Lightning stroke active characteristic identification method for power transmission line image

Abstract

The invention discloses a lightning stroke active feature recognition method for a power transmission line image, and the method comprises the following steps: converting an image from rgb into a yuv matrix through matrix multiplication and addition, taking a brightness matrix for extracting edge information, and then extracting the edge feature information of the brightness matrix through a candy algorithm; traversing on the brightness matrix Y by using a 3 * 3 Gaussian kernel to perform convolution operation; calculating the horizontal axis gradient and the longitudinal axis gradient of each pixel of the brightness matrix by taking 3 * 3 as a window; traversing each pixel in the brightness matrix; in the horizontal axis gradient matrix, if the horizontal axis gradient value of the current pixel is greater than the gradient value of adjacent points in the horizontal direction or the brightness value of the current pixel is greater than 0.8 times of the brightness maximum value. According to the method, any data pre-annotation and model training work are not needed, and the overall development cost is lower; the calculated amount is small, the system can be deployed in terminal equipment, and the system complexity and communication cost are reduced; and the interpretability is high, the correlation of input and output is more intuitive, and the operation and maintenance work can be better guided.

Description

Technical Field

[0001] The invention relates to a lightning strike active feature recognition method, and in particular to a lightning strike active feature recognition method for a power transmission line image. Background Art

[0002] Transmission line online monitoring equipment has been widely installed on lines or towers in various regions and at various voltage levels to collect real-time data, including current, voltage, meteorology, images and other data. Among them, image-based online monitoring equipment can clearly reflect the actual situation on site because its output results are more intuitive. For example, line damage caused by large mechanical equipment or abnormal activities near towers can be better presented, so that the line operation and maintenance personnel can be notified as soon as possible to handle them and ensure the power supply reliability of the power grid. Among transmission line faults, tripping faults caused by lightning strikes account for the vast majority. Therefore, capturing pictures or videos of the moment when lightning flashes over and hits the line can greatly accelerate fault diagnosis and line maintenance, thereby providing a method for active lightning feature recognition of transmission line images to improve line operation stability.

[0003] Reference Figure 3 , one solution in the prior art is to use machine learning or deep learning algorithms to achieve this. The specific solution of this method is:

[0004] 1. Organize samples: collect a large number of historical pictures, combine them with the actual lightning tripping situation, and manually give each picture a label (for example, "lightning strike" or "no lightning strike");

[0005] 2. Feature extraction: Convert all pixel values ​​of each historical image sample into a one-dimensional vector x = [x1, x2, …, xn] as the input of the neural network model;

[0006] 3. Build a model: Establish a multi-layer neural network, including input layer, hidden layer, and output layer, where: the input layer is related to the number of image pixels; the hidden layer is composed of multiple layers, each layer contains multiple neurons, each neuron contains a linear function (i.e. y = wx + b, x is input, y is output, w is weight matrix, b is bias matrix) and an activation function (e.g. g(x) = max(0,x), used to control the output within a specific range to prevent the parameters from expanding out of control); the output layer contains 2 nodes, a and b, representing the labels of the algorithm output, if a>b, it means "there is a lightning strike", otherwise it means "there is no lightning strike";

[0007] 4. Training model: By feeding a large number of training samples into the built model, the back propagation algorithm is used to continuously optimize the parameters in the model, so that the model output labeling category is as close as possible to the manual classification category, and the trained model is saved;

[0008] 5. Model classification: Extract features from new samples and input them into the trained model to obtain the labeling result of the sample in the algorithm.

[0009] However, the principle of the neural network algorithm in the above-mentioned lightning active feature recognition method determines that the parameters in the network do not have the ability to reason. On the contrary, it is essentially data-driven and must use massive data to allow the parameters to continuously iterate to the most appropriate values. The neural network model often involves dozens or even hundreds of hidden layers, and each layer contains thousands of neurons; therefore, the amount of calculation for each image is very large, and the processor of the general terminal device is difficult to support, so most of them are uploaded to the cloud server first and then calculated; this increases the communication cost on the one hand, and increases the system complexity and the burden on the cloud server on the other hand.

[0010] Deep learning is essentially a black box. The algorithm is driven by data rather than mechanism, so it lacks explainability and the correlation between input and output is not intuitive enough. If errors are identified, it is difficult to make targeted improvements unless the data samples are significantly upgraded and retrained. Therefore, there is room for improvement. Summary of the invention

[0011] The present invention overcomes the shortcomings of the prior art and provides a method for identifying active lightning strike features of a power transmission line image.

[0012] To achieve the above object, the technical solution adopted by the present invention is: a method for identifying active lightning strike features of a transmission line image, comprising the following steps:

[0013] Step 1: Use matrix multiplication and addition to convert the image from RGB to YUV matrix, take the brightness matrix to extract edge information, and then use the candy algorithm to extract edge feature information of the brightness matrix; use a 3*3 Gaussian kernel to traverse the brightness matrix Y for convolution operation; use a 3*3 window to calculate the horizontal and vertical gradients of each pixel in the brightness matrix;

[0014] Traverse each pixel in the brightness matrix; in the horizontal gradient matrix, if the horizontal gradient value of the current pixel is greater than the gradient value of its horizontal adjacent point or its brightness value exceeds 0.8 times the brightness maximum, then retain the horizontal gradient value of the pixel, otherwise set to 0; in the vertical gradient matrix, if the vertical gradient value of the current pixel is greater than the gradient value of its vertical adjacent point or its brightness value exceeds 0.8 times the brightness maximum, then retain the vertical gradient value of the pixel, otherwise set to 0; combine the horizontal gradient matrix and the vertical gradient matrix, that is, only save the pixels with values ​​in the horizontal gradient matrix or the vertical gradient matrix in the brightness matrix Y, thereby completing non-maximum suppression;

[0015] Traverse each pixel in the luminance matrix Y. If the horizontal gradient value or the vertical gradient value of the pixel is greater than the high threshold, it is a strong edge point, and its pixel value is retained; if the horizontal gradient value or the vertical gradient value is greater than the low threshold and less than the high threshold, it is a weak edge point. If there is a strong edge point around it, its pixel value is retained, otherwise it is set to 0; the pixel values of other pixels are all set to 0, thus completing the double-threshold detection.

[0016] Step 2: Set a window of 11*11 to slide on the image matrix, with a horizontal step size of 1 and a vertical step size of 11. Calculate the mean value u within each window, record the position corresponding to the maximum mean value d, and calculate the standard deviation s of the abscissa; when a lightning current channel is formed, there is an optical reaction, and the flashover trajectory is tracked according to the luminance values of each pixel.

[0017] Step 3: Since the lightning flashover trajectory has the characteristics of high luminance and vertical downward, it can be judged whether it is a lightning flashover according to the luminance mean value u and its abscissa standard deviation s in the trajectory extracted by the above sliding window; taking the maximum mean value d of the window mean value above as a benchmark, search for pixels whose luminance values exceed 0.5 times of its maximum mean value around it, and record the positions of the pixels whose luminance values meet the conditions, which are the lightning flashover trajectories.

[0018] If u > 0.5d and s < p, the output is "lightning strike", otherwise the output is "no lightning strike".

[0019] p is the set threshold of 1.5.

[0020] In a preferred embodiment of the present invention, rgb respectively represent red, green, and blue, and yuv respectively represent luminance, hue, and saturation.

[0021] In a preferred embodiment of the present invention, the yuv matrix calculation formula is:

[0022]

[0023] In a preferred embodiment of the present invention, the Gaussian kernel matrix expression is:

[0024]

[0025] In a preferred embodiment of the present invention, the horizontal axis gradient expression is: G x = sum(-x(i - 1, j - 1) + x(i + 1, j - 1) - 2x(i - 1, j) + 2x(i + 1, j) - x(i - 1, j + 1) + x(i + 1, j + 1)); where x(i, j) is the current pixel luminance matrix value, and i, j represent the horizontal and vertical coordinate positions of the pixel.

[0026] In a preferred embodiment of the present invention, the vertical axis gradient expression is: G y=sum(x(i-1,j-1)+2x(i,j-1)+x(i+1,j-1)-x(i-1,j+1)-2x(i,j+1)-x(i+1,j+1)); where x(i,j) is the current pixel brightness matrix value, and i,j represent the horizontal and vertical coordinate positions of the pixel.

[0027] In a preferred embodiment of the present invention, the mean value u represents the average brightness value of a pixel set in a lightning flashover trajectory, and its calculation formula is: Where N represents the number of all pixels in the sliding window; I(x i ,y i ) represents the brightness value of each pixel.

[0028] In a preferred embodiment of the present invention, the horizontal coordinate standard deviation s represents the discrete degree of pixel distribution in the horizontal direction of the lightning flashover trajectory, which is calculated based on the horizontal coordinates of pixels whose brightness values ​​exceed a certain threshold in the flashover trajectory, and the calculation formula is: Among them, x i The x coordinates of these pixels are 1 , x 2 , ..., x n ; represents the mean of the horizontal axis, The calculation formula is:

[0029] A computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the steps of the above method are implemented when the processor executes the computer program.

[0030] A computer-readable storage medium stores a computer program, which implements the steps of the above method when executed by a processor.

[0031] The present invention solves the defects existing in the background technology and has the following beneficial effects:

[0032] The present invention is based on the image features of lightning, performs calculation operations on the image at the pixel level, and outputs an identification of whether the image has the features, thereby distinguishing whether the image contains lightning strikes; it does not require any data pre-labeling and model training, and the overall development cost is lower; the amount of calculation is small, and it can be deployed on terminal devices to reduce system complexity and communication costs; it has strong interpretability, and the correlation between input and output is more intuitive, which can better guide operation and maintenance work. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] 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 are briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative work.

[0034] Figure 1 A schematic diagram of the recognition process of a method for identifying active lightning strike features of a power transmission line image proposed by the present invention;

[0035] Figure 2 A schematic diagram of an image preprocessing process of a lightning strike active feature recognition method for a power transmission line image proposed by the present invention;

[0036] Figure 3 The present invention provides a schematic diagram of the prior art recognition process of a lightning strike active feature recognition method for a power transmission line image. DETAILED DESCRIPTION

[0037] 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.

[0038] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0039] In the description of the present application, it should be understood that the terms "center", "longitudinal", "lateral", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the scope of protection of the present application. In addition, the terms "first", "second", etc. are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, features defined as "first", "second", etc. may explicitly or implicitly include one or more of the features. In the description of the invention, unless otherwise specified, "multiple" means two or more.

[0040] In the description of this application, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium, and it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in this application can be understood by specific circumstances.

[0041] Transmission line online monitoring equipment has been widely used, especially visualization, including two visualization forms: photos and videos, which can very intuitively reflect the line operation status within the monitoring range of the equipment, so as to better guide the line operation and maintenance work. However, since the line distance can be as long as hundreds of kilometers and there are a large number of cameras deployed online, a large amount of image information will be continuously generated. In order to conduct rapid fault diagnosis and maintenance, line operation and maintenance personnel need to screen out a very small part of the pictures or video frames that can capture the flashover moment of the lightning fault, so the workload is very huge. Therefore, it is necessary to develop a lightning flashover recognition method, which automatically searches for pictures containing lightning flashover moments based on the image information of the transmission line visualization online monitoring equipment, thereby reducing the manpower burden.

[0042] However, there are a large number of such devices, which generate a huge amount of image data all the time, and the operation and maintenance personnel need to identify them one by one. Only by finding out the very few that actually have lightning flashover moments can further diagnosis and maintenance be meaningful. Therefore, it is necessary to design an algorithm to filter out the pictures without lightning flashover features in the on-site pictures of the transmission lines, and only give the pictures with lightning flashover features to the relevant personnel for further identification, which can greatly reduce the labor cost. Therefore, the following technical solutions are specially disclosed:

[0043] like Figure 1 and Figure 2 As shown, a method for active lightning feature recognition of a transmission line image comprises the following steps:

[0044] Use matrix multiplication and addition to convert the image from RGB to YUV matrix, where RGB stands for red, green and blue, and YUV stands for brightness, hue and saturation. The YUV matrix calculation formula is: The brightness matrix is ​​used to extract edge information, and then the candy algorithm is used to extract edge feature information of the brightness matrix; a 3*3 Gaussian kernel is used to traverse the brightness matrix Y for convolution operation; the Gaussian kernel matrix expression is: Calculate the horizontal and vertical gradients of each pixel in the brightness matrix using a 3*3 window;

[0045] The horizontal axis gradient expression is:

[0046] G x =sum(-x(i-1,j-1)+x(i+1,j-1)-2x(i-1,j)+2x(i+1,j)-x(i-1,j+1)+x(i+1,j+1))

[0047] ; Where x(i,j) is the current pixel brightness matrix value, and i,j represent the horizontal and vertical coordinate positions of the pixel.

[0048] The vertical axis gradient expression is:

[0049] G y =sum(x(i-1,j-1)+2x(i,j-1)+x(i+1,j-1)-x(i-1,j+1)-2x(i,j+1)-x(i+1,j+1))

[0050] ; Where x(i,j) is the current pixel brightness matrix value, and i,j represent the horizontal and vertical coordinate positions of the pixel.

[0051] The original non-maximum suppression is to refine the edge, but it is easy to cause edge discontinuity. And because the lightning flashover is narrow, in order to preserve the flashover trajectory, an optimization process is performed here:

[0052] Traverse each pixel in the brightness matrix; in the horizontal gradient matrix, if the horizontal gradient value of the current pixel is greater than the gradient value of its horizontal adjacent point or its brightness value exceeds 0.8 times the maximum brightness, then retain the horizontal gradient value of the pixel, otherwise set it to 0; in the vertical gradient matrix, if the vertical gradient value of the current pixel is greater than the gradient value of its vertical adjacent point or its brightness value exceeds 0.8 times the maximum brightness, then retain the vertical gradient value of the pixel, otherwise set it to 0;

[0053] Combine the horizontal gradient matrix and the vertical gradient matrix, that is, only save the pixels with values ​​in the horizontal gradient matrix or the vertical gradient matrix in the brightness matrix Y, thereby completing non-maximum suppression;

[0054] Traverse each pixel in the brightness matrix Y. If the horizontal gradient value or vertical gradient value of the pixel is greater than the high threshold, it is a strong edge point and its pixel value is retained; if the horizontal gradient value or vertical gradient value is greater than the low threshold and less than the high threshold, it is a weak edge point. If there are strong edge points around it, its pixel value is retained, otherwise it is set to 0; other pixel values ​​are all set to 0, thereby completing the double threshold detection;

[0055] Set an 11*11 window to slide on the image matrix with a horizontal step size of 1 and a vertical step size of 11. Calculate the mean value u within each window, record the position corresponding to the maximum mean value d, and calculate the standard deviation s of the abscissa. When a lightning current channel is formed, there is an optical reaction accompanied by it. Track the flashover trajectory according to the brightness values of each pixel.

[0056] Since the lightning flashover trajectory has the characteristics of high brightness and vertical downward direction, it is possible to judge whether it is a lightning flashover according to the mean brightness u and the standard deviation s of its abscissa in the trajectory extracted by the above sliding window. Based on the maximum value d of the window mean above, search for pixels whose brightness values exceed 0.5 times the maximum mean value around it, and record the positions of the pixels whose brightness values meet the conditions, which are the lightning flashover trajectories.

[0057] The mean value u represents the average brightness value of the pixel set in the lightning flashover trajectory, and its calculation formula is: Among them, N represents the number of all pixels in the sliding window; I(xi,yi) represents the brightness value of each pixel.

[0058] The standard deviation s of the abscissa represents the degree of dispersion of the pixel distribution of the lightning flashover trajectory in the horizontal direction. Its calculation is based on the abscissas of the pixels whose brightness values in the flashover trajectory exceed a certain threshold, and the calculation formula is: Among them, xi represents the abscissa x of these pixels 1 , x 2 ,..., x n ; represents the mean value of the abscissa, The calculation formula of

[0059] If u > 0.5d and s < p, the output is "lightning strike", otherwise the output is "no lightning strike"; p is the set threshold 1.5.

[0060] A computer device includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the above method.

[0061] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it implements the steps of the above method.

[0062] In the picture of the lightning strike moment, there is often a white high-brightness line, and the direction is basically perpendicular to the ground. Therefore, this method specifically proposes a method that can extract this feature to identify the lightning flashover moment. In principle, this method can effectively avoid the problems of large data volume, large calculation amount, and poor interpretability existing in the prior art.

[0063] In order to verify the advantages of the present invention over the prior art, the following comparative experiments are conducted:

[0064] The data of each category is divided into 150 groups, 90 groups of data are used for training, and 60 groups of data are used for testing. Among them, the labels of lightning strike and no lightning strike are defined as 0 and 1 respectively. A bidirectional LSTM layer with 95 hidden units is specified, and the last element of the sequence is output. Finally, 2 classes are specified by including a fully connected layer of size 4, followed by a softmax layer and a classification layer. The LSTM solver is specified as adam, the gradient threshold is 11, and the maximum number of rounds is 450. A total of 2800 iterations are performed. As the training progresses, the accuracy continues to improve, and the loss function tends to converge stably. The final training accuracy reaches 98.7%. Under the same configuration, the feature vectors are input into the BP neural network, support vector machine (SVM) and random forest algorithm (RF) for training and testing.

[0065] Table 1 shows the classification results of each method:

[0066]

[0067]

[0068] The results show that the LSTM classification method has a training set accuracy of 98.70% and a test set accuracy of 96.76%. Both results are significantly better than the other three methods. This proves that the LSTM classification method has higher accuracy and has a greater advantage over other classification algorithms, making significant progress.

[0069] The above is based on the ideal embodiment of the present invention. Through the above description, relevant personnel can make various changes and modifications without departing from the technical concept of the present invention. The technical scope of the present invention is not limited to the content in the specification, and the technical scope must be determined according to the scope of the claims.

Claims

1. A method for identifying active lightning strike features of a power transmission line image, characterized in that: It includes the following steps: Step 1: Convert the picture from RGB to a YUV matrix using matrix multiplication and addition. Take the luminance matrix for edge information extraction, and then use the Canny algorithm to extract the edge feature information of the luminance matrix; Perform a convolution operation by traversing with a 3×3 Gaussian kernel on the luminance matrix Y; Calculate the horizontal axis gradient and vertical axis gradient of each pixel in the luminance matrix with a 3×3 window; Traverse each pixel in the luminance matrix; In the horizontal axis gradient matrix, if the horizontal axis gradient value of the current pixel is greater than the gradient value of its adjacent point in the horizontal direction or its luminance value exceeds 0.8 times the maximum luminance value, then retain the horizontal axis gradient value of this pixel, otherwise set it to 0; In the vertical axis gradient matrix, if the vertical axis gradient value of the current pixel is greater than the gradient value of its adjacent point in the vertical direction or its luminance value exceeds 0.8 times the maximum luminance value, then retain the vertical axis gradient value of this pixel, otherwise set it to 0; Combine the horizontal axis gradient matrix and the vertical axis gradient matrix, that is, only save the pixels with values in either the horizontal axis gradient matrix or the vertical axis gradient matrix in the luminance matrix Y, thus completing non-maximum suppression; Traverse each pixel in the luminance matrix Y. If the horizontal gradient value or vertical gradient value of this pixel is greater than the high threshold, it is a strong edge point and retain its pixel value; If the horizontal gradient value or vertical gradient value is greater than the low threshold and less than the high threshold, it is a weak edge point. If there is a strong edge point around it, then retain its pixel value, otherwise set it to 0; The pixel values of others are all set to 0, thus completing double-threshold detection; Step 2: Set a 11×11 window to slide on the image matrix, with a horizontal step size of 1 and a vertical step size of 11. Calculate the mean value u within each window, record the position corresponding to the maximum mean value d, and calculate the standard deviation s of the abscissa; When a lightning current channel is formed, there is an optical reaction accompanied by it, and track its flashover trajectory according to the luminance values of each pixel; Step 3: Since the lightning flashover trajectory has the characteristics of high luminance and vertically downward, it is possible to judge whether it is a lightning flashover according to the mean luminance u and its abscissa standard deviation s in the trajectory extracted by the above sliding window; Based on the maximum window mean value d above, search for pixels whose luminance values exceed 0.5 times its maximum mean value around, and record the positions of the pixels whose luminance values meet the conditions, which are the lightning flashover trajectories; If u > 0.5d and s < p, then output "There is a lightning strike", otherwise output "There is no lightning strike"; p is the set threshold of 1.

5.

2. The method for identifying active lightning strike features of a power transmission line image according to claim 1, characterized in that: The RGB respectively represent red, green, and blue, and the YUV respectively represent luminance, hue, and saturation.

3. The method for identifying active lightning strike features of a power transmission line image according to claim 1, characterized in that: The calculation formula of the YUV matrix is:

4. The method for identifying active lightning strike features of a power transmission line image according to claim 3 is characterized in that: The expression of the Gaussian kernel matrix is:

5. The method for identifying active lightning strike features of a power transmission line image according to claim 1, characterized in that: The expression of the horizontal axis gradient is: G x =sum(-x(i-1,j-1)+x(i+1,j-1)-2x(i-1,j)+2x(i+1,j)-x(i-1,j+1)+x(i+1,j+1)); where x(i,j) is the current pixel brightness matrix value, and i,j represent the horizontal and vertical coordinate positions of the pixel.

6. The method for identifying active lightning strike features of a power transmission line image according to claim 5, characterized in that: The expression of the vertical axis gradient is: G y =sum(x(i-1,j-1)+2x(i,j-1)+x(i+1,j-1)-x(i-1,j+1)-2x(i,j+1)-x(i+1,j+1)); where x(i,j) is the current pixel brightness matrix value, and i,j represent the horizontal and vertical coordinate positions of the pixel.

7. The method for identifying active lightning strike features of a power transmission line image according to claim 1, characterized in that: The mean value u represents the average luminance value of the pixel set in the lightning flashover trajectory, and its calculation formula is: Where N represents the number of all pixels in the sliding window; I(x i ,y i ) represents the brightness value of each pixel.

8. The method for identifying active lightning strike features of a power transmission line image according to claim 7, characterized in that: The abscissa standard deviation s represents the degree of dispersion of the pixel distribution of the lightning flashover trajectory in the horizontal direction, and its calculation is based on the abscissas of the pixels whose luminance values in the flashover trajectory exceed a certain threshold, and the calculation formula is: Among them, xi represents the horizontal coordinates x1, x2, ..., x n ; represents the mean of the horizontal axis, The calculation formula is:

9. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, it implements the steps of the method described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the method described in any one of claims 1 to 8.

Images