Power transmission line surrounding hidden danger dynamic detection system based on image difference identification

Abstract

The invention provides a power transmission line surrounding hidden danger dynamic detection system based on image difference recognition, and relates to the technical field of data processing, and the system comprises a data collection module which is used for collecting video and image data around a power transmission line, and the data comes from video monitoring equipment arranged around the power transmission line; the image preprocessing module is used for preprocessing the collected video and image data to obtain preprocessed video and image data; the image difference recognition module is used for carrying out difference recognition on the preprocessed images by utilizing a machine learning algorithm, and detecting an abnormal value of the surrounding environment of the power transmission line by comparing continuous frames of images; adjusting the abnormal value of the surrounding environment of the power transmission line according to the dynamic factor to obtain a correction index; and the hidden danger analysis module is used for analyzing the correction index and determining the type, the position and the severity of the hidden danger. According to the invention, the efficiency and accuracy of power transmission line surrounding hidden danger detection are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and particularly to a dynamic detection system for hidden dangers around transmission lines based on image difference recognition. Background Art

[0002] With the rapid development of the power industry, the safe operation of transmission lines is crucial for ensuring the stable supply of the entire power system. However, since transmission lines usually pass through complex geographical environments, changes in their surrounding environments may pose potential threats to line safety. Therefore, timely detection and identification of hidden dangers around transmission lines are of great significance for preventing accidents and ensuring the continuity of power supply.

[0003] Although some existing hidden danger detections around transmission lines can automatically identify some hidden dangers, some lack consideration of the dynamic changes between consecutive frame images, resulting in missed or misdetected of some instantaneous or progressive hidden dangers. At the same time, the robustness of these systems in dealing with complex backgrounds and changing lighting conditions also needs to be improved. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a dynamic detection system for hidden dangers around transmission lines based on image difference recognition, which improves the efficiency and accuracy of hidden danger detection around transmission lines.

[0005] To solve the above technical problem, the technical solution of the present invention is as follows:

[0006] In a first aspect, a dynamic detection system for hidden dangers around transmission lines based on image difference recognition includes:

[0007] A data acquisition module for collecting video and image data around the transmission line, and the data is from video monitoring devices arranged around the transmission line;

[0008] An image preprocessing module for preprocessing the collected video and image data to obtain preprocessed video and image data;

[0009] An image difference recognition module for using machine learning algorithms to perform difference recognition on the preprocessed images, detecting abnormal values in the surrounding environment of the transmission line by comparing consecutive frame images; adjusting the abnormal values in the surrounding environment of the transmission line according to dynamic factors to obtain a correction index;

[0010] A hidden danger analysis module for analyzing the correction index to determine the type, location and severity of the hidden danger;

[0011] An early warning prompt module for generating early warning information according to the type, location and severity of the hidden danger, and real-time prompting operation and maintenance personnel through a user interface.

[0012] Furthermore, video and image data around the transmission line are collected. The data is sourced from video monitoring devices arranged around the transmission line, including:

[0013] In the arrangement space of video monitoring devices around the transmission line, a certain number of grey wolf individuals are randomly initialized, i.e., candidate positions of video monitoring devices. Each grey wolf individual represents a solution, including the position, orientation, and focal length parameters of the video monitoring device.

[0014] Define the objective function, which is used to evaluate the quality of grey wolf individuals, that is, the quality of data collected by video monitoring devices under specific positions and parameter configurations.

[0015] Grey wolf individuals update their positions according to the current final solution, simulating the hunting behavior of grey wolves, searching for corresponding solutions in the solution space. Grey wolf individuals gradually approach the current corresponding solution. When a grey wolf individual approaches the final solution, a search is executed to find the global final solution. According to the output result of the grey wolf optimization algorithm, the position and parameter configuration of the video monitoring device are adjusted.

[0016] Under the updated video monitoring device configuration, actual video and image data collection is carried out to obtain video and image data around the transmission line.

[0017] Furthermore, the collected video and image data are preprocessed to obtain preprocessed video and image data, including:

[0018] In the space of preprocessing parameters, a certain number of particles are randomly initialized, i.e., candidate preprocessing parameter combinations. Each particle represents a set of preprocessing parameters, including the type, size, and threshold of the filter.

[0019] Define the fitness function, which is used to evaluate the quality of preprocessed video and image data.

[0020] According to the individual historical final position and the group historical final position, the velocity of each particle is updated. According to the updated velocity, the position of each particle, i.e., the preprocessing parameters, is adjusted.

[0021] Perform preprocessing operations on each new position and use the fitness function to evaluate the preprocessing effect. Repeat the steps of velocity update, position update, and evaluation update until the preset number of iterations is reached to obtain the preprocessing parameter combination corresponding to the group historical final position.

[0022] Use the preprocessing parameter combination corresponding to the final position to perform preprocessing operations on the collected video and image data to obtain preprocessed video and image data.

[0023] Furthermore, a machine learning algorithm is used to perform difference recognition on the preprocessed image. By comparing consecutive frames of images, outliers in the surrounding environment of the transmission line are detected, including:

[0024] Extract consecutive image frames from the video stream at a set frame rate;

[0025] Extract features from each frame of the image to obtain a feature vector representing the image content;

[0026] Determine the number of initial frames for establishing the Gaussian mixture model. Use the feature vectors of the initial frames as training data and input them into the Gaussian mixture model. Learn the distribution of the feature vectors through the Gaussian mixture model algorithm to establish a Gaussian mixture model representing the "normal" state;

[0027] Set the parameters of the Gaussian mixture model, including the number of Gaussian distributions and the type of covariance matrix;

[0028] For each new frame of the video stream, extract its feature vector. Input the feature vector of the new image into the established Gaussian mixture model. Calculate the Mahalanobis distance between the feature vector of the new image and each Gaussian distribution in the Gaussian mixture model to obtain its difference value from the Gaussian mixture model.

[0029] Furthermore, extract features from each frame of the image to obtain a feature vector representing the image content, including:

[0030] Receive the preprocessed single-frame image data as input and initialize the SIFT feature detection algorithm, including setting the number of layers and scale factor for scale space extreme value detection;

[0031] Construct the scale space of the image by performing Gaussian blur and downsampling on the input image at different scales to find stable feature points at different scales;

[0032] In the scale space, find local extreme points by comparing the pixel values of each point with those of adjacent scales and adjacent positions. The extreme points are considered candidate key points;

[0033] Screen and locate the candidate key points detected in the previous step. Assign a main direction to each key point by calculating the gradient direction and magnitude of the pixels around the key point;

[0034] In the neighborhood around the key point, statistically analyze the gradient information of the pixels according to the assigned main direction to generate a descriptor vector of a fixed length;

[0035] Organize the descriptor vectors of all key points into a feature vector set, and the feature vector set represents the content of the entire image.

[0036] Further, the distribution of feature vectors is learned through the Gaussian mixture model algorithm to establish a Gaussian mixture model representing the "normal" state, including:

[0037] Collect feature vector data representing the "normal" state, and determine the number of Gaussian distributions to be used, i.e., the number of components of the mixture model; initialize the parameters for each Gaussian distribution, including the mean vector, covariance matrix, and mixing coefficient, i.e., the weight of each Gaussian distribution in the mixture model;

[0038] Use the expectation-maximization algorithm to iteratively update the Gaussian mixture model parameters. According to the current Gaussian mixture model parameters, calculate the posterior probability, i.e., the responsibility degree, of each data point belonging to each Gaussian distribution;

[0039] Based on the calculated posterior probability, update the mean, covariance, and mixing coefficient of each Gaussian distribution, and iterate until the Gaussian mixture model parameters converge;

[0040] Use the silhouette coefficient to evaluate the fitting effect of the Gaussian mixture model to obtain an evaluation result;

[0041] According to the evaluation result, determine the corresponding Gaussian mixture model as the Gaussian mixture model representing the "normal" state.

[0042] Further, the calculation formula for the dynamic factor is:

[0043]

[0044] where A(t) represents the dynamic adjustment factor; α represents the adjustment coefficient; I i (t) and I i (t + 1) respectively represent the gray values of the i-th pixel in the image I(t) and the image I(t + 1); N is the total number of pixels in the image; β represents the adjustment coefficient; v(t) represents the wind speed at time t; κ is a constant representing the magnification factor of the wind speed on the image difference; γ represents the adjustment coefficient; b i (t) is the actual value of the i-th pixel in the image b(t); is the denoising estimate value of the i-th pixel in the image b(t); i represents the index value.

[0045] Further, adjust the outliers in the surrounding environment of the transmission line according to the dynamic factor to obtain a correction index, including:

[0046] Multiply the dynamic factor by the corresponding difference value to obtain the correction index.

[0047] In the second aspect, a method for dynamically detecting potential hazards around a transmission line based on image difference recognition, the method includes the following steps:

[0048] Collect video and image data around the transmission line, where the data is sourced from video monitoring devices arranged around the transmission line;

[0049] Preprocess the collected video and image data to obtain preprocessed video and image data;

[0050] Use machine learning algorithms to perform difference recognition on the preprocessed images. By comparing consecutive frames of images, detect outliers in the environment around the transmission line; Adjust the outliers in the environment around the transmission line according to dynamic factors to obtain a correction index;

[0051] Analyze the correction index to determine the type, location, and severity of the potential hazards;

[0052] Generate a warning message based on the type, location, and severity of the potential hazards, and prompt the operation and maintenance personnel in real time through the user interface.

[0053] In a third aspect, a computing device includes:

[0054] One or more processors;

[0055] A storage device for storing one or more programs, which when executed by the one or more processors cause the one or more processors to implement the method described above.

[0056] In a fourth aspect, a computer-readable storage medium stores a program that, when executed by a processor, implements the method described above.

[0057] The above solution of the present invention has at least the following beneficial effects:

[0058] By collecting data in real time through video monitoring devices arranged around the transmission line, this system can continuously monitor the environment around the line and detect potential hazards in a timely manner. This automated data collection method greatly improves the efficiency and real-time performance of detection, and reduces the frequency and cost of manual inspections.

[0059] The image difference recognition module uses machine learning algorithms to perform difference recognition on the preprocessed images, which can accurately detect minute changes in the environment around the transmission line, and effectively identify outliers through the comparison of consecutive frames. This technical means improves the accuracy and sensitivity of potential hazard recognition.

[0060] Adjust the detected outliers through dynamic factors to obtain a correction index, which helps to reduce false alarms and missed detections, making the system more intelligent and adaptive. The dynamic adjustment mechanism enables the system to cope with complex and changing environmental conditions and improves the robustness of the system.

[0061] The hidden danger analysis module analyzes the correction index, and can accurately determine the type, location and severity of the hidden danger, which provides detailed hidden danger information for the operation and maintenance personnel and helps them make a quick and accurate response. The early warning prompt module can generate early warning information according to the specific situation of the hidden danger and prompt the operation and maintenance personnel in real time through the user interface. This instant feedback mechanism greatly shortens the time from discovering the hidden danger to taking countermeasures and improves the safety and stability of the power system. Brief Description of the Drawings

[0062] Figure 1 is a schematic diagram of a dynamic hidden danger detection system for the periphery of a transmission line based on image difference recognition provided by an embodiment of the present invention.

[0063] Figure 2 is a schematic flowchart of a dynamic hidden danger detection method for the periphery of a transmission line based on image difference recognition provided by an embodiment of the present invention. Detailed Embodiments

[0064] Hereinafter, exemplary embodiments of the present disclosure will be described in more detail with reference to the drawings. Although the exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided so that the present disclosure can be more thoroughly understood and the scope of the present disclosure can be fully conveyed to those skilled in the art.

[0065] As Figure 1 shown, an embodiment of the present invention provides a dynamic hidden danger detection system for the periphery of a transmission line based on image difference recognition, including:

[0066] A data acquisition module for acquiring video and image data around the transmission line, and the data is from video monitoring devices arranged around the transmission line;

[0067] An image preprocessing module for preprocessing the acquired video and image data to obtain preprocessed video and image data;

[0068] An image difference recognition module for using a machine learning algorithm to perform difference recognition on the preprocessed image, detecting outliers in the environment around the transmission line by comparing consecutive frames of images; adjusting the outliers in the environment around the transmission line according to a dynamic factor to obtain a correction index;

[0069] A hidden danger analysis module for analyzing the correction index to determine the type, location and severity of the hidden danger;

[0070] An early warning prompt module for generating early warning information according to the type, location and severity of the hidden danger and prompting the operation and maintenance personnel in real time through the user interface.

[0071] In the embodiment of the present invention, by arranging the video monitoring equipment around the transmission line, the video and image data around the transmission line can be collected in real time and continuously, providing a rich data source for subsequent image processing and hidden danger detection, which is helpful to timely discover and deal with potential safety hazards and improve the safety and stability of the transmission line. The collected original video and image data are preprocessed, such as denoising, enhancement, scaling and other operations, which can improve the image quality and reduce the influence of environmental factors (such as illumination changes, shadows, etc.) on subsequent difference recognition. The preprocessed image data is clearer and more accurate, which helps to improve the accuracy and efficiency of hidden danger detection. The machine learning algorithm is used to perform difference recognition on the preprocessed image. By comparing the image changes of continuous frames, the abnormal values ​​of the surrounding environment of the transmission line can be sensitively captured. This dynamic detection method can quickly discover hidden dangers such as excessive growth of trees, illegal construction of buildings, and external force damage. At the same time, the abnormal values ​​are adjusted according to the dynamic factors to obtain the correction index, which can further improve the accuracy and reliability of detection. Through in-depth analysis of the correction index, the system can accurately determine the type of hidden danger (such as natural disasters, man-made damage, etc.), specific location and severity, which helps operation and maintenance personnel quickly understand the hidden danger status and formulate reasonable treatment plans, so as to eliminate safety hazards in time and prevent accidents. According to the results of the hidden danger analysis, the system can automatically generate early warning information and prompt the operation and maintenance personnel in real time through the user interface. This timely early warning mechanism can ensure that the operation and maintenance personnel are aware of the hidden danger situation at the first time and take countermeasures quickly, effectively improving the safety management level of the transmission line. At the same time, the detail and accuracy of the early warning information also help to improve the pertinence and efficiency of the operation and maintenance work.

[0072] In a preferred embodiment of the present invention, video and image data around the transmission line are collected, and the data comes from video monitoring equipment arranged around the transmission line, including:

[0073] In the layout space of video surveillance equipment around the power transmission line, a certain number of gray wolf individuals are randomly initialized, i.e., candidate video surveillance equipment locations; each gray wolf individual represents a solution, including the location, orientation, and focal length parameters of the video surveillance equipment, specifically including: determining the spatial range where the video surveillance equipment may be deployed, which is usually based on factors such as the direction of the power transmission line, the surrounding environment, and safety regulations. Within this spatial range, a certain number of gray wolf individuals are randomly generated, each gray wolf individual represents a candidate video surveillance equipment location and parameter configuration, and each gray wolf individual contains not only location information (such as longitude and latitude or three-dimensional coordinates), but also the orientation of the equipment (such as azimuth, pitch angle) and focal length parameters. These parameters will affect the field of view and monitoring effect of the monitoring equipment.

[0074] Define the objective function, which is used to evaluate the quality of gray wolf individuals, that is, the quality of data collected by video surveillance devices under specific positions and parameter configurations. The calculation formula of the objective function is as follows:

[0075]

[0076] Among them, F(x) represents the objective function, which is used to evaluate the quality of data collected by video surveillance devices under specific positions and parameter configurations; w 1 、w 2 and w 3 are weight coefficients, which are used to balance the importance of three factors: image clarity, monitoring range coverage rate, and weighted target visibility. These weights can be adjusted according to actual needs; represents image clarity, which is measured by calculating the sum of the gradients of each pixel point in the x and y directions in the image. The larger the gradient, the clearer the image edge, and thus the higher the image clarity; This part represents the coverage rate of the monitoring range. Among them, A is the overlapping area between the field of view of the monitoring device and the area to be monitored, and T is the total area to be monitored. The larger this ratio, the wider the coverage range of the monitoring device; This part represents the unweighted target visibility. Among them, P is the number of pixels of the target in the image, S is the total number of pixels in the image, C is the average contrast of the target, and M is the possible maximum contrast. The larger this value, the more obvious the target in the image; d is the distance between the monitoring device and the key target, μ is set as the center position of the key target (usually taken as 0), and σ is the standard deviation of the Gaussian function, which is used to control the speed of weight decay with distance. This formula makes the device configuration closer to the key target obtain a higher evaluation; ∑ i,j is a summation symbol, indicating the summation of all pixel points (i, j) in the image. Among them, i and j represent the abscissa and ordinate of the pixel point respectively; This is the gradient of image I at pixel point (i, j) in the x direction (i.e., the brightness change rate in the horizontal direction). The gradient reflects the change degree and direction of pixel values in the image and is one of the important indicators for evaluating image clarity. If the image has clear edges and rich details in the horizontal direction, then this gradient value will be relatively large; This is the gradient of image I at pixel point (i, j) in the y direction (i.e., the brightness change rate in the vertical direction). Similar to it reflects the clarity of the image in the vertical direction.

[0077] Grey wolf individuals update their positions according to the current final solution, simulating the hunting behavior of grey wolves to search for corresponding solutions in the solution space. Grey wolf individuals gradually approach the current corresponding solution. When a grey wolf individual approaches the final solution, a search is executed to find the global final solution. According to the output results of the grey wolf optimization algorithm, the positions and parameter configurations of video surveillance devices are adjusted, specifically including: taking the generated grey wolf individuals as the initial wolf pack, evaluating the fitness of each grey wolf individual according to the objective function, and selecting the individual with the highest fitness as the leading wolf (i.e., the current optimal solution); other grey wolf individuals update their positions based on the position of the leading wolf and their own positions, simulating the hunting behavior of grey wolves, and repeating the process of evaluating fitness and updating positions until the stopping condition is met (such as reaching the maximum number of iterations or the fitness improvement is less than a certain threshold). When a grey wolf individual approaches the final solution, a fine search is executed to find the global optimal solution; according to the output results of the grey wolf optimization algorithm (i.e., the position and parameter configuration of the optimal grey wolf individual), the position, orientation, and focal length of the actual video surveillance device are adjusted. After the configuration is adjusted, actual video surveillance tests are carried out to verify whether the effect of the new configuration meets the expectations.

[0078] Under the updated video surveillance device configuration, actual video and image data are collected to obtain video and image data around the transmission line, specifically including: under the updated and verified video surveillance device configuration, the device is started to collect actual video and image data, the collected video and image data are saved, and necessary preprocessing and analysis work are carried out for subsequent use or further analysis.

[0079] In the embodiments of the present invention, by randomly initializing a certain number of grey wolf individuals (i.e., candidate positions of video surveillance devices), the diversity and extensiveness of the search process can be ensured, thereby increasing the possibility of finding the global optimal solution. This initialization method helps to avoid the algorithm falling into a local optimal solution and improves the overall optimization effect. Each grey wolf individual is designed to include the position, orientation, and focal length parameters of the video surveillance device. Such a design can more comprehensively reflect the actual deployment situation of the video surveillance device, making the optimization process closer to the actual requirements. By considering multiple parameters simultaneously, the comprehensive optimization of the video surveillance device configuration can be achieved, and the quality of data collection can be improved. By defining an objective function to evaluate the quality of grey wolf individuals, that is, the quality of data collected by the video surveillance device under specific positions and parameter configurations, the setting of the objective function is directly related to the quality requirements of data collection, ensuring that the optimization process is consistent with the actual application requirements. By simulating the hunting behavior of grey wolves to update the positions of grey wolf individuals, this search strategy can effectively find better solutions in the solution space. The bionic principle of the grey wolf optimization algorithm makes the search process adaptive and intelligent, capable of improving the search efficiency and the probability of finding the global optimal solution; according to the output results of the grey wolf optimization algorithm, the positions and parameter configurations of the video surveillance device are adjusted, which ensures that the optimization results can be directly applied to the actual system, improving the performance of the system. By guiding the actual configuration adjustment with the algorithm, the precise optimization of the video surveillance system can be achieved, and the efficiency and accuracy of data collection can be improved. Conduct actual video and image data collection under the updated video surveillance device configuration, which can verify the effectiveness of the optimization algorithm and ensure that the collected data meets the application requirements. By feeding back the optimization effect through actual data collection, a closed-loop optimization process is formed, which helps to continuously improve and enhance the performance of the video surveillance system.

[0080] In a preferred embodiment of the present invention, the collected video and image data are preprocessed to obtain preprocessed video and image data, including:

[0081] In the space of preprocessing parameters, a certain number of particles are randomly initialized, that is, candidate combinations of preprocessing parameters. Each particle represents a set of preprocessing parameters, including the type, size, and threshold of the filter, specifically including: determining the range of preprocessing parameters, which includes the type of filter (such as Gaussian filter, median filter, etc.), the size of the filter (i.e., the size of the kernel), and the threshold (used for certain types of filters, such as edge detection filters); in the space of preprocessing parameters, a certain number of particles are randomly generated, and each particle represents a candidate combination of a set of preprocessing parameters. An initial position and velocity are assigned to each particle. Here, "position" refers to the specific values of the preprocessing parameters, and "velocity" is the direction and step size of parameter adjustment.

[0082] Define a fitness function to evaluate the quality of the preprocessed video and image data. The calculation formula of the fitness function is as follows:

[0083]

[0084] where μ X and μ Y are the means of the original image and the preprocessed image respectively; σ X and σ Y are the standard deviations of the original image and the preprocessed image respectively, and σ XY is the covariance of the original image and the preprocessed image; C 1 and C 2 are constants; G x (i,j) and G y (i,j) are the gradient values of the image in the x and y directions respectively; M×N is the total number of pixels of the image; σ n is the standard deviation of the noise in the original image (calculated without denoising); σ p is the standard deviation of the noise in the preprocessed image; w 1 、w 2 and w 3 are weight coefficients. The fitness function will guide the particles to move towards the optimal combination of preprocessing parameters to achieve the best data preprocessing effect.

[0085] Update the velocity of each particle according to the individual historical final position and the group historical final position. Adjust the position of each particle, that is, the preprocessing parameters, according to the updated velocity. Specifically, for each particle, record the position with the highest fitness function value in its history, that is, the individual historical best position. Among all the particles, find the position with the highest fitness function value, that is, the group historical best position. Update the velocity of each particle through the standard velocity update formula of PSO according to the individual historical best position, the group historical best position, and the current velocity of the particle. Adjust the position of each particle, that is, the value of the preprocessing parameters, according to the updated velocity.

[0086] Perform preprocessing operations on each new position and use the fitness function to evaluate the preprocessing effect. Repeat the steps of velocity update, position update, and evaluation update until the preset number of iterations is reached to obtain the combination of preprocessing parameters corresponding to the group historical final position. Specifically, perform preprocessing operations on each new position (i.e., the new combination of preprocessing parameters), such as filtering, denoising, etc., use the fitness function to evaluate the quality of the preprocessed video and image data, and obtain a fitness value; continuously repeat the steps of updating the particle velocity and position and evaluating the preprocessing effect until the preset number of iterations is reached or other stopping conditions are met. After the iteration ends, output the combination of preprocessing parameters corresponding to the group historical best position.

[0087] Preprocess the collected video and image data using the preprocessing parameter combination corresponding to the final position to obtain the preprocessed video and image data, specifically including: using the output optimal preprocessing parameter combination to preprocess the collected original video and image data, and after preprocessing, obtain video and image data with improved quality.

[0088] In the embodiment of the present invention, the particle swarm optimization algorithm is used to search for the best preprocessing parameter combination, including the type, size, and threshold of the filter, etc., which can more effectively remove noise in the image, enhance the edges and details of the image, thereby significantly improving the quality of video and image data. Since the particle swarm optimization algorithm is a population-based stochastic search algorithm, it can adaptively adjust the search strategy and does not depend on the specific form of the problem. Therefore, for different types of video and image data, as well as data collected in different environments, this algorithm can show good adaptability and robustness. Traditional preprocessing parameter settings often rely on experienced professionals for manual adjustment, which is not only time-consuming and laborious but also difficult to ensure the best processing effect every time. However, using the particle swarm optimization algorithm can automatically find the optimal preprocessing parameter combination, greatly reducing the need for manual intervention and improving the processing efficiency. By automatically adjusting the preprocessing parameters through the algorithm, it is possible to avoid image information loss or redundant calculations caused by inappropriate processing, thereby making more reasonable use of computing resources and achieving efficient video and image processing. The particle swarm optimization algorithm can be combined with other image processing technologies to form a more complex preprocessing process. At the same time, with the continuous iteration and optimization of the algorithm, there is still potential for further improvement in its processing effect.

[0089] In a preferred embodiment of the present invention, a machine learning algorithm is used to perform difference recognition on the preprocessed image. By comparing consecutive frames of images, outliers in the surrounding environment of the transmission line are detected, including:

[0090] Extract consecutive image frames from the video stream at a set frame rate, specifically including: use a video processing library (such as OpenCV) to open the video stream, set a frame rate, for example, extract 5 frames of images per second; traverse the video stream, extract image frames at the set frame rate interval, and save them as image files or directly process them in memory.

[0091] Extract features from each frame of the image to obtain a feature vector representing the image content;

[0092] Determine the number of initial frames for establishing the Gaussian mixture model, for example, the first 100 frames, use the feature vectors of the initial frames as training data, input them into the Gaussian mixture model, and learn the distribution of the feature vectors through the Gaussian mixture model algorithm to establish a Gaussian mixture model representing the "normal" state;

[0093] Set the parameters of the Gaussian mixture model, including the number of Gaussian distributions and the type of covariance matrix. Specifically, according to the application requirements and experimental results, set the number of Gaussian distributions in the Gaussian mixture model and select the type of covariance matrix, such as a diagonal matrix. These parameters will affect the complexity of the model and its ability to fit the data.

[0094] For each new image in the video stream, extract its feature vector and input the feature vector of the new image into the established Gaussian mixture model. By calculating the Mahalanobis distance between the feature vector of the new image and each Gaussian distribution in the Gaussian mixture model, obtain the difference value between it and the Gaussian mixture model. Specifically, a new frame of image is extracted from the video stream, and the feature vector representing the content of this image has been obtained through a feature extraction method (such as SIFT). This feature vector will be used as the basis for analyzing whether this frame of image conforms to the "normal" state. Next, input this feature vector into the previously trained Gaussian mixture model (GMM). GMM is a probability model composed of multiple Gaussian distributions, and each Gaussian distribution has its mean and covariance matrix. These parameters have been estimated through algorithms during the training phase. For the input feature vector, calculate its Mahalanobis distance from each Gaussian distribution in the GMM. The Mahalanobis distance is a distance metric that takes into account the covariance structure of the data. The Mahalanobis distance considers the correlation between dimensions. By calculating the Mahalanobis distance between the input feature vector and each Gaussian distribution, a set of distance values can be obtained. These distance values reflect the similarity degree between the input feature vector and each Gaussian distribution in the GMM. After obtaining the Mahalanobis distance, further process these distance values to obtain a comprehensive difference value. This difference value can be the average of all Mahalanobis distances. The size of the difference value reflects the deviation degree of the input feature vector from the "normal" state model. Finally, set a threshold to determine whether the input feature vector is abnormal. This threshold is usually determined through experiments or experience, and it represents the boundary between the "normal" state and the "abnormal" state. If the calculated difference value exceeds this threshold, then it can be considered that this image frame is abnormal because it has a significant difference from the "normal" state represented by the training data.

[0095] In the embodiments of the present invention, by setting the frame rate to continuously extract image frames from the video stream, the system can monitor the surrounding environment of the transmission line in real time. Once an anomaly is detected, an alarm can be immediately triggered or other countermeasures can be taken, thus greatly reducing potential safety risks. Feature extraction is performed on each frame of the image to obtain a feature vector representing the image content, which helps to compress the data and focus on the key information in the image. The use of the feature vector improves the efficiency and accuracy of the subsequent anomaly detection process. By learning the distribution of the feature vectors through the Gaussian Mixture Model (GMM), the system can establish a model representing the "normal" state. This model is adaptive and can be updated according to changes in the actual environment to maintain an accurate description of the "normal" state. The user can set the parameters of the Gaussian Mixture Model, such as the number of Gaussian distributions and the type of covariance matrix, to adjust the complexity and sensitivity of the model according to the specific application scenario. This flexibility and configurability enable the system to be widely applicable to different monitoring environments and requirements. By calculating the Mahalanobis distance between the new image feature vector and each Gaussian distribution in the Gaussian Mixture Model, the system can accurately identify the outliers that are significantly different from the "normal" state. This method is highly robust to interference factors such as illumination changes, shadows, and occlusions, and can reduce false alarms and missed detections. The difference values and anomaly detection results provided by the system can be used as data-driven decision support to help the operation and maintenance personnel more accurately judge the safety status of the transmission line and formulate targeted maintenance strategies.

[0096] In a preferred embodiment of the present invention, feature extraction is performed on each frame of the image to obtain a feature vector representing the image content, including:

[0097] Receiving the preprocessed single-frame image data as input and initializing the SIFT feature detection algorithm, including setting the number of layers and scale factor for scale space extreme value detection, specifically including: receiving the preprocessed single-frame image as the input of the algorithm. The preprocessing includes steps such as grayscale conversion, noise reduction, and contrast enhancement to improve the accuracy of feature extraction; initializing the SIFT algorithm and setting the algorithm parameters. Among them, the important parameters include the number of layers and scale factor for scale space extreme value detection. The number of layers determines the depth of the scale space, while the scale factor controls the scale change between adjacent scale layers.

[0098] By performing Gaussian blur and downsampling on the input image at different scales, the scale space of the image is constructed to find stable feature points at different scales, specifically including: performing Gaussian blur on the input image at different scales. The purpose of Gaussian blur is to simulate the performance of the image at different scales, which is achieved by convolving the image with Gaussian kernels of different scales. Downsampling the blurred image to generate a series of images at different scales. These images constitute the scale space of the image, where each scale layer represents the information of the image at different resolutions.

[0099] In the scale space, local extreme points are found by comparing the pixel values of each point with those of adjacent scales and adjacent positions. These extreme points are considered candidate key points, specifically including: traversing each point in the scale space and comparing it with points of adjacent scales and adjacent positions, which typically involves comparing the pixel values of the current point with those of its upper and lower scales and the surrounding 8 points. If the current point is an extreme value (maximum or minimum) in all these comparisons, it is marked as a local extreme point. These extreme points are regarded as candidate key points because they exhibit significant features at different scales.

[0100] The candidate key points detected in the previous step are filtered and localized. By calculating the gradient direction and magnitude of the pixels around the key points, a main direction is assigned to each key point, specifically including: further filtering the candidate key points detected in the previous step to remove points with low contrast and unstable edge response points, which is usually achieved by calculating the Hessian matrix (or a similar method) at the key points to evaluate their stability and significance. Precise localization of the filtered key points is carried out, which includes adjusting the position and scale of the key points to make them more accurately correspond to the actual features in the image. By calculating the gradient direction and magnitude of the pixels around the key points, a main direction is assigned to each key point. This main direction is used to ensure the rotational invariance of the SIFT features.

[0101] Within the neighborhood around the key points, according to the assigned main direction, the gradient information of the pixels is statistically analyzed to generate a descriptor vector of a fixed length, specifically including: within the neighborhood around the key points, the gradient information of the pixels is statistically analyzed according to the assigned main direction, which typically involves dividing the neighborhood into multiple sub-regions and calculating the statistical information (such as histograms) of the gradient direction and magnitude within each sub-region. These statistical information are integrated into a descriptor vector of a fixed length. This descriptor vector captures the local image structure information around the key points and has a certain stability against image rotation, scale change, and illumination change.

[0102] The descriptor vectors of all key points are organized into a set of feature vectors, which represents the content of the entire image, specifically including: organizing the descriptor vectors of all key points into a set of feature vectors. This set represents the content of the entire image and can be used for subsequent tasks such as image matching and recognition.

[0103] In the embodiments of the present invention, SIFT features have good stability against scale changes and rotations of images. This means that even if the objects in the image appear in different sizes or orientations due to changes in shooting distance or angle, SIFT can still accurately identify and extract the features of these objects, which greatly enhances the robustness and applicability of the feature extraction method. SIFT focuses on local features of images rather than global features. This enables it to accurately identify the objects or regions of interest in complex backgrounds, even if these objects or regions are partially occluded or overlapped with other objects. In addition, local features are more conducive to subsequent image matching and recognition tasks. By searching for stable feature points at different scales and through a screening and localization process, SIFT can extract the most prominent and stable features in the image. These features not only have a moderate quantity but also high quality, and can effectively represent the content of the image, which improves the stability and reliability of the feature vectors. SIFT generates a descriptor vector with a fixed length for each key point. This fixed-length representation method is not only convenient for storage and management but also conducive to the processing of subsequent machine learning algorithms. For example, when training a classifier or constructing an image retrieval system, these fixed-length feature vectors can be directly used as inputs. Due to the above advantages of SIFT features, they are widely used in various computer vision tasks such as image matching, object tracking, and 3D reconstruction. Therefore, using SIFT for feature extraction can provide rich choices and flexibility for subsequent application development.

[0104] In a preferred embodiment of the present invention, the distribution of feature vectors is learned through the Gaussian mixture model algorithm to establish a Gaussian mixture model representing the "normal" state, including:

[0105] Collect feature vector data representing the "normal" state and determine the number of Gaussian distributions to be used, i.e., the number of components of the mixture model; initialize the parameters for each Gaussian distribution, including the mean vector, covariance matrix, and mixing coefficient, i.e., the weight of each Gaussian distribution in the mixture model. Specifically, it includes: collecting a large amount of data generated in the "normal" state from actual application scenarios (such as surveillance videos, sensor data, etc.), preprocessing these data, including cleaning, standardization, etc., to eliminate the influence of noise and outliers, and extracting the feature vectors of the data. These feature vectors should be able to effectively represent the essential attributes of the data; according to the complexity of the data and actual requirements, select an appropriate number of Gaussian distributions (i.e., the number of components of the mixture model), and initialize the parameters of each Gaussian distribution, including the mean vector, covariance matrix, and mixing coefficient. These parameters can be initialized randomly or based on the statistical characteristics of the data.

[0106] Iteratively update the Gaussian mixture model parameters using the Expectation-Maximization algorithm. According to the current Gaussian mixture model parameters, calculate the posterior probability (responsibility) of each data point belonging to each Gaussian distribution. Based on the calculated posterior probabilities, update the mean, covariance, and mixing coefficients of each Gaussian distribution. Iterate until the Gaussian mixture model parameters converge. Specifically, it includes: According to the current Gaussian mixture model parameters, calculate the posterior probability (responsibility) of each data point belonging to each Gaussian distribution, which involves using the probability density function of the Gaussian distribution to calculate the likelihood of the data point under each distribution and weighting according to the mixing coefficients; Based on the calculated posterior probabilities, update the mean, covariance, and mixing coefficients of each Gaussian distribution. These updates are done by maximizing the expected likelihood function of the data, where the updates of the mean and covariance involve weighted average and weighted covariance calculations, and the update of the mixing coefficient is achieved by normalizing the sum of the posterior probabilities. Repeat the above steps until the parameters of the Gaussian mixture model converge. The convergence can be judged by monitoring the magnitude of the parameter changes or setting the maximum number of iterations.

[0107] Use the silhouette coefficient to evaluate the fitting effect of the Gaussian mixture model to obtain the evaluation result. Specifically, it includes: Use the Silhouette Coefficient to quantify the fitting effect of the Gaussian mixture model. The silhouette coefficient measures the clustering quality of the model by comparing the distance of a data point to other points within its own class and to points in other classes. Calculate the silhouette coefficient and evaluate the performance of the model based on its value. A higher silhouette coefficient value usually means a better fitting effect.

[0108] According to the evaluation result, determine the corresponding Gaussian mixture model as the Gaussian mixture model representing the "normal" state. Specifically, it includes: According to the evaluation result, select the Gaussian mixture model with the best performance as the model representing the "normal" state. This can be done by comparing the silhouette coefficients of different models (with different numbers of Gaussian distributions or initialization parameters). Save the selected Gaussian mixture model for subsequent tasks such as anomaly detection and data generation.

[0109] In the embodiments of the present invention, the Gaussian mixture model can flexibly fit complex data distributions. Since it uses the weighted sum of multiple Gaussian distributions to represent data, it can capture the multimodal characteristics that cannot be described by a single Gaussian distribution, which enables the model to more accurately represent the true distribution of "normal" state data. By establishing a Gaussian mixture model representing the "normal" state, it is easy to detect abnormal data that does not conform to the normal mode, which is crucial for many application scenarios (such as fault detection, security monitoring, etc.), because timely identification of anomalies helps prevent potential problems or risks. The Expectation-Maximization (EM) algorithm is used to iteratively update the parameters of the Gaussian mixture model. This is a robust parameter estimation method. Even in the presence of noise or missing data, the EM algorithm can effectively estimate the model parameters, thus ensuring the robustness of the model. Evaluation metrics such as the silhouette coefficient are used to quantify the fitting effect of the Gaussian mixture model, which provides an objective basis for model selection and optimization. By comparing the silhouette coefficients of different models, the model that best represents the "normal" state can be selected, thereby improving the accuracy of subsequent analysis. Each Gaussian distribution in the Gaussian mixture model can correspond to a specific group or pattern in the data, which enhances the interpretability of the model. By examining the parameters (such as the mean and covariance) of each Gaussian distribution, different characteristics and patterns of the "normal" state can be deeply understood.

[0110] In a preferred embodiment of the present invention, the calculation formula of the dynamic factor is:

[0111]

[0112] where, A(t) represents the dynamic adjustment factor; α represents the adjustment coefficient; I i (t) and I i (t + 1) respectively represent the gray values of the i-th pixel in the image I(t) and the image I(t + 1); N is the total number of pixels in the image; β represents the adjustment coefficient; v(t) represents the wind speed at time t; κ is a constant representing the magnification factor of the wind speed on the image difference; γ represents the adjustment coefficient; b i (t) is the actual value of the i-th pixel in the image b(t); is the denoising estimate value of the i-th pixel in the image b(t); i represents the index value.

[0113] In the embodiments of the present invention, A(t) can be dynamically adjusted according to actual situations, which means it can remain effective under different environments and conditions. For example, in a scenario with large wind speed variations, A(t) can be adjusted accordingly to reflect such changes, thereby improving the adaptability and robustness of the overall system. By calculating the changes in pixel gray values between adjacent image frames, A(t) can accurately capture the dynamic information in the image, which is crucial for application scenarios that require real-time monitoring of image changes (such as video surveillance, motion analysis, etc.) because it provides a direct and quantitative metric for the changes in image content. Incorporating wind speed data into the calculation makes A(t) not only dependent on image information but also consider environmental factors. The fusion of such multi-source data enhances the representation ability of A(t), especially in scenarios where environmental factors have a significant impact on image quality (such as outdoor surveillance, remote sensing imaging, etc.). By introducing the difference term between the actual value and the denoising estimated value of the pixels in the image, A(t) can suppress the influence of noise to a certain extent, which helps improve the stability and accuracy of the system in a noisy environment. Especially in image processing and analysis tasks, noise is usually a non-negligible problem.

[0114] In a preferred embodiment of the present invention, adjusting the outliers in the surrounding environment of the transmission line according to the dynamic factor to obtain a correction index includes:

[0115] Multiplying the dynamic factor by the corresponding difference value to obtain the correction index.

[0116] In the embodiments of the present invention, by using the dynamic factor to adjust the outliers, the actual situation of the surrounding environment of the transmission line can be more accurately reflected. The dynamic factor takes into account various factors, including image changes, wind speed, etc., so that it can more precisely correct the outliers in the original data, making the corrected index closer to the real situation. In the monitoring of transmission lines, changes in the surrounding environment may cause data fluctuations, which in turn affect the accurate judgment of the line state. By introducing the dynamic factor for outlier adjustment, such fluctuations can be reduced, the stability of the monitoring data can be enhanced, and the reliability of the system for evaluating the state of the transmission line can be improved. Since the dynamic factor can reflect the changes in environmental factors in real time (such as wind speed, image differences, etc.), the outlier adjustment method based on the dynamic factor can respond more promptly to the changes in the surrounding environment. The corrected index can provide more accurate data support for the operation and maintenance management of the transmission line. By dynamically adjusting the outliers, the operation and maintenance personnel can make decisions based on more real data, thereby optimizing resource allocation and improving operation and maintenance efficiency. Introducing the dynamic factor for outlier adjustment is part of an intelligent monitoring system. The application of this method helps to improve the intelligent level of the entire monitoring system, enabling it to more autonomously adapt to environmental changes and improve the accuracy of early warning and response.

[0117] In a preferred embodiment of the present invention, the correction index is analyzed to determine the type, location, and severity of potential hazards, including:

[0118] The potential hazard analysis module first receives the correction index. Thresholds corresponding to different potential hazard types are preset in advance. These thresholds are obtained based on historical data and are used to distinguish normal changes from potential hazards. Each received correction index is compared one by one with the threshold of its corresponding potential hazard type. If a certain correction index exceeds the threshold of its corresponding potential hazard type, the index is marked as abnormal, and the corresponding potential hazard type is recorded. The image difference recognition module has output an image frame containing the abnormal area or the coordinate information of the abnormal area. If an image frame is output, image processing techniques (such as edge detection, color segmentation, etc.) are used to automatically identify and locate the abnormal area and extract its coordinate information. If the coordinate information of the abnormal area is directly output, this coordinate information is directly used.

[0119] Determine the geographical location (such as longitude and latitude) of the video surveillance device and its shooting range. The shooting range may be affected by various factors such as the camera direction, focal length, and lens type. Based on this information, a mapping table can be created to map the pixel coordinates (x, y) in the image to the geographical coordinates (longitude, latitude) in the real world. For example, by placing reference points with known geographical locations within the camera's field of view, the mapping relationship can be calibrated and adjusted. After establishing the mapping relationship, it is used to convert the pixel coordinates of the abnormal area detected in the image into actual geographical coordinates. This usually involves inserting the pixel coordinates into the mapping function or looking up the mapping table to output the corresponding longitude and latitude coordinates. Combining the converted geographical coordinates with the potential hazard type determined previously by the correction index, the specific location of the potential hazard in the real world can now be determined. This location is represented as a point. Next, the correction index is used to evaluate the severity of the potential hazard. The correction index is a quantitative indicator that reflects the degree of abnormality detected by the image difference recognition module. Generally, the higher the correction index, the more obvious the abnormality and the more serious the potential hazard may be. Different potential hazard levels (such as low, medium, high) can be set according to the size of the correction index to trigger different levels of responses or alarms.

[0120] In a preferred embodiment of the present invention, according to the type, location, and severity of the potential hazard, a warning message is generated and the operation and maintenance personnel are prompted in real time through the user interface, including:

[0121] Obtain information such as the potential hazard type, location, and severity determined in the previous steps. These information are the basis for generating the warning message to ensure the accuracy and specificity of the warning content. According to the summarized potential hazard information, the system automatically generates a warning message. The warning message should contain the following key elements:

[0122] Hazard type, indicating which type the detected hazard belongs to, such as tree fall, foreign object intrusion, fire, etc.

[0123] Location description, providing the specific location of the hazard, which may include latitude and longitude coordinates, relative location description (such as "500 meters away from a certain substation") or marked points on the map.

[0124] Severity, indicating the severity of the hazard, which can use qualitative descriptions (such as "minor", "severe") or quantitative indicators (such as specific values of the correction index).

[0125] Suggested measures, based on the hazard type and severity, providing preliminary suggested handling measures to guide the operation and maintenance personnel to respond quickly.

[0126] The format of the warning information can be adjusted according to actual needs to ensure that the information is clear, easy to understand, and can quickly convey key information.

[0127] After generating the warning information, the system needs to prompt the operation and maintenance personnel in real time through the user interface. This can be achieved in the following ways:

[0128] Interface pop-up window, pop up a window containing the warning information in the user interface to ensure that the operation and maintenance personnel can notice it immediately when using the system.

[0129] Sound or vibration reminder, if the system supports, it can remind the operation and maintenance personnel of new warning information through sound or vibration.

[0130] Mobile application push, if the system integrates a mobile application, it can send the warning information to the mobile devices of the operation and maintenance personnel through the application push function.

[0131] Email or SMS notification, as an alternative, the system can also send the warning information to the operation and maintenance personnel through email or SMS to ensure that they can receive the notification in time even when they are not in front of the system.

[0132] After receiving the warning information, the operation and maintenance personnel should take corresponding actions according to the content of the information and the suggested measures. They may need to go to the site for further inspection, initiate the emergency response process or coordinate with other relevant departments. Through this step, the dynamic detection system for potential hazards around transmission lines based on image difference recognition can achieve the rapid discovery and effective disposal of potential hazards, thereby improving the safety and stability of transmission lines.

[0133] As Figure 2 shown, a dynamic detection method for potential hazards around transmission lines based on image difference recognition includes:

[0134] Collect video and image data around the transmission line, and the data is sourced from video monitoring devices arranged around the transmission line;

[0135] Preprocess the collected video and image data to obtain preprocessed video and image data;

[0136] Use machine learning algorithms to perform difference recognition on the preprocessed images. By comparing consecutive frames of images, detect outliers in the environment around the transmission line; Adjust the outliers in the environment around the transmission line according to dynamic factors to obtain a correction index;

[0137] Analyze the correction index to determine the type, location, and severity of potential hazards;

[0138] Generate a warning message based on the type, location, and severity of the potential hazard, and prompt the operation and maintenance personnel in real time through the user interface.

[0139] It should be noted that this system corresponds to the above method. All implementation manners in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0140] An embodiment of the present invention also provides a computing device, including: a processor and a memory storing a computer program. When the computer program is run by the processor, it executes the method described above. All implementation manners in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0141] An embodiment of the present invention also provides a computer-readable storage medium storing instructions. When the instructions are run on a computer, the computer is made to execute the method described above. All implementation manners in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0142] The above is the preferred implementation manner of the present invention. It should be pointed out that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. A dynamic detection system for hidden dangers around power transmission lines based on image difference recognition, characterized in that: include: A data acquisition module, used to collect video and image data around the transmission line, the data comes from video monitoring equipment arranged around the transmission line; An image preprocessing module is used to preprocess the collected video and image data to obtain preprocessed video and image data; The image difference recognition module is used to perform difference recognition on the pre-processed image using a machine learning algorithm, and detect abnormal values ​​of the environment around the transmission line by comparing images of consecutive frames; the abnormal values ​​of the environment around the transmission line are adjusted according to dynamic factors to obtain a correction index; Hidden danger analysis module, used to analyze the correction index and determine the type, location and severity of hidden dangers; The early warning prompt module is used to generate early warning information according to the type, location and severity of hidden dangers, and prompt operation and maintenance personnel in real time through the user interface.

2. The dynamic detection system for hidden dangers around power transmission lines based on image difference recognition according to claim 1 is characterized in that: Collect video and image data around the transmission line, the data comes from video monitoring equipment arranged around the transmission line, including: In the layout space of video surveillance equipment around the transmission line, a certain number of gray wolf individuals are randomly initialized, i.e., candidate video surveillance equipment locations; each gray wolf individual represents a solution, including the location, orientation, and focal length parameters of the video surveillance equipment; Define an objective function, which is used to evaluate the quality of individual gray wolves, that is, the quality of data collected by video surveillance equipment under specific locations and parameter configurations; The individual gray wolf updates its position according to the current final solution, simulates the hunting behavior of the gray wolf, searches for the corresponding solution in the solution space, and gradually approaches the current corresponding solution. When the individual gray wolf is close to the final solution, it performs a search to find the global final solution. According to the output results of the gray wolf optimization algorithm, the position and parameter configuration of the video surveillance equipment are adjusted; Under the updated video surveillance equipment configuration, actual video and image data acquisition is performed to obtain video and image data around the transmission line.

3. The dynamic detection system for hidden dangers around power transmission lines based on image difference recognition according to claim 2 is characterized in that: The collected video and image data are preprocessed to obtain preprocessed video and image data, including: In the space of preprocessing parameters, a certain number of particles, i.e. candidate preprocessing parameter combinations, are randomly initialized. Each particle represents a set of preprocessing parameters, including the type, size and threshold of the filter. Define a fitness function to evaluate the quality of preprocessed video and image data; According to the final historical position of the individual and the final historical position of the group, the speed of each particle is updated, and according to the updated speed, the position of each particle is adjusted, i.e., the preprocessing parameters; Perform preprocessing operations on each new position, and use the fitness function to evaluate the preprocessing effect, and repeat the speed update, position update and evaluation update steps until the preset number of iterations is reached to obtain the preprocessing parameter combination corresponding to the final historical position of the group; The collected video and image data are preprocessed using a combination of preprocessing parameters corresponding to the final position to obtain preprocessed video and image data.

4. The system for dynamic detection of hidden dangers around power transmission lines based on image difference recognition according to claim 3 is characterized in that: The machine learning algorithm is used to identify the differences of the pre-processed images and detect abnormal values ​​of the surrounding environment of the transmission line by comparing the images of consecutive frames, including: Extract continuous image frames from the video stream at a set frame rate; Perform feature extraction on each frame of image to obtain a feature vector representing the image content; Determine the number of initial frames used to establish the Gaussian mixture model, use the feature vectors of the initial frames as training data, input them into the Gaussian mixture model, and learn the distribution of the feature vectors through the Gaussian mixture model algorithm to establish a Gaussian mixture model representing a "normal" state; Set the parameters of the Gaussian mixture model, including the number of Gaussian distributions and the type of covariance matrix; For each new image frame in the video stream, its feature vector is extracted and input into the established Gaussian mixture model. The difference between the new image feature vector and each Gaussian distribution in the Gaussian mixture model is calculated to obtain the difference between the new image feature vector and the Gaussian mixture model.

5. The system for dynamic detection of hidden dangers around power transmission lines based on image difference recognition according to claim 4 is characterized in that: Perform feature extraction on each frame of the image to obtain a feature vector representing the image content, including: Receive the preprocessed single-frame image data as input and initialize the SIFT feature detection algorithm, including setting the number of layers and scale factor of scale space extreme value detection; By performing Gaussian blurring and downsampling on the input image at different scales, the scale space of the image is constructed to find stable feature points at different scales. In the scale space, by comparing the pixel values ​​of each point with those of adjacent scales and adjacent positions, local extreme points are found, and the extreme points are considered as candidate key points; Screen and locate the candidate key points detected in the previous step, and assign a main direction to each key point by calculating the gradient direction and amplitude of the pixels around the key point; In the neighborhood around the key point, the gradient information of the pixels is counted according to the assigned main direction to generate a descriptor vector of fixed length; The descriptor vectors of all key points are organized into a feature vector set, which represents the content of the entire image.

6. The system for dynamic detection of hidden dangers around power transmission lines based on image difference recognition according to claim 4 is characterized in that: The distribution of feature vectors is learned through the Gaussian mixture model algorithm to build a Gaussian mixture model representing the "normal" state, including: Collect feature vector data representing the "normal" state and determine the number of Gaussian distributions used, i.e., the number of components of the mixture model; initialize parameters for each Gaussian distribution, including the mean vector, covariance matrix, and mixture coefficient, i.e., the weight of each Gaussian distribution in the mixture model; Use the expectation maximization algorithm to iteratively update the Gaussian mixture model parameters, and calculate the posterior probability that each data point belongs to each Gaussian distribution, i.e., the degree of responsibility, based on the current Gaussian mixture model parameters; Based on the calculated posterior probability, the mean, covariance and mixing coefficient of each Gaussian distribution are updated, and the iteration converges until the Gaussian mixture model parameters converge; Use the silhouette coefficient to evaluate the fitting effect of the Gaussian mixture model to obtain the evaluation results; According to the evaluation results, the corresponding Gaussian mixture model is determined as the Gaussian mixture model representing the "normal" state.

7. The system for dynamic detection of hidden dangers around power transmission lines based on image difference recognition according to claim 4 is characterized in that: The calculation formula of dynamic factor is: Where A(t) represents the dynamic adjustment factor; α represents the adjustment coefficient; I i (t) and I i (t+1) represents the gray value of the i-th pixel in image I(t) and image I(t+1) respectively; N is the total number of pixels in the image; β represents the adjustment coefficient; v(t) represents the wind speed at time t; κ is a constant, which represents the magnification of the wind speed to the image difference; γ represents the adjustment coefficient; b i (t) is the actual value of the i-th pixel in image b(t); is the denoised estimate of the i-th pixel in image b(t); i represents the index value.

8. The dynamic detection system for hidden dangers around power transmission lines based on image difference recognition according to claim 7 is characterized in that: The abnormal values ​​of the surrounding environment of the transmission line are adjusted according to the dynamic factors to obtain the correction index, including: Multiply the dynamic factor by the corresponding difference value to obtain the correction index.

9. A method for dynamic detection of hidden dangers around power transmission lines based on image difference recognition, characterized in that: The method is used to execute the system according to any one of claims 1 to 8, and the method comprises the following steps: Collecting video and image data around the transmission line, the data comes from video monitoring equipment arranged around the transmission line; Preprocessing the collected video and image data to obtain preprocessed video and image data; The machine learning algorithm is used to identify the difference of the preprocessed images, and the abnormal values ​​of the environment around the transmission line are detected by comparing the images of consecutive frames; the abnormal values ​​of the environment around the transmission line are adjusted according to the dynamic factors to obtain the correction index; Analyze the modified index to determine the type, location and severity of hidden dangers; According to the type, location and severity of hidden dangers, early warning information is generated and the operation and maintenance personnel are prompted in real time through the user interface.

10. A computing device, characterized in that include: one or more processors; A storage device for storing one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors implement the method as claimed in claim 9.

Images