Online detection method, device and system for pantograph-catenary arcing, and storage medium

Abstract

The invention discloses a pantograph-catenary arc discharge online detection method, device and system and a storage medium, and the method comprises the steps: S1, obtaining a pantograph-catenary arc discharge detection model according to a pantograph-catenary data set; and S2, inputting a pantograph-catenary image acquired in real time into the pantograph-catenary arcing detection model, and outputting and judging whether arcing exists or not. By adopting the technical scheme of the invention, the method can adapt to different scenes and conditions, reduces the dependence on a fixed algorithm and parameters, is higher in robustness, and can better maintain the detection performance.

Description

Technical Field

[0001] The present invention belongs to the technical field of rail transit safety, and in particular relates to a method and device, a system and a storage medium for online detection of arcing of a bow-net. Background Art

[0002] In rail transit systems, bow-net arcing is an unavoidable hazardous phenomenon determined by the train's power transmission method. Timely and accurate detection of bow-net arcing is crucial to ensuring driving safety and improving operational efficiency. At present, bow-net arcing detection is mainly achieved through sensor acquisition and analysis and traditional digital image processing. However, the sensor acquisition and analysis method uses sensor information such as temperature, current, and vibration for analysis and judgment, with high response delay, poor environmental robustness, and difficulty in feature extraction; traditional digital image processing is easily affected by environmental interference, lacks adaptive capabilities, and lacks universality. Summary of the invention

[0003] The technical problem to be solved by the present invention is to provide a method and device, a system and a storage medium for online detection of bow-catenary arcing.

[0004] To achieve the above object, the present invention adopts the following technical solution:

[0005] A method for online detection of bow-net arcing, comprising:

[0006] Step S1, obtaining a bow-catcher arc detection model according to a bow-catcher data set;

[0007] Step S2: input the real-time acquired bow-net image into the bow-net arc detection model, output and judge whether there is arc.

[0008] Preferably, step S1 comprises:

[0009] Step S11, constructing a detection converter network structure;

[0010] Step S12: train the detection converter network structure according to the bow-net data set to obtain the bow-net arc detection model

[0011] Preferably, the detection transformer network structure includes: a backbone network, an encoder, a decoder and a prediction head; wherein,

[0012] The backbone network module includes: a ResNet18 network connected in sequence, a convolution layer with a convolution kernel size of 1×1, and a Flatten layer; the encoder module includes: 1 embedding layer and 3 encoding layers; the encoding layer includes: a first multi-head self-attention layer, a first residual connection and normalization layer, a first feedforward network layer, and a second residual connection and normalization layer connected in sequence; the position encoding of the encoding layer uses sinusoidal position encoding; the decoder module includes: 3 decoding layers; the decoding layer includes: a second multi-head self-attention layer, a third residual connection and normalization layer, a second multi-head attention layer, a fourth residual connection and normalization layer, a first feedforward network layer, and a fifth residual connection and normalization layer connected in sequence; the number of queries in the decoding layer is 2; the prediction head module includes: a second feedforward network layer and a third feedforward network layer connected in parallel;

[0013] The first feedforward network layer includes: a first fully connected layer, a LeakyReLU activation function layer, a discard layer, and a second fully connected layer connected in sequence; the second feedforward network layer includes: a third fully connected layer and a Softmax layer connected in sequence; the third feedforward network layer includes: a fourth fully connected layer, a fifth fully connected layer, and a Sigmoid activation function layer connected in sequence.

[0014] Preferably, the loss function of the detection transformer network structure includes: position loss and illumination loss.

[0015] The present invention also provides a bow-cat arc online detection device, comprising:

[0016] A training module is used to obtain a bow-net arc detection model based on a bow-net data set;

[0017] The detection module is used to input the real-time acquired bow-net image into the bow-net arc detection model, and output and judge whether there is arc.

[0018] Preferably, the training module includes:

[0019] A construction unit, used for constructing a detection transformer network structure;

[0020] The training unit is used to train the detection transformer network structure according to the bow-net data set to obtain the bow-net arc detection model

[0021] Preferably, the detection transformer network structure includes: a backbone network, an encoder, a decoder and a prediction head; wherein,

[0022] The backbone network module includes: a ResNet18 network connected in sequence, a convolution layer with a convolution kernel size of 1×1, and a Flatten layer; the encoder module includes: 1 embedding layer and 3 encoding layers; the encoding layer includes: a first multi-head self-attention layer, a first residual connection and normalization layer, a first feedforward network layer, and a second residual connection and normalization layer connected in sequence; the position encoding of the encoding layer uses sinusoidal position encoding; the decoder module includes: 3 decoding layers; the decoding layer includes: a second multi-head self-attention layer, a third residual connection and normalization layer, a second multi-head attention layer, a fourth residual connection and normalization layer, a first feedforward network layer, and a fifth residual connection and normalization layer connected in sequence; the number of queries in the decoding layer is 2; the prediction head module includes: a second feedforward network layer and a third feedforward network layer connected in parallel;

[0023] The first feedforward network layer includes: a first fully connected layer, a LeakyReLU activation function layer, a discard layer, and a second fully connected layer connected in sequence; the second feedforward network layer includes: a third fully connected layer and a Softmax layer connected in sequence; the third feedforward network layer includes: a fourth fully connected layer, a fifth fully connected layer, and a Sigmoid activation function layer connected in sequence.

[0024] Preferably, the loss function of the detection transformer network structure includes: position loss and illumination loss.

[0025] An embodiment of the present invention further provides a bow-catenary arc online detection system, comprising: a memory and a processor, wherein the memory stores a computer program executed by the processor, and the computer program executes a bow-catenary arc online detection method when executed by the processor.

[0026] An embodiment of the present invention further provides a storage medium, on which a computer program is stored, and when the computer program is run, a method for online detection of arcing between a bow and a catenary is executed.

[0027] The present invention is based on the online detection of bow-net arcing of the detection converter, and improves the model performance by optimizing the backbone network structure, the number of converter layers, and the customized loss function. It can adapt to different scenarios and conditions, reduce dependence on fixed algorithms and parameters, has stronger robustness, and can better maintain detection performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0029] Figure 1 This is a flow chart of the on-line detection method for bow-net arcing according to an embodiment of the present invention;

[0030] Figure 2 It is a data processing flow chart of online detection of bow-net arcing according to an embodiment of the present invention. DETAILED DESCRIPTION

[0031] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0032] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0033] Embodiment 1:

[0034] like Figure 1 , 2 As shown, an embodiment of the present invention provides a method for online detection of bow-net arcing, comprising:

[0035] Step S1, obtaining a bow-catcher arc detection model according to a bow-catcher data set;

[0036] Step S2: input the real-time acquired bow-net image into the bow-net arc detection model, output and judge whether there is arc.

[0037] As an implementation of the embodiment of the present invention, step S1 includes:

[0038] Step S11, constructing a detection converter network structure;

[0039] Step S12: train the detection converter network structure according to the bow-net data set to obtain the bow-net arc detection model

[0040] As an implementation method of the present invention, in step S11, the transformer network structure is detected to include: a backbone network, an encoder, a decoder and a prediction head; wherein,

[0041] The backbone network module includes: a ResNet18 network connected in sequence, a convolution layer with a convolution kernel size of 1×1, and a Flatten layer; the encoder module includes: 1 embedding layer and 3 encoding layers; the encoding layer includes: a first multi-head self-attention layer, a first residual connection and normalization layer, a first feedforward network layer, and a second residual connection and normalization layer connected in sequence; the position encoding of the encoding layer uses sinusoidal position encoding; the decoder module includes: 3 decoding layers; the decoding layer includes: a second multi-head self-attention layer, a third residual connection and normalization layer, a second multi-head attention layer, a fourth residual connection and normalization layer, a first feedforward network layer, and a fifth residual connection and normalization layer connected in sequence; the number of queries in the decoding layer is 2; the prediction head module includes: a second feedforward network layer and a third feedforward network layer connected in parallel;

[0042] The first feedforward network layer includes: a first fully connected layer, a LeakyReLU activation function layer, a discard layer, and a second fully connected layer connected in sequence; the second feedforward network layer includes: a third fully connected layer and a Softmax layer connected in sequence; the third feedforward network layer includes: a fourth fully connected layer, a fifth fully connected layer, and a Sigmoid activation function layer connected in sequence.

[0043] As an implementation method of an embodiment of the present invention, in step S12, the bow-net arc drawing data set includes images and corresponding labels, and the image shooting conditions include different time periods and different environments. The label data consists of the center point coordinates, target frame height and width. The embodiment of the present invention has 5,300 pictures and labels, with a training set ratio of 80% and a test set ratio of 20%.

[0044] Further, the training process of step S12 includes:

[0045] Step S121: Take the training set as input, extract features through ResNet18, embed the spatial position through the embedding layer, pass through the encoder, decoder, and prediction head, until the prediction result is obtained according to the confidence threshold; in the prediction result, They are the horizontal and vertical coordinates of the center point of the arc detection frame, are the width and height of the arc detection box respectively; cls is the category index of the prediction result, and conf is the confidence level of the model in the current prediction result;

[0046] Step 122, design the loss function L as follows:

[0047]

[0048] Among them, the superscript ^ indicates the predicted value; is the confidence of the prediction result; L1 is the position loss; L2 is the illumination loss; λ is the proportional coefficient of illumination loss; log(·) represents the logarithmic function;

[0049] The position loss L1 formula is as follows:

[0050]

[0051] Among them, b, are the position information of the real box and the predicted box, respectively. They are normalized vectors, including the horizontal coordinate of the center point of the box, the vertical coordinate of the center point of the box, the width of the box, and the height of the box; L giou is the GIoU loss between the real box and the predicted box; L L1 is the L1 loss between the real box and the predicted box; giou ,λ L1 They are the proportional coefficients of GIoU loss and L1 loss respectively;

[0052] The formula for illumination loss L2 is as follows:

[0053]

[0054] Among them, μ, are the mean brightness values ​​of the images inside the real box and the predicted box respectively; c is the proportional coefficient of local contrast loss; c, are the local contrasts of the real box and the predicted box, respectively, and the calculation formulas are

[0055]

[0056] Among them, M, are the number of pixels in the real frame and the predicted frame respectively; I c,j , are the values ​​of the jth pixel in the real box and the predicted box respectively; I s , are the average values ​​of the pixels outside the real box and the predicted box respectively;

[0057] Step 123, using the Adam optimization algorithm, continuously update the network parameters according to the latest calculated loss function value until the loss function value converges or reaches the preset number of iterations, and obtain the bow-net arc detection model; wherein, the learning rate is 0.0005, and the preset number of iterations is 300 steps.

[0058] As an implementation of the embodiment of the present invention, step S2 includes:

[0059] Step 21, input the real-time image of the bow-net into the bow-net arc detection model, obtain the feature vector through ResNet18, and then pass through the encoder, decoder, and prediction head of the transformer in sequence until the final detection result is obtained according to the confidence threshold screening;

[0060] Step 22, the detection result is a vector of length 6, the first to fourth values ​​are the horizontal coordinate of the center point of the arc detection frame, the vertical coordinate of the center point, the width, and the height, that is, The fifth and sixth values ​​are the category index of the prediction result and the confidence level of the model in the current prediction result, namely cls and conf;

[0061] Step 23: Determine whether there is an arc according to the category index cls of the prediction result.

[0062] Embodiment 2:

[0063] The embodiment of the present invention further provides a device for detecting an arc between a bow and a catenary on-line, comprising:

[0064] A training module is used to obtain a bow-net arc detection model based on a bow-net data set;

[0065] The detection module is used to input the real-time acquired bow-net image into the bow-net arc detection model, and output and judge whether there is arc.

[0066] As an implementation of an embodiment of the present invention, the training module includes:

[0067] A construction unit, used for constructing a detection transformer network structure;

[0068] The training unit is used to train the detection transformer network structure according to the bow-net data set to obtain the bow-net arc detection model

[0069] As an implementation method of the present invention, the detection transformer network structure includes: a backbone network, an encoder, a decoder and a prediction head; wherein,

[0070] The backbone network module includes: a ResNet18 network connected in sequence, a convolution layer with a convolution kernel size of 1×1, and a Flatten layer; the encoder module includes: 1 embedding layer and 3 encoding layers; the encoding layer includes: a first multi-head self-attention layer, a first residual connection and normalization layer, a first feedforward network layer, and a second residual connection and normalization layer connected in sequence; the position encoding of the encoding layer uses sinusoidal position encoding; the decoder module includes: 3 decoding layers; the decoding layer includes: a second multi-head self-attention layer, a third residual connection and normalization layer, a second multi-head attention layer, a fourth residual connection and normalization layer, a first feedforward network layer, and a fifth residual connection and normalization layer connected in sequence; the number of queries in the decoding layer is 2; the prediction head module includes: a second feedforward network layer and a third feedforward network layer connected in parallel;

[0071] The first feedforward network layer includes: a first fully connected layer, a LeakyReLU activation function layer, a discard layer, and a second fully connected layer connected in sequence; the second feedforward network layer includes: a third fully connected layer and a Softmax layer connected in sequence; the third feedforward network layer includes: a fourth fully connected layer, a fifth fully connected layer, and a Sigmoid activation function layer connected in sequence.

[0072] As an implementation manner of the embodiment of the present invention, the loss function of detecting the transformer network structure includes: position loss and illumination loss.

[0073] Embodiment 3:

[0074] An embodiment of the present invention further provides a bow-catenary arc online detection system, comprising: a memory and a processor, wherein the memory stores a computer program executed by the processor, and the computer program executes a bow-catenary arc online detection method when executed by the processor.

[0075] Embodiment 4:

[0076] An embodiment of the present invention further provides a storage medium, on which a computer program is stored, and when the computer program is run, the method for online detection of arcing between a bow and a catenary is executed.

[0077] The embodiments described above are only descriptions of the preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A method for online detection of bow-net arcing, characterized in that: include: Step S1, obtaining a bow-catcher arc detection model according to a bow-catcher data set; Step S2: input the real-time acquired bow-net image into the bow-net arc detection model, output and judge whether there is arc.

2. The on-line detection method for bow-net arcing according to claim 1, characterized in that: Step S1 includes: Step S11, constructing a detection converter network structure; Step S12: training the detection converter network structure according to the bow-catenary data set to obtain a bow-catenary arc detection model.

3. The on-line detection method for bow-net arcing according to claim 2, characterized in that: The detection transformer network structure includes: backbone network, encoder, decoder and prediction head; among them, The backbone network module includes: a ResNet18 network connected in sequence, a convolution layer with a convolution kernel size of 1×1, and a Flatten layer; the encoder module includes: 1 embedding layer and 3 encoding layers; the encoding layer includes: a first multi-head self-attention layer, a first residual connection and normalization layer, a first feedforward network layer, and a second residual connection and normalization layer connected in sequence; the position encoding of the encoding layer uses sinusoidal position encoding; the decoder module includes: 3 decoding layers; the decoding layer includes: a second multi-head self-attention layer, a third residual connection and normalization layer, a second multi-head attention layer, a fourth residual connection and normalization layer, a first feedforward network layer, and a fifth residual connection and normalization layer connected in sequence; the number of queries in the decoding layer is 2; the prediction head module includes: a second feedforward network layer and a third feedforward network layer connected in parallel; The first feedforward network layer includes: a first fully connected layer, a LeakyReLU activation function layer, a discard layer, and a second fully connected layer connected in sequence; the second feedforward network layer includes: a third fully connected layer and a Softmax layer connected in sequence; the third feedforward network layer includes: a fourth fully connected layer, a fifth fully connected layer, and a Sigmoid activation function layer connected in sequence.

4. The on-line detection method for bow-net arcing according to claim 3, characterized in that: The loss functions for detecting the transformer network structure include: position loss and illumination loss.

5. An online detection device for bow-net arcing, characterized in that: include: A training module is used to obtain a bow-net arc detection model based on a bow-net data set; The detection module is used to input the real-time acquired bow-net image into the bow-net arc detection model, and output and judge whether there is arc.

6. The on-line detection device for arcing of a bow-net as claimed in claim 5, characterized in that: The training modules include: A construction unit, used for constructing a detection transformer network structure; The training unit is used to train the detection transformer network structure according to the bow-net data set to obtain the bow-net arc detection model.

7. The on-line detection device for bow-cat arcing according to claim 6, characterized in that: The detection transformer network structure includes: backbone network, encoder, decoder and prediction head; among them, The backbone network module includes: a ResNet18 network connected in sequence, a convolution layer with a convolution kernel size of 1×1, and a Flatten layer; the encoder module includes: 1 embedding layer and 3 encoding layers; the encoding layer includes: a first multi-head self-attention layer, a first residual connection and normalization layer, a first feedforward network layer, and a second residual connection and normalization layer connected in sequence; the position encoding of the encoding layer uses sinusoidal position encoding; the decoder module includes: 3 decoding layers; the decoding layer includes: a second multi-head self-attention layer, a third residual connection and normalization layer, a second multi-head attention layer, a fourth residual connection and normalization layer, a first feedforward network layer, and a fifth residual connection and normalization layer connected in sequence; the number of queries in the decoding layer is 2; the prediction head module includes: a second feedforward network layer and a third feedforward network layer connected in parallel; The first feedforward network layer includes: a first fully connected layer, a LeakyReLU activation function layer, a discard layer, and a second fully connected layer connected in sequence; the second feedforward network layer includes: a third fully connected layer and a Softmax layer connected in sequence; the third feedforward network layer includes: a fourth fully connected layer, a fifth fully connected layer, and a Sigmoid activation function layer connected in sequence.

8. The on-line detection device for arcing of a bow-net as claimed in claim 7, characterized in that: The loss functions for detecting the transformer network structure include: position loss and illumination loss.

9. An online detection system for bow-net arcing, characterized in that: include: A memory and a processor, wherein the memory stores a computer program executed by the processor, and when the computer program is executed by the processor, the method for online detection of bow-net arcing as claimed in any one of claims 1 to 4 is executed.

10. A storage medium, characterized in that: The storage medium stores a computer program, which, when running, executes the on-line detection method for bow-catenary arcing according to any one of claims 1 to 4.

Images