application of image processing and deep learning models, realize real-time contact network geometric parameter detection, and reduce detection costs and manpower investment.

[0033] 3. Adaptability to complex environments: This technical solution has low dependence on the environment. Even in environments with large changes in lighting conditions or strong external noise, the pre-processed image data can still be effectively analyzed by the Swin Transformer-CNN model, improving the robustness and reliability of the system.

[0034] 4. Automated detection and alarm: When abnormal geometric parameters of the contact network are detected, this method can automatically generate a detection report and trigger an alarm prompt, promptly reminding operation and maintenance personnel to perform maintenance and avoid potential safety hazards, thereby improving the safety and maintenance efficiency of the railway system.

[0035] 5. Efficient visual analysis: Based on the detection results, this method can generate visual images and reports of the contact network geometric parameters, intuitively display the geometric status of the contact network, facilitate rapid location and solution of problems, and provide convenience for the daily operation and maintenance of the contact network.

[0036] In summary, this technical solution is highly practical and innovative, and can significantly improve the detection efficiency, accuracy and stability of contact network geometric parameters. It is suitable for daily maintenance and safety assurance of railway electrification systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 The present invention is a schematic flow chart of a method for detecting static geometric parameters of a contact network based on Swin Transformer-CNN. DETAILED DESCRIPTION

[0038] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention is further described below with reference to the accompanying drawings.

[0039] In this embodiment, see Figure 1 As shown, the present invention proposes a method for detecting static geometric parameters of a contact network based on Swin Transformer-CNN, comprising the steps of:

[0040] S100, collecting static image data of the contact network, wherein the image data includes geometric structure information of the contact network;

[0041] S200, preprocessing the collected image data, including image denoising, grayscale conversion and normalization, to improve the image processing effect;

[0042] S300, inputting the preprocessed image data into a hybrid model based on Swin Transformer and convolutional neural network (CNN), wherein the model includes a Swin Transformer module and a CNN module, wherein the Swin Transformer module is used to extract global geometric features, and the CNN module is used to extract local geometric features;

[0043] S400, Swin Transformer module extracts global features of the input image through the sliding window self-attention mechanism, using the formula:

[0044]

[0045] Among them, Q i , K i , V i are query, key and value matrices respectively, d k is the dimension of the attention head;

[0046] At S500, the CNN module extracts local features through convolution operations, which are defined as:

[0047] Z conv =σ(W*X+b)

[0048] Among them, W is the convolution kernel weight, X is the input image, b is the bias term, * represents the convolution operation, and σ is the activation function;

[0049] S600, combines global features and local features to obtain static geometric parameters of the contact network, including the height, inclination angle, tension, etc. of the contact network;

[0050] S700 outputs the test results and generates a visual report of the contact network geometric parameters.

[0051] Furthermore, the image preprocessing includes performing edge enhancement operation on the image and performing edge detection using the Sobel operator, which is defined as:

[0052]

[0053] Among them, I is the input image, G x and G y are the gradients in the horizontal and vertical directions respectively.

[0054] Furthermore, the Swin Transformer module uses a multi-head self-attention mechanism, where each head is calculated by the formula:

[0055] Z multi-head =Concat(head1,head2,...,head h )W O

[0056] Where h is the number of attention heads, W O is the output weight matrix.

[0057] Furthermore, the geometric parameter detection includes the detection of key points of the contact network and the calculation of geometric parameters, and the contact network geometric parameters include the height, position offset, inclination angle, etc. of the contact line;

[0058] Furthermore, the detection results are subjected to geometric parameter anomaly detection by generating a reconstruction error map, and the reconstruction error is defined as:

[0059]

[0060] Among them, x i and are the pixel values ​​of the input image and the reconstructed image respectively, and n is the total number of image pixels.

[0061] Furthermore, it includes real-time monitoring of the contact network based on the detection results, and triggering an alarm prompt or maintenance suggestion when the geometric parameters exceed the preset threshold.

[0062] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A method for detecting static geometric parameters of contact network based on Swin Transformer-CNN, characterized in that: The following steps are involved: S100, collecting static image data of the contact network, wherein the image data includes geometric structure information of the contact network; S200, preprocessing the collected image data, including image denoising, grayscale conversion and normalization, to improve the image processing effect; S300, inputting the preprocessed image data into a hybrid model based on Swin Transformer and convolutional neural network (CNN), wherein the model includes a Swin Transformer module and a CNN module, wherein the Swin Transformer module is used to extract global geometric features, and the CNN module is used to extract local geometric features; S400, Swin Transformer module extracts global features of the input image through the sliding window self-attention mechanism, using the formula: Among them, Q i , K i , V i are query, key and value matrices respectively, d k is the dimension of the attention head; At S500, the CNN module extracts local features through convolution operations, which are defined as: Z conv =σ(W*X+b) Among them, W is the convolution kernel weight, X is the input image, b is the bias term, * represents the convolution operation, and σ is the activation function; S600, combines global features and local features to obtain static geometric parameters of the contact network, including the height, inclination angle, and tension of the contact network; S700 outputs the test results and generates a visual report of the contact network geometric parameters.

2. The method for detecting static geometric parameters of contact network based on Swin Transformer-CNN according to claim 1 is characterized in that: The image preprocessing includes performing edge enhancement operation on the image and using the Sobel operator for edge detection, which is defined as: Among them, I is the input image, G x and G y are the gradients in the horizontal and vertical directions respectively.

3. The method for detecting static geometric parameters of contact network based on Swin Transformer-CNN according to claim 1 is characterized in that: The Swin Transformer module uses a multi-head self-attention mechanism, where each head is calculated by the formula: Z multi-head =Concat(head1,head2,...,head h )W O Where h is the number of attention heads, W O is the output weight matrix.

4. The method for detecting static geometric parameters of contact network based on Swin Transformer-CNN according to claim 1, characterized in that: The geometric parameter detection includes the detection of key points of the contact network and the calculation of geometric parameters. The contact network geometric parameters include the height, position offset and inclination angle of the contact line.

5. The method for detecting static geometric parameters of contact network based on Swin Transformer-CNN according to claim 1, characterized in that: The detection results are used to detect geometric parameter anomalies through the generated reconstruction error map, and the reconstruction error is defined as: Among them, x i and are the pixel values ​​of the input image and the reconstructed image respectively, and n is the total number of image pixels.

6. The method for detecting static geometric parameters of contact network based on Swin Transformer-CNN according to claim 1, characterized in that: It further includes real-time monitoring of the contact network based on the detection results, and triggering alarm prompts or maintenance suggestions when geometric parameters exceed preset thresholds.

Images