A cable external damage early warning method and device and a storage medium

Abstract

The present application relates to the field of cable monitoring, and discloses a cable external damage early warning method, device and storage medium, the method comprising the following steps: deploying a camera in a preset range of a cable to be monitored; the camera communicates with a cloud server; the camera is used to obtain a monitoring image; a target detection model is constructed on the cloud server side and pre-trained; the monitoring image is input into the trained target detection model to obtain a target detection result; if the target detection result belongs to a preset target, the cloud server sends a warning to a terminal; otherwise, a sound signal is obtained according to the camera, and a vibration signal is extracted according to adjacent frames of monitoring images; a pre-trained machine learning model is used to generate a warning result according to the vibration signal and the sound signal. The present application has the advantages of low cost, strong environmental adaptability, and good balance between detection effect and speed.

Description

Technical Field

[0001] This invention relates to the field of cable monitoring, and in particular to a method, device and storage medium for early warning of cable damage from external sources. Background Technology

[0002] Against the backdrop of rapid urbanization, underground cables, as the arteries of urban power transmission, are threatened by external damage, especially by construction excavation, illegal intrusion, and other activities that lead to cable damage, affecting the stability of urban power supply and public safety.

[0003] Traditional monitoring methods that rely on manual inspections and fixed-point vibration sensors are not only costly, but also difficult to achieve real-time monitoring with full coverage and all-time hours. Especially in complex and ever-changing urban environments, factors such as visual obstruction and nighttime hours limit the effectiveness of monitoring.

[0004] An existing cable line anti-vandalism monitoring system and method based on DAS target sensing technology, especially distributed fiber optic sensing technology, is powerful but may involve high initial construction and maintenance costs, including specialized equipment (such as lasers, EDFA amplifiers, couplers, etc.) and complex system integration, as well as the need for professional technicians, which increases the overall complexity and economic burden of the system.

[0005] In addition, a monitoring method, device, equipment, and storage medium for external force damage to optical cables are also provided. This method involves a large amount of data analysis and processing, including steps such as vibration waveform analysis, risk value calculation, event point identification, and loss risk assessment. Natural vibrations in marine environments (tidal surges, biological activities) or non-hazardous and irrelevant events may also trigger alarms. Precise threshold calibration and algorithms are needed to reduce false positive results; otherwise, it may affect operation and maintenance efficiency and waste resources.

[0006] In summary, providing a low-cost, widely applicable cable anti-cable early warning method with a certain level of detection accuracy and real-time capability is an urgent problem to be solved. Summary of the Invention

[0007] The purpose of this invention is to propose a method, device, and storage medium for early warning of cable damage from external sources, in order to solve the technical problems of high cost and inability to simultaneously achieve detection accuracy and real-time performance in existing technologies for cable damage prevention.

[0008] A method for early warning of external damage to cables includes the following steps:

[0009] S1: Deploy cameras within a preset range of the cable to be monitored; the cameras communicate with the cloud server;

[0010] S2: Use a camera to acquire surveillance images;

[0011] S3: Build an object detection model on the cloud server and pre-train it to obtain a pre-trained object detection model;

[0012] S4: Input the monitored image into the trained target detection model to obtain the target detection result;

[0013] S5: If the target detection result belongs to the preset target, then issue an alert to the terminal through the cloud server; otherwise, proceed to step S6.

[0014] S6: Acquire sound signals from the camera and extract vibration signals from adjacent frames of the monitoring image;

[0015] S7: Based on the vibration and sound signals, a pre-trained machine learning model is used to generate early warning results.

[0016] A storage medium storing instructions and data for implementing a cable damage prevention early warning method.

[0017] A cable damage prevention early warning device includes: a processor and a storage medium; the processor loads and executes instructions and data in the storage medium to implement a cable damage prevention early warning method.

[0018] The beneficial effects provided by this invention are:

[0019] 1) Non-contact and non-destructive monitoring:

[0020] The system of this invention does not require direct contact with cables or related equipment, thereby avoiding potential physical damage to the cables or equipment. This non-contact monitoring method not only ensures the integrity of cables and equipment but also significantly reduces installation and maintenance costs, and improves the reliability and service life of the system.

[0021] 2) Comprehensive monitoring integrating visual and acoustic technologies:

[0022] By integrating video image processing and sound feature extraction technologies, this invention achieves dual visual and acoustic monitoring. Within the visible range, the system can directly observe through video images; in areas with limited or no line of sight, potential threats are identified through image analysis and sound feature extraction. This comprehensive monitoring approach greatly improves the comprehensiveness and accuracy of the early warning system.

[0023] 3) Strong environmental adaptability:

[0024] Even in the presence of obstructions or complex environments, such as tree branches or eaves blocking the line of sight, the system of this invention can still extract vibration and sound features through image processing technology, effectively overcoming the line-of-sight limitations of traditional monitoring systems. This strong environmental adaptability enables the invention to achieve excellent monitoring results in various practical application scenarios.

[0025] 4) Efficient early warning and preventative maintenance:

[0026] By analyzing image and sound data in real time, the system can promptly detect potential vibration and sound anomalies, such as the operating noise of construction machinery, and issue early warnings. This efficient early warning mechanism enables relevant departments to take timely measures to effectively prevent cable breakage accidents and ensure the safe and stable supply of power.

[0027] 5) Intelligent analysis and decision support:

[0028] This invention integrates advanced image processing technology and intelligent algorithms, enabling automatic identification, classification, and judgment of various situations without human intervention. This not only improves processing speed and accuracy but also significantly reduces the possibility of human error and false alarms. Simultaneously, the system can provide decision support for emergency response, further enhancing the efficiency and effectiveness of cable protection work.

[0029] 6) Cost-effectiveness and scalability:

[0030] Compared to traditional monitoring methods, this invention eliminates the need for additional physical sensors installed on or around cables, significantly reducing hardware costs and deployment complexity. Furthermore, because the system is based on video surveillance, it can be easily expanded to monitor a wider area, providing more comprehensive protection for the safe operation and maintenance of urban power distribution networks. Attached Figure Description

[0031] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0032] Figure 2 This is a schematic diagram showing the camera's field of view cable being unobstructed.

[0033] Figure 3 This is a diagram illustrating a situation where there are obstructions in the camera's field of view.

[0034] Figure 4 This is a schematic diagram of the hardware device of the present invention. Detailed Implementation

[0035] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0036] Before formally describing the present invention, a general description of the solution of the present invention will be given first to facilitate understanding.

[0037] Please refer to Figure 1 , Figure 1 This is a flowchart illustrating the method of the present invention;

[0038] This invention provides a method for early warning of cable damage from external sources, comprising the following steps:

[0039] S1: Deploy cameras within a preset range of the cable to be monitored; the cameras communicate with the cloud server;

[0040] It should be noted that the camera deployment of this invention takes into account two scenarios: one is that there are obstructions within the preset range of the cable to be monitored, and the other is that there are no obstructions within the preset range of the cable to be monitored.

[0041] Please refer to the following: Figure 2-3 ; Figure 2 This is a schematic diagram showing the camera's field of view cable being unobstructed. Figure 3 This is a diagram illustrating a situation where there are obstructions in the field of view of a camera.

[0042] Depending on whether there are obstructions, this invention employs two different monitoring methods. In the case of obstructions, a comprehensive judgment is made using vibration and sound characteristics. In the case of no obstructions, a target detection model is used to detect external targets for judgment. These will be explained in detail below.

[0043] S2: Use a camera to acquire surveillance images;

[0044] It should be noted that this invention continuously captures video data of key areas of cable lines using high-definition video surveillance equipment, forming a time-series set of video frames.

[0045] {I(t)|t=1,2,...,T}(1)

[0046] Where I(t) represents the video frame captured at time t.

[0047] S3: Build an object detection model on the cloud server and pre-train it to obtain a pre-trained object detection model;

[0048] It should be noted that the object detection models include: R-CNN model and YOLO model.

[0049] As an example, the present invention will be described using the YOLO model.

[0050] The YOLO (You Only Look Once) algorithm performs exceptionally well in object detection tasks, particularly in identifying objects that may pose a threat to cables (such as pile drivers, excavators, cranes, hoists, forklifts, etc., which may accidentally bump into or dig up cables).

[0051] S4: Input the monitored image into the trained target detection model to obtain the target detection result;

[0052] Step S4 is as follows:

[0053] S41. After scaling and preprocessing the monitoring image, input it into the trained target detection model.

[0054] As one example, the original input image I(t) needs to be scaled to fit the model's input requirements. The scaled image I′(t) is typically 416×416 pixels. The scaling formula is as follows:

[0055] I′(t)=ResizeGunction(I(t),TargetSize)(2)

[0056] TargetSize is the target size, for example, 416×416 pixels.

[0057] S42, The image is divided into multiple grids;

[0058] As one example, YOLO uses a unified convolutional neural network to simultaneously predict multiple bounding boxes and class probabilities. The image I′(t) is divided into S×S grids (e.g., 13×13), and each grid is responsible for detecting objects whose centers fall within that grid.

[0059] S43. For each grid cell, the trained object detection model predicts its object bounding box and class probability.

[0060] ① As one embodiment, each grid predicts B bounding boxes, and each bounding box contains the following parameters:

[0061] The probability that an object exists within the b-th bounding box of the grid (i,j).

[0062] Normalized offset of the bounding box center.

[0063] in:

[0064] S is the size of each grid cell (assuming the grid is square), (x 预测框中心， y 预测框中心 ) is the coordinate of the center of the predicted bounding box, while (x) is the coordinate of the center of the predicted bounding box网格左上角， y 网格左上角 ) is the coordinate of the top left corner of the current grid.

[0065] The relative proportions of the width and height of the bounding box. Where:

[0066]

[0067] w 预测框 and h 预测框 These are the width and height of the prediction box, respectively, and S is the size of each grid cell (assuming the grid is square).

[0068] For each grid cell, the probabilities of C categories are also predicted:

[0069]

[0070] p ij It is a vector whose elements are from arrive Each element represents the probability or score of the corresponding category C at position (i,j).

[0071] The raw data output by the object detection model needs to be decoded into specific bounding box coordinates and class probabilities.

[0072] The decoding formula is as follows:

[0073]

[0074] σ is the Sigmoid function, (c x ,c y ) is the coordinate of the top left corner of the grid, (p w ,p h ) is a predefined anchor frame size.

[0075]

[0076] ① Loss function: Taking into account coordinate error, confidence error, and classification error, the loss function can be expressed as:

[0077] The loss function is a common multi-task loss function in object detection tasks. It usually includes localization loss, confidence loss, and classification loss.

[0078] Loss=λ coord ∑L loc +ΣL conf +λ cls ∑L cls (9)

[0079] in,

[0080] ∑L loc : is the sum of all positioning losses.

[0081] ∑L conf It is the sum of all confidence losses.

[0082] ∑L cls : is the sum of all classification losses.

[0083] λ coord and λ cls These are hyperparameters used to balance the losses for different tasks, so that different types of losses are given different weights during training.

[0084] Furthermore, the specific form of each loss term can be clarified. For example:

[0085] L loc The mean squared error (MSE) or other forms of regression loss function are used to measure the difference between the predicted bounding box and the true bounding box.

[0086] L conf The commonly used method is the binary cross-entropy loss, which measures the difference between the predicted target confidence and the true value.

[0087] L cls The method likely used is Categorical Cross Entropy Loss, which measures the difference between the predicted target class and the true class.

[0088] Therefore, the entire loss function can be viewed as a weighted sum of these three parts of the loss, by adjusting λ. coord and λ cls To optimize the overall performance of the model.

[0089] S44. Non-maximum suppression of overlapping prediction boxes is used to obtain the target detection results.

[0090] As one example, Non-maximum Suppression (NMS) is used to remove overlapping predicted boxes, retaining only the most likely correct detections. The NMS process is as follows:

[0091] Sort the predicted boxes in descending order of confidence level. Calculate the intersection-union ratio (IUU) for each pair of predicted boxes:

[0092]

[0093] AIB represents the intersection of set A and set B. In object detection, this typically corresponds to the overlapping region of two predicted boxes (or bounding boxes). This intersection, as the numerator of the formula, represents the region commonly covered by the two predicted boxes.

[0094] AΥB represents the union of sets A and B. In object detection, it represents the total area occupied by the two predicted boxes, including overlapping and non-overlapping portions. This union serves as the denominator in the formula to normalize the intersection, resulting in a value between 0 and 1 that measures the degree of overlap between the two predicted boxes.

[0095] Therefore, the IoU formula measures the degree of overlap between two predicted boxes; a value closer to 1 indicates greater overlap, and a value closer to 0 indicates less overlap. In object detection tasks, IoU is often used to evaluate the similarity between predicted and ground truth boxes, and to remove overlapping predicted boxes in post-processing steps using non-maximum suppression (NMS). If the IoU between two boxes exceeds a set threshold, the box with higher confidence is retained, and the other is removed.

[0096] S5: If the target detection result belongs to the preset target, then issue an alert to the terminal through the cloud server; otherwise, proceed to step S6.

[0097] It should be noted that if the target detection result is a threatening object such as a pile driver, excavator, crane, hoist, or forklift that may accidentally hit or dig up the cable, an early warning will be issued.

[0098] When the cable monitored by the camera is obstructed, this invention provides early warning detection by combining vibration and sound characteristics. Specifically:

[0099] S6: Acquire sound signals from the camera and extract vibration signals from adjacent frames of the monitoring image;

[0100] It should be noted that step S6 is as follows:

[0101] S61. Extract edges from adjacent frames of the monitoring images;

[0102] As an example, edge detection is an important step in image processing, used to detect areas of significant brightness variation in an image, i.e., edges. The Canny edge detection algorithm is a commonly used edge detection method, and its main steps are as follows:

[0103] Gaussian filtering: A Gaussian filter is used to smooth images and reduce noise. The kernel function of a Gaussian filter is:

[0104]

[0105] Where σ is the standard deviation of the Gaussian kernel, which controls the degree of filtering.

[0106] Gradient calculation: The first or second derivative of the image gray level is calculated using the Sobel operator or other differential operators to determine the intensity and direction of the edge.

[0107] Sobel operator:

[0108]

[0109] This operator contains two 3×3 matrices, one horizontal and one vertical. Convolving the operator with the image plane yields the horizontal and vertical brightness difference values, respectively. If A represents the original image, G... x G y The image grayscale values ​​representing the horizontal and vertical edges, respectively, are given by the following formula:

[0110]

[0111] It is worth noting that while a typical convolution operation involves flipping the convolution kernel, Sobel does not require flipping because rotating it 180° around the center point does not affect the result.

[0112] S62. Perform differential processing on adjacent frame images after edge extraction to obtain the differential result;

[0113] It should be noted that after extracting the edges of each frame, a difference operation is performed on the edge images of adjacent frames. This is achieved by calculating the difference in corresponding pixel values, thereby obtaining the edge changes between adjacent frames.

[0114] S63. Extract vibration-related information based on the difference results;

[0115] It should be noted that the non-zero value regions in the difference image represent edge movement or change, which may be related to vibration. Further analysis of these difference images, such as calculating the magnitude and direction of the change, can extract vibration-related information.

[0116] S64. Extract vibration-related features to obtain vibration features;

[0117] Various features, such as amplitude, frequency, and waveform, can be extracted from the processed vibration signal to enable a deeper analysis and understanding of the vibration.

[0118] S65. Acquire sound signals through a camera and extract sound features.

[0119] It should be noted that MFCC (Mel-Frequency Cepstral Coefficients) is a widely used feature extraction method in sound recognition technology. It simulates the auditory characteristics of the human ear, converting sound signals into feature vectors through a series of mathematical transformations (such as Fast Fourier Transform, Mel-filter, logarithmic transform, and discrete cosine transform). These feature vectors can then be used for sound classification and recognition.

[0120] S7: Based on the vibration and sound signals, a pre-trained machine learning model is used to generate early warning results.

[0121] Step S7 is as follows:

[0122] S71. Pre-construct vibration signal dataset and sound feature dataset;

[0123] S72. Train a vibration detection model using a vibration signal dataset and train a sound detection model using a sound feature dataset.

[0124] S73. Input the vibration features and sound features obtained in steps S64 to S65 into the corresponding vibration detection model and sound detection model respectively to obtain the vibration detection result and sound detection result.

[0125] S74. If the vibration detection results and the sound detection results can match each other, an early warning will be sent to the terminal through the cloud server.

[0126] It should be noted that the results of image detection (such as edge extraction), vibration signal (extracted by optical flow method), and sound matching (through methods such as MFCC) can be used for intelligent analysis using logical judgment rules or machine learning-based classifiers (such as support vector machines, neural networks, etc.).

[0127] This invention integrates this information to determine the existence of potential threats and triggers an early warning mechanism when a threat is detected. For example, in surveillance video, if abnormal edge activity, abnormal vibration signals, or a match with a specific sound pattern are detected, the system may automatically send an alarm for timely response.

[0128] Please see Figure 4 , Figure 4 This is a schematic diagram of the hardware device in operation according to an embodiment of the present invention. The hardware device specifically includes: a cable anti-external damage early warning device 401, a processor 402, and a storage medium 403.

[0129] A cable external damage prevention early warning device 401: The cable external damage prevention early warning device 401 implements the cable external damage prevention early warning method.

[0130] Processor 402: The processor 402 loads and executes the instructions and data in the storage medium 403 to implement the cable anti-external damage early warning method.

[0131] Storage medium 403: The storage medium 403 stores instructions and data; the storage medium 403 is used to implement the cable anti-external damage early warning method.

[0132] The beneficial effects of this invention are:

[0133] 1) Non-contact and non-destructive monitoring:

[0134] The system of this invention does not require direct contact with cables or related equipment, thereby avoiding potential physical damage to the cables or equipment. This non-contact monitoring method not only ensures the integrity of cables and equipment but also significantly reduces installation and maintenance costs, and improves the reliability and service life of the system.

[0135] 2) Comprehensive monitoring integrating visual and acoustic technologies:

[0136] By integrating video image processing and sound feature extraction technologies, this invention achieves dual visual and acoustic monitoring. Within the visible range, the system can directly observe through video images; in areas with limited or no line of sight, potential threats are identified through image analysis and sound feature extraction. This comprehensive monitoring approach greatly improves the comprehensiveness and accuracy of the early warning system.

[0137] 3) Strong environmental adaptability:

[0138] Even in the presence of obstructions or complex environments, such as tree branches or eaves blocking the line of sight, the system of this invention can still extract vibration and sound features through image processing technology, effectively overcoming the line-of-sight limitations of traditional monitoring systems. This strong environmental adaptability enables the invention to achieve excellent monitoring results in various practical application scenarios.

[0139] 4) Efficient early warning and preventative maintenance:

[0140] By analyzing image and sound data in real time, the system can promptly detect potential vibration and sound anomalies, such as the operating noise of construction machinery, and issue early warnings. This efficient early warning mechanism enables relevant departments to take timely measures to effectively prevent cable breakage accidents and ensure the safe and stable supply of power.

[0141] 5) Intelligent analysis and decision support:

[0142] This invention integrates advanced image processing technology and intelligent algorithms, enabling automatic identification, classification, and judgment of various situations without human intervention. This not only improves processing speed and accuracy but also significantly reduces the possibility of human error and false alarms. Simultaneously, the system can provide decision support for emergency response, further enhancing the efficiency and effectiveness of cable protection work.

[0143] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for early warning of cable damage from external sources, characterized in that: The method includes the following steps: S1: Deploy cameras within a preset range of the cable to be monitored; the cameras communicate with the cloud server; S2: Use a camera to acquire surveillance images; S3: Build an object detection model on the cloud server and pre-train it to obtain a pre-trained object detection model; S4: Input the monitored image into the trained target detection model to obtain the target detection result; S5: If the target detection result belongs to the preset target, an alert will be sent to the terminal through the cloud server; Otherwise, proceed to step S6; S6: Acquire sound signals from the camera and extract vibration signals from adjacent frames of the monitoring image; S7: Based on the vibration and sound signals, a pre-trained machine learning model is used to generate early warning results; The target detection models include: R-CNN model and YOLO model; Step S4 is as follows: S41. After scaling and preprocessing the monitoring image, input it into the trained target detection model. S42, The image is divided into multiple grids; S43. For each grid cell, the trained object detection model predicts its object bounding box and class probability. S44. Non-maximum suppression of overlapping prediction boxes is used to obtain the target detection results; Step S6 is as follows: S61. Extract edges from adjacent frames of the monitoring images; S62. Perform differential processing on adjacent frame images after edge extraction to obtain the differential result; S63. Extract vibration-related information based on the difference results; S64. Extract vibration-related features to obtain vibration features; S65. Acquire sound signals through a camera and extract sound features; In step S64, the vibration characteristics include: amplitude and frequency; In step S65, the MFCC method is used to extract sound features; Step S7 is as follows: S71. Pre-construct vibration signal dataset and sound feature dataset; S72. Train a vibration detection model using a vibration signal dataset and train a sound detection model using a sound feature dataset. S73. Input the vibration features and sound features obtained in steps S64-S65 into the corresponding vibration detection model and sound detection model respectively to obtain the vibration detection result and sound detection result. S74. If the vibration detection results and the sound detection results can match each other, an early warning will be sent to the terminal through the cloud server.

2. A storage medium, characterized in that: The storage medium stores instructions and data to implement the cable damage prevention early warning method as described in claim 1.

3. A cable damage prevention early warning device, characterized in that: include: A processor and a storage medium; the processor loads and executes instructions and data in the storage medium to implement the cable damage prevention early warning method according to claim 1.

Images