Cable identification and positioning method combining data fusion and image identification

Abstract

The invention relates to a cable identification and positioning method combining data fusion and image identification, and belongs to the field of power cables. According to the method, a denoising algorithm based on a Gaussian function and an improved Sobel edge detection algorithm are adopted for image preprocessing, and cable features are extracted. And then, fusing multi-source data by using an extended Kalman filtering algorithm based on a multi-innovation theory to obtain accurate position information of the cable. The system is high in automation degree, high in identification and positioning efficiency, high in positioning precision and high in adaptability, can obtain cable position information in real time, and provides powerful support for maintenance and repair of electric power and communication facilities.

Description

Technical Field

[0001] The invention belongs to the field of power cables and relates to a cable identification and positioning method combining data fusion and image recognition. Background Art

[0002] Underground cables are a key part of the power transmission process. The normal transportation of electricity ensures that the power grid can provide reliable power supply services to all customers and has a strong power supply guarantee capability in emergency situations. In order to ensure the normal operation of electrical equipment, timely grasp of the operating data and status of the cable is a basic requirement.

[0003] Currently, the identification and location of cables in cable trenches are crucial for maintenance and repair work. Cable trenches are usually laid underground to accommodate power cables, communication cables and other types of cables. Due to the large number of cables, complex types and harsh environmental conditions, traditional cable identification and location methods face many challenges.

[0004] Based on the above background, the present invention proposes a method for identifying and locating cables in a cable trench, which combines a data fusion algorithm to integrate data from different sensors such as GPS to improve the accuracy and stability of cable positioning, and then combines an image processing method to provide an efficient and accurate cable identification and positioning method.

[0005] (1) Manual recording method

[0006] Manual recording is the simplest and most primitive inspection method. The staff needs to list the detailed list of inspection tasks in the workbook, and then for each inspection item, the staff will check whether the parameters of the item in the workbook meet the standards and record them in the workbook.

[0007] (2) RFID-based inspection method

[0008] RFID, or radio frequency identification technology, is an electronic tag communication technology based on the Internet of Things. It can identify specific targets and read related data through radio signals without any mechanical or optical contact.

[0009] Compared with the existing cable positioning and identification methods, the cable identification and positioning method in the cable trench provided by the present invention adopts automatic identification, reduces manual intervention, and improves the efficiency of identification and positioning. In addition, the data fusion algorithm is used to improve the accuracy and robustness of identification and positioning. It is applicable to various types of cables and complex cable trench environments. Through the method of the present invention, the efficiency and accuracy of cable identification and positioning in the cable trench can be significantly improved, the error of manual operation can be reduced, and strong support can be provided for the maintenance and repair of power and communication facilities.

[0010] This method can ensure high-precision positioning information in various environments, especially in densely populated urban areas and underground environments. Real-time acquisition and update of cable location data ensures the accuracy and timeliness of positioning information. Develop reliable image recognition algorithms that can accurately distinguish and identify different types of cables, process cable images in complex backgrounds, and improve recognition accuracy and robustness, including recognition in low light, occlusion, and cluttered environments. Summary of the invention

[0011] In view of this, an object of the present invention is to provide a cable identification and positioning method combining data fusion and image recognition.

[0012] In order to achieve the above object, the present invention provides the following technical solutions:

[0013] A cable identification and positioning method combining data fusion and image recognition, the method comprising the following steps:

[0014] Get images of cables in the cable trench;

[0015] The cable image is preprocessed, including:

[0016] The cable image is denoised using a denoising algorithm based on a Gaussian function;

[0017] The improved Sobel edge detection algorithm is used to perform edge detection on the denoised cable image;

[0018] Extracting cable features from the edge-detected cable image;

[0019] The cable features are fused using an extended Kalman filter algorithm based on multiple innovation theory to obtain the location information of the cable.

[0020] Furthermore, the denoising algorithm based on the Gaussian function comprises the following steps:

[0021] Step 1: Move the central element in the cable image to be tested so that the central element of the cable image is directly above the image;

[0022] Step 2: The pixel value of the input cable surface image is used as a weight value and multiplied by its corresponding kernel function to obtain a corresponding value;

[0023] Step 3: Add the values ​​obtained in step 1 and step 2 and output them;

[0024] Step 4: Classify the noise of the cable image according to the following formula;

[0025]

[0026] In the formula, (x, y) represents the points of all pixels of the cable image, f(x, y) represents the gray value of the cable surface image; the value of A is the point of the cable image pixel value determined to contain other noise; the value of B is the point of the cable image pixel value containing Gaussian noise; Min[f(x, y)] is the minimum gray value in the cable image within the 5×5 window of the improved denoising algorithm; Max[f(x, y)] is the maximum gray value in the cable image within the 5×5 window of the improved denoising algorithm; according to the above formula, the suspicious other noise or Gaussian noise points in the cable gray image can be analyzed one by one;

[0027] In order to determine the range of cable grayscale image difference values, a judgment threshold is defined:

[0028]

[0029] Where: g is the defined judgment threshold; h is the offset in the horizontal direction (x-axis); j is the offset in the vertical direction (y-axis); n is the window size of the neighborhood;

[0030] Step 5: De-noise the classified cable image according to the following formula;

[0031]

[0032] In the formula, S represents the output cable image; median filtering represents the median filtering, which performs median filtering on (x, y); A represents the point containing other noise in the cable image pixel, and B represents the point containing Gaussian noise in the cable image pixel; the noise of the cable image is filtered out by the improved denoising algorithm to obtain a high-quality cable image.

[0033] Further, the improved Sobel edge detection algorithm comprises the following steps:

[0034] Step 11: Use an eight-directional Sobel operator template of size 3x3 to perform convolution processing with the cable image;

[0035] Step 12: Find the maximum value of all elements of the array n obtained in step 11;

[0036] Step 13: Assign the maximum value to the central pixel as the output value.

[0037] Further, the extended Kalman filter algorithm based on multiple innovation theory includes the following steps:

[0038] Prediction stage:

[0039] Based on the current system state and system control input, predict the system state at the next moment;

[0040] According to the predicted system state, predict the observation information at the next moment;

[0041] Observation phase:

[0042] Obtain actual observation information;

[0043] Update phase:

[0044] Calculate the difference between the observed information and the predicted observed information, i.e. the new information;

[0045] Use the new information to update the system state and obtain a more accurate estimate of the system state.

[0046] Furthermore, the cable characteristics include characteristics of shape, size and color of the cable.

[0047] The beneficial effects of the present invention are: in view of the problem that the identification and positioning of cables inside the existing cable trench rely on manual work, which consumes time and manpower, the cable identification and positioning method in the cable trench provided by the present invention adopts automatic identification, reduces manual intervention, improves the identification and positioning efficiency, and provides strong support for the maintenance and inspection of power and communication facilities.

[0048] Other advantages, objectives and features of the present invention will be described in the following description to some extent, and to some extent, will be obvious to those skilled in the art based on the following examination and study, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below in conjunction with the accompanying drawings, wherein:

[0050] Figure 1 This is the flow chart of cable image recognition and positioning;

[0051] Figure 2 It is the filtering flow chart of Gaussian function;

[0052] Figure 3 It is the Sobel eight-direction edge detection algorithm template. DETAILED DESCRIPTION

[0053] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner, and the following embodiments and features in the embodiments can be combined with each other without conflict.

[0054] Among them, the drawings are only used for illustrative explanations, and they only represent schematic diagrams rather than actual pictures, and should not be understood as limitations on the present invention. In order to better illustrate the embodiments of the present invention, some parts of the drawings may be omitted, enlarged or reduced, and do not represent the size of actual products. For those skilled in the art, it is understandable that some well-known structures and their descriptions in the drawings may be omitted.

[0055] The same or similar numbers in the drawings of the embodiments of the present invention correspond to the same or similar parts; in the description of the present invention, it should be understood that if the terms "upper", "lower", "left", "right", "front", "rear", etc. indicate the orientation or position relationship, they are based on the orientation or position relationship shown in the drawings, which is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation. Therefore, the terms describing the position relationship in the drawings are only used for illustrative purposes and cannot be understood as limiting the present invention. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to specific circumstances.

[0056] Figure 1 Flow chart of cable image recognition and positioning

[0057] 1. Improved EKF algorithm based on multi-innovation theory

[0058] EKF is an extension of Kalman filtering and is suitable for nonlinear systems. Unlike Kalman filtering, EKF handles nonlinear systems by performing Taylor expansion on the nonlinear state equation and linearizing it. In cable positioning, if nonlinear dynamic or measurement models are involved, EKF can more effectively integrate the data from various sensors.

[0059] The multi-innovation identification theory is to generalize the single-innovation correction technology, propose the concept of multi-innovation correction technology identification from the perspective of innovation correction, and establish the multi-innovation correction identification method, referred to as the multi-innovation identification method. As the name suggests, the multi-innovation algorithm generalizes the innovation. For a scalar system, the scalar innovation e(t)∈R in the algorithm is generalized to the innovation vector E(p,t)∈R p, that is, multiple innovations, in order to make the matrix multiplication dimension compatible, the gain vector L(t)∈R n “It must be generalized to the gain matrix Γ(p,t)∈R n×p , so the innovation identification algorithm can be written as follows:

[0060]

[0061] Where Γ(p,t)∈R n×p is the gain matrix, E(p,t)∈R p is the new information vector, and p≥1 is the new information length.

[0062] EKF assumes that the sensor noise information follows a Gaussian distribution, so it uses the system state X and the covariance matrix P to describe it, as shown in formula (3):

[0063] X(x+1)=f(X(x),μ(x),v(x)) (2)

[0064] Among them, X(x) represents the system state at time x, μ(x) represents the system control input vector, v(x) represents Gaussian white noise with zero mean, and the transformation matrix of v(x) is Q(x). Then it is expressed as shown in formulas (3) and (4):

[0065] E(v(x))=0 (3)

[0066] E(v(x)v(x) T )=Q(x) (4)

[0067] EKF is divided into three stages: prediction, observation, and update. In the prediction process, the system state at time x+1 is expressed as The observation information of the system is expressed as In the observation phase, the observation model of the system is described as follows:

[0068] Z(x+1)=h(X(x+1))+w(x+1) (5)

[0069] Among them, Z(x+1) is the observation data obtained in the observation phase, and w(x+1) is Gaussian white noise with a mean of zero.

[0070] Finally, in the update phase, the state of the system at time k+1 is updated through the observation information Z(x+1). The expression of the scalar new information e(x) in the update equation is as follows:

[0071]

[0072] The scalar innovation e(x) is expanded to the innovation matrix E(p,x), and the system observation data Z(x) is expanded to Z(p,x), the measurement noise v(x) is expanded to V(p,x), and the expanded filter gain matrix is ​​K(p,x). They are expressed as follows:

[0073]

[0074] Where p is the length of the new information.

[0075] The prediction equation of the standard EKF is shown as follows:

[0076]

[0077]

[0078] Among them, x k is the system state matrix at time k, f(·) is a nonlinear function called the state transfer function, which calculates the state at time k based on the state at time k-1. is a priori estimate The corresponding covariance matrix.

[0079] Substituting formula (7) into (8) and (9), it can be expanded to:

[0080]

[0081] in, It is the prior estimate at time k, that is, the value obtained by solving the state prediction equation using the best estimate at time k-1, which represents the system estimate. i (k)(i=1,2…,P) represents the Kalman gain matrix. E(p,k) is the innovation matrix composed of innovations at multiple moments. This is the best estimate of the system state after correction.

[0082] 2. Improved denoising algorithm based on Gaussian function

[0083] Whether it is median filtering or mean filtering, it is not possible to filter out the noise of cable images containing Gaussian and other noises of different concentrations at the same time, and the cable edge information is not clear, indicating that a single noise removal algorithm cannot meet the actual needs of production and life. Therefore, it is necessary to design an algorithm that can filter out the noise of cable images containing other noises and Gaussian noise, so as to make the cable image quality and cable edge information closer to reality, so it is necessary to perform Gaussian smoothing first and then make a judgment.

[0084] The Gaussian smoothing filter is a very effective low-pass filter in the actual image processing algorithm in the spatial domain or frequency domain. The following are the important properties of the Gaussian function: When the Gaussian function acts on discrete signal data, even if the data has various directions, the Gaussian function can still maintain consistency in any direction. When a signal in the image is affected by the Gaussian function, the weight may be different, but more image data information can be retained. Because of its own unbounded nature, the Gaussian function allows users to continuously expand it to obtain the required image information.

[0085] like Figure 2 As shown, the main steps of Gaussian function filtering are as follows:

[0086] Step 1: Move the central element in the cable image to be tested so that the central element of the cable image is directly above the image.

[0087] Step 2: Take the pixel value of the input cable surface image as the weight value and multiply it with the corresponding kernel function to obtain the corresponding value.

[0088] Step 3: Add the values ​​obtained in step 1 and step 2 and output them.

[0089] According to the following formula, we can judge which pixels in the cable image are Gaussian noise and which are other noise. After distinguishing them, we can apply the corresponding denoising algorithm according to the corresponding pixels. Generally speaking, the pixel value of other noise is obviously different from the value of other pixels in the same image. If the pixel value of the image at this point is found to be significantly higher or lower than the pixel value of other points during retrieval, then it can be determined that the pixel at this point is the other noise point of the image. Otherwise, it is considered to be the Gaussian noise point of the image. Formula (10) can be used to make a preliminary judgment on other noise points and Gaussian noise points in the cable image:

[0090]

[0091] In the formula: (x, y) represents the points of all the pixels of the cable image, f(x, y) represents the gray value of the cable surface image; the value of A is the point of the cable image pixel value determined to contain other noise; the value of B is the point of the cable image pixel value containing Gaussian noise; Min[f(x, y)] is the minimum gray value in the cable image within the 5×5 window of the improved denoising algorithm; Max[f(x, y)] is the maximum gray value in the cable image within the 5×5 window of the improved denoising algorithm. Based on the above formula, the suspicious other noise or Gaussian noise points in the cable gray image can be analyzed one by one.

[0092] In order to determine the range of cable grayscale image difference values, a judgment threshold needs to be defined:

[0093]

[0094] Where: g is the defined judgment threshold.

[0095] If the absolute value of the difference between the most central value of all the grayscale values ​​in the grayscale image pixels of the cable surface and the grayscale value of any other pixel in the neighborhood of the cable image center is greater than the threshold value g, it is determined that there is an obvious difference between the two points of the cable grayscale image. According to the above method, the values ​​of other points of the image are subtracted from the center point value in turn, and statistics are made. If the number of points with subtracted values ​​greater than the threshold value is greater, it means that the possibility of other noise in the area is greater, otherwise it is other noise.

[0096] The pixels of the classified cable surface grayscale image are processed according to the following formula:

[0097]

[0098] Where: S represents the output cable image; A represents the point in the cable image pixel containing other noise, and B represents the point in the cable image pixel containing Gaussian noise. According to the above formula, the improved denoising algorithm can filter out the noise of the cable image and obtain a high-quality cable image.

[0099] 3. Based on the improved Sobel edge detection algorithm

[0100] The weighted average filtering algorithm, also known as linear filtering, mainly replaces the grayscale value of each pixel with the average grayscale value of its neighboring pixels. The basic principle is: for any edge point to be processed in an image, a filter pattern is constructed with it as the core, and all signal points on the pattern are weighted averaged, that is, the grayscale value of the noise point in the center of the model, and then the filtered image is output. For example, for an image F with a size of P×Q and a window size of K×K, the weighted average filtering formula is:

[0101]

[0102] Where 1 = (k-1) / 2, f(x,y) is the image pixel value of image F at point (x,y), and w ij is the weight coefficient of the image pixel.

[0103] When the Sobel edge detection operator is used for detection, relatively continuous information about the edge direction is given. This method can well maintain the contour features of the original image. The Sobel operator selects any position in the operator image to generate a corresponding gradient vector and uses a discrete difference operator to calculate the image brightness gradient. This operator does not introduce any additional noise parameters.

[0104] The algorithm is improved based on the Sobel operator, that is, the Sobel algorithm is first used for edge detection after weighted average filtering. Previous literature believes that the Sobel operator is extended to four directions of 0°, 45°, 90° and 135°. Although the edge detection effect of the image has been improved, some image details are still lost. The Sobel eight-direction edge detection algorithm detects image edges that are relatively complete, with clear contours and strong continuity. Figure 3 It is the template of the eight-directional Sobel operator.

[0105] The algorithm in this paper uses an eight-direction edge detection template of size 5×5. The eight directions are 0°, 22.5°, 45°, 67.5°, 90°, 112.5°, 135° and 157.5°. Figure 3 In the given template, the weights w(m,n) of different positions can be calculated by the following formula:

[0106]

[0107] lng(m,n)=-[d(m,n) 2 -u]ln2 (16)

[0108] w(m,n)=[g(m,n)] (17)

[0109] Where d(m,n) is the Euclidean distance from the model unit with coordinates (m,n) to the model center with coordinates (i,j), u is the adjustment coefficient, and g(m,n) is the effective weight in (m,n). In the formula, "[]" indicates rounding operation.

[0110] The smaller the absolute value of the distance from the model pixel to the center pixel, the higher the weight; conversely, the larger the absolute value of the distance from the model pixel to the center pixel, the lower the weight. The algorithm uses the maximum grayscale value in the eight buffers as the output value of the current image pixel.

[0111] First, let the matrix [f(i,j)] be the image data matrix to be detected (where (i,i) represents the coordinates of each pixel). f(i,j) represents the gray value (i,i) of the image, w[f(i,j)] represents a 5×5 window centered at point (i,j), and g(i,Y) is the edge detected at point (i,Z).

[0112] Algorithm execution steps: set m k (k=1,2...,8) are the given 8 5×5 templates, and n is an integer array containing 8 elements. Each pixel in the image (with coordinates (i,j)) is processed as follows:

[0113] Step 11: Using Template m k(k=1,2...,8) is convolved with w[f(i,j)] and the absolute value of the convolution result to form (i,j);

[0114] Step 12: Find the maximum value of all elements of the array n obtained in the first step, initialize max = n[0], k = 1; if n[k]>max, set max = n[k]; if k<8, execute k=k+1 and go to step 12, otherwise go to step 13;

[0115] Step 13: Assign max to g(i,j) as the pixel output value.

[0116] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution, which should be included in the scope of the claims of the present invention.

Claims

1. A cable identification and positioning method combining data fusion and image recognition, characterized in that: The method comprises the following steps: Get images of cables in the cable trench; The cable image is preprocessed, including: The cable image is denoised using a denoising algorithm based on a Gaussian function; The improved Sobel edge detection algorithm is used to perform edge detection on the denoised cable image; Extracting cable features from the edge-detected cable image; The cable features are fused using an extended Kalman filter algorithm based on multiple innovation theory to obtain the location information of the cable.

2. The cable identification and positioning method combining data fusion and image recognition according to claim 1 is characterized in that: The Gaussian function-based denoising algorithm comprises the following steps: Step 1: Move the central element in the cable image to be tested so that the central element of the cable image is directly above the image; Step 2: The pixel value of the input cable surface image is used as a weight value and multiplied by its corresponding kernel function to obtain a corresponding value; Step 3: Add the values ​​obtained in step 1 and step 2 and output them; Step 4: Classify the noise of the cable image according to the following formula; In the formula, (x, y) represents the points of all pixels of the cable image, f(x, y) represents the gray value of the cable surface image; the value of A is the point of the cable image pixel value determined to contain other noise; the value of B is the point of the cable image pixel value containing Gaussian noise; Min[f(x, y)] is the minimum gray value in the cable image within the 5×5 window of the improved denoising algorithm; Max[f(x, y)] is the maximum gray value in the cable image within the 5×5 window of the improved denoising algorithm; according to the above formula, the suspicious other noise or Gaussian noise points in the cable gray image can be analyzed one by one; In order to determine the range of cable grayscale image difference values, a judgment threshold is defined: Among them, g represents the judgment threshold; h represents the offset on the horizontal x-axis; j represents the offset on the vertical y-axis; n represents the window size of the neighborhood; Step 5: De-noise the classified cable image according to the following formula; Where S represents the output cable image; medianfiltering(x,y) represents the median filter, which performs median filtering on (x,y); A represents the point containing other noise in the cable image pixel, and B represents the point containing Gaussian noise in the cable image pixel; the noise of the cable image is filtered out by the improved denoising algorithm to obtain a high-quality cable image.

3. The cable identification and positioning method combining data fusion and image recognition according to claim 1 is characterized in that: The improved Sobel edge detection algorithm comprises the following steps: Step 11: Use an eight-directional Sobel operator template of size 3x3 to perform convolution processing with the cable image; Step 12: Find the maximum value of all elements of the array n obtained in step 11; Step 13: Assign the maximum value to the central pixel as the output value.

4. The cable identification and positioning method combining data fusion and image recognition according to claim 1, characterized in that: The extended Kalman filter algorithm based on multiple innovation theory includes the following steps: Prediction stage: Based on the current system state and system control input, predict the system state at the next moment; According to the predicted system state, predict the observation information at the next moment; Observation phase: Obtain actual observation information; Update phase: Calculate the difference between the observed information and the predicted observed information, i.e. the new information; Use the new information to update the system state and obtain a more accurate estimate of the system state.

5. The cable identification and positioning method combining data fusion and image recognition according to claim 1, characterized in that: The cable characteristics include characteristics of the shape, size and color of the cable.

Images