A method, equipment, and medium for detecting the condition of high-risk objects in a power transmission line corridor.

Abstract

This invention discloses a method, equipment, and medium for detecting the status of high-risk objects in power transmission line corridors, belonging to the field of intelligent operation and maintenance technology for power system transmission lines. The method includes: mining the impact characteristics of high-risk objects based on image and video data of the transmission line corridor; spatially labeling the high-risk objects based on these characteristics to construct a high-risk object resource database; identifying high-risk objects and initially delineating suspected threat areas based on video data; identifying conductor status changes based on image data; determining whether to issue a high-risk object threat warning for the transmission line corridor based on the high-risk object identification results and conductor status change identification results; and performing time-domain analysis on the threat warning data to generate a risk prediction report. This invention achieves a leap from passive response to proactive warning, significantly improving the accuracy and proactivity of external force risk prevention and control in transmission line corridors.

Description

Technical Field

[0001] This invention relates to the field of intelligent operation and maintenance technology for power transmission lines, specifically to a method, equipment, and medium for detecting the status of high-risk objects in transmission line corridors. Background Technology

[0002] Early monitoring relied primarily on manual inspections and fixed sensors such as LiDAR. Today, we've entered an era of visual perception centered on video surveillance. On one hand, deep learning object detection algorithms, such as YOLO and Faster R-CNN, are maturing and can accurately identify common high-risk objects like cranes and excavators. On the other hand, multi-source information fusion is becoming a trend, with technicians attempting to combine video footage with SCADA systems and meteorological data for comprehensive analysis. More cutting-edge explorations point to three-dimensional and refined methods, such as using binocular vision or laser point clouds to measure distances, and even constructing digital twin scenarios for simulation and deduction.

[0003] Overall, current technological trends remain focused on more accurately detecting objects and more precisely measuring static distances. However, a crucial issue is often overlooked: existing technologies are far from adequate for assessing the dynamic and destructive behavior of these objects, and the complex interactions between this behavior and the state of the conductor.

[0004] For this reason, existing technologies have revealed several significant weaknesses in practical applications. First, most systems rely on general detection models, whose judgments are based on the appearance of objects, failing to deeply integrate the unique operational and maintenance knowledge and historical fault data of the power system. This leads to an awkward situation: the system can identify a crane, but struggles to determine whether a specific angle of its boom poses a substantial threat at a given wind speed; it can detect a pile driver, but cannot assess the impact of its vibration frequency on nearby conductors. The consequence is often that, for safety's sake, the system can only use overly conservative fixed distance thresholds for alarms, resulting in frequent false alarms, while truly dangerous behaviors may be missed because the threshold is not met. Second, in the core aspect of risk quantification, existing methods typically use simple two-dimensional pixel distances or coarse three-dimensional spatial distances, ignoring the perspective distortion of monocular vision, the true three-dimensional posture of objects, and the dynamic changes in conductors caused by wind deflection and sag. This disconnect between the model and physical reality significantly reduces the reliability of risk assessment. More importantly, the existing system lacks a smart closed loop for continuous learning from historical lessons. When a power outage occurs, maintenance personnel often have to search through massive amounts of video footage to trace the cause, which is time-consuming and laborious. Moreover, valuable handling experience is difficult to be accumulated and transformed into quantifiable rules that the system can understand and iterate upon. Summary of the Invention

[0005] In view of the above-mentioned problems, the present invention is proposed.

[0006] Therefore, this invention aims to solve the problem of inaccurate early warning caused by the inability to accurately identify specific high-risk objects and their actual threats in existing power transmission line corridors, as well as the multi-dimensional coupling effect of actual response status and environmental factors.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a method for detecting the status of high-risk objects in a power transmission line corridor, comprising, The impact characteristics of high-risk objects are analyzed based on image and video data of the transmission line corridor; based on these impact characteristics, high-risk objects are spatially labeled to construct a high-risk object resource database; high-risk object identification is performed on video data, and conductor state change identification is performed on image data; based on the identification results, it is determined whether to issue a high-risk object threat warning for the transmission line corridor; time-domain analysis is performed on the threat warning data to generate a risk prediction report.

[0008] As a preferred embodiment of the method for detecting the state of high-risk objects in a power transmission line corridor according to the present invention, the analysis of the influence characteristics of high-risk objects includes: Obtain the first and second databases of the transmission line corridor; Retrieve suspected causative image sequences from the first database; Preprocessing of suspected causative image sequences; Moving target detection and trajectory tracking are performed on the preprocessed suspected causative image sequence to extract target behavior image fragments; The system receives descriptive information from target behavior image fragments as input, representing the impact characteristics of high-risk objects.

[0009] As a preferred embodiment of the method for detecting the status of high-risk objects in a power transmission line corridor according to the present invention, the step of constructing a high-risk object resource database includes: Images that meet the aforementioned influence characteristics are selected as samples, and the sample images are aligned with a pre-constructed 3D scene model; Adjust the 3D placeholder frame of the object in the 3D scene model to match the 2D outline of the object in the sample image; Calculate the minimum three-dimensional spatial distance between the three-dimensional coordinates of the overlapping object and the three-dimensional coordinates of the guide wire, and construct a high-risk object resource database.

[0010] As a preferred embodiment of the high-risk object status detection method for power transmission line corridors according to the present invention, the high-risk object identification includes: Foreground extraction is performed on real-time monitoring images to obtain candidate regions, and features of the candidate regions are extracted. The features are compared with those in the high-risk object resource database to generate a preliminary category hypothesis. For each of the primary category hypotheses, perform Level 1, Level 2, and Level 3 verifications sequentially. The fusion confidence level of each candidate region corresponding to each category is calculated based on the validation results at each level. When a candidate region whose fusion confidence exceeds the global determination threshold is identified as a high-risk object, a suspected threat region is defined with the high-risk object as the center.

[0011] As a preferred embodiment of the method for detecting the state of high-risk objects in a power transmission line corridor according to the present invention, the identification of conductor state changes includes: Maintain the data of the conductor to obtain the pixel-level centerline; Based on pixel-level centerlines, a non-interference-resistant strip search strategy is used to extract wires and obtain sub-pixel precision centerlines. The unilateral visual tension index of the current conductor segment is calculated based on the subpixel precision centerline and real-time data. The unilateral visual tension index is compared with a dynamic threshold to determine whether there is a change in the state of the conductor.

[0012] As a preferred embodiment of the method for detecting the state of high-risk objects in a power transmission line corridor according to the present invention, the identification result includes: Calculate the shortest spatial distance from the high-risk object to the conductor, query the corresponding impact characteristics of the high-risk object, and obtain the basic warning distance; Determine the object state factor based on the behavioral state identifier of high-risk objects; Calculate the conductor state factor based on the conductor state change judgment results; Obtain real-time wind speed and calculate environmental factors based on the real-time wind speed and reference wind speed; The comprehensive threat index is obtained by multiplying the ratio of the basic warning distance to the shortest spatial distance as the base, and then multiplying it by the object state factor, the conductor state factor, and the environmental factor.

[0013] As a preferred embodiment of the method for detecting the status of high-risk objects in a transmission line corridor according to the present invention, the step of determining whether to issue a high-risk object threat warning for the transmission line corridor includes: The comprehensive threat index is compared with the first threshold and the second threshold; When the comprehensive threat index is less than the first threshold, it is determined to be safe; When the comprehensive threat index is greater than or equal to the first threshold and less than the second threshold, it is determined to be a yellow warning level; When the comprehensive threat index is greater than or equal to the second threshold, it is determined to be a red alert level.

[0014] As a preferred embodiment of the method for detecting the state of high-risk objects in a power transmission line corridor according to the present invention, the time-domain analysis includes: Based on historical risk warnings of transmission line corridors, obtain continuously updated data streams; Define a grid for historical risk warning data, calculate the density of warning events within each grid, and merge adjacent grids as risk hotspot areas; For each identified risk hotspot area, periodic and trend analyses are conducted. Spatial analysis is performed on the aforementioned risk hotspot areas to infer the causes of the risks and generate a risk prediction report.

[0015] The present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, characterized in that the processor executes the computer program to implement the steps of the method for detecting the state of high-risk objects in a power transmission line corridor.

[0016] The present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that, when the computer program is executed by a processor, it implements the steps of the method for detecting the state of high-risk objects in a power transmission line corridor.

[0017] The beneficial effects of this invention are as follows: By retrospectively analyzing historical tripping events and associated image data, this invention transforms implicit operational experience into structured rules governing the impact characteristics of high-risk objects. Based on these rules, samples are selected and three-dimensional spatial annotations are completed, establishing a resource library that integrates object visual features and real geographic information, providing a "visual-spatial" standard for identification. In the real-time detection phase, a three-level verification mechanism is used to identify high-risk objects and generate three-dimensional threat areas. Simultaneously, by analyzing the local morphological balance of conductors, its external force state is innovatively perceived. Subsequently, object information, conductor response, and environmental data are integrated, and a dynamic threat index is calculated through a multi-factor coupling model to achieve accurate hierarchical early warning. Furthermore, spatiotemporal clustering and trend analysis are performed on the early warning data to achieve risk hotspot prediction and operational strategy optimization. This invention thus achieves a leap in external force risk prevention and control for transmission line corridors from post-event response to real-time perception and then to pre-event prediction, significantly improving the accuracy of early warning and the initiative of operation and maintenance. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1The above is a flowchart of a method for detecting the status of high-risk objects in a power transmission line corridor, provided as an embodiment of the present invention.

[0020] Figure 2 This is a flowchart illustrating the high-risk object identification method for a high-risk object status detection method in a power transmission line corridor, as provided in one embodiment of the present invention.

[0021] Figure 3 The flowchart illustrates the time-domain analysis of a method for detecting the state of high-risk objects in a power transmission line corridor, as provided in one embodiment of the present invention. Detailed Implementation

[0022] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0023] Example 1, referring to Figure 1 This is one embodiment of the present invention, which provides a method for detecting the status of high-risk objects in a power transmission line corridor, including: S100. Analyze the impact characteristics of high-risk objects based on image and video data of the power transmission line corridor.

[0024] S200. Based on impact characteristics, high-risk objects are spatially labeled to construct a high-risk object resource database.

[0025] S300: Identify high-risk objects from video data and identify changes in conductor status from image data.

[0026] S400: Based on the identification results, determine whether to issue a high-risk object threat warning for the power transmission line corridor.

[0027] S500 performs time-domain analysis on threat warning data and generates risk prediction reports.

[0028] It should be noted that in the existing technology, there are still some problems such as implicit experience and unstructured information; the disconnect between two-dimensional visual information and three-dimensional spatial information; and the high false alarm rate of target recognition.

[0029] Therefore, to address the aforementioned problems, a knowledge base on the impact characteristics of high-risk objects is constructed through steps S100-S500 and correlation analysis of historical tripping events. Based on this, samples are selected and three-dimensional spatial annotations are performed to establish a resource library integrating visual and geographic information. In real-time detection, a three-level verification mechanism identifies high-risk objects and generates three-dimensional threat areas, while simultaneously analyzing the local morphological balance of conductors to perceive external forces. Subsequently, object state, conductor response, and environmental data are integrated, and a dynamic threat index is calculated using a multi-factor coupling model to achieve precise graded early warning. Finally, spatiotemporal clustering and trend analysis are performed on the early warning data to achieve risk prediction.

[0030] Example 2, refer to Figures 1-3 This is one embodiment of the present invention, which provides a method for detecting the status of high-risk objects in a power transmission line corridor, including: In this embodiment of the invention, step S100, based on image and video data of the power transmission line corridor, mines the impact characteristics of high-risk objects, including the following steps S101-S104: S101. Obtain the first and second databases of the transmission line corridor. The specific operation steps are as follows: This invention takes high-risk objects around power transmission line corridors, such as pile drivers, excavators, cranes, and trucks, as the research object, and obtains two structured databases from the internal server of the power transmission line corridor, including a first database and a second database. The first database is a large sample database of historical images and videos of power transmission line corridors. Each record includes, but is not limited to: image data, video data, acquisition timestamp (down to the year, month, day, hour, minute, and second), unique identifier of the acquisition device, and pre-calibrated internal and external parameters of the current camera.

[0031] The second database is a statistical database of line tripping events. Each specific record includes, but is not limited to: a unique tripping event number, a tripping timestamp, the line section or tower number where the trip occurred, the tripping type, and associated meteorological station data records.

[0032] It should be noted that, in an optional embodiment, the tripping type can be selected from a predefined enumeration list, including but not limited to lightning strike, wind-induced flashover, external force short circuit, tree obstruction, bird damage, etc. The data in the first database is preprocessed to standardize, including scaling images of different resolutions to a standard size and standardizing the color space to reduce the impact of lighting changes.

[0033] S102. Retrieve suspected causative image sequences from the first database. The specific steps are as follows: Read a tripping event record from the second database, and parse the tripping time and geographical location in the record. It should be noted that in the embodiments of the present invention, the geographical location is determined by the tower number. The time backtracking window length is set, which is based on the empirical value of the reaction time for transmission line fault early warning, and can be 30 minutes; the retrieval start time is calculated by multiplying the tripping time and the time backtracking window length. Furthermore, in the first database, image records that meet all of the following conditions are retrieved: The image acquisition timestamp is within the closed interval from the start time of retrieval to the tripping time, and the tower number corresponding to the acquisition device identification number and the geographical location of the tripping are within the preset range of adjacent towers. It should be noted that, in an optional embodiment, the preset range of adjacent towers can be set to two towers before and after. Furthermore, all retrieved images and video records are arranged in chronological order of collection time to form a sequence of suspected causal images associated with the current power outage event.

[0034] S103. Preprocess the suspected causative image sequence. The specific steps are as follows: The suspected causative image sequence is processed, and the earliest frame in the sequence is selected as the initial background frame. Each subsequent frame in the sequence is then subjected to pixel-level grayscale difference operation with the initial background frame to obtain a difference image. The difference image is then binarized using a grayscale threshold. It should be noted that the grayscale threshold value is based on an 8-bit grayscale image and is set to a range of 15 to 30. This range is tested experimentally to separate moving targets from noise caused by slight changes in illumination.

[0035] Furthermore, morphological operations are performed on the binarized image. First, a 3x3 pixel structuring element is used to perform a closing operation to fill the holes inside the target. Then, the same structuring element is used to perform an opening operation to eliminate small noise points. In the binary image after morphological operations, all connected components are identified, and the number of pixels in each connected component is calculated. Using an area filter, connected components with an area less than 0.02% of the total image pixels and greater than 20% of the total image pixels are removed. It should be noted that 0.02% of the total image pixels is mainly used to filter out small objects such as flying birds and drifting leaves, while 20% of the total image pixels is used to filter out interference such as large areas of cloud shadows and sudden changes in overall illumination.

[0036] Furthermore, the filtered connected components are marked as the initial target and recorded as the position of the bounding rectangle in the current frame image.

[0037] S104. Perform moving target detection and trajectory tracking on the preprocessed suspected causative image sequence, extract target behavior image segments, and receive the descriptive information input from the target behavior image segments as the influence characteristics of high-risk objects. The specific operation steps are as follows: The image sequence is subjected to moving target detection and trajectory tracking to extract target behavior image segments; descriptive information input from the target behavior image segments is received as the influence characteristics of high-risk objects. The specific operation steps are as follows: The initial target appearing in multiple consecutive frames is correlated across frames to form a motion trajectory; It should be noted that the association rules for cross-frame association are based on the appearance similarity and motion continuity of the initial targets; For the initial target in frame The location, in In this process, all initial targets are searched within a dynamically expanded prediction region based on the motion velocity of the initial target in the previous frame. The appearance similarity between the candidate targets and the initial targets is calculated. The appearance similarity is determined by comparing the Bach distance of the color histograms within the rectangular regions of the two targets. Simultaneously, the image pixel distance between the initial target position and the predicted position is calculated using Euclidean distance; by weighting the appearance similarity score and the image pixel distance, the candidate target with the highest total score is selected as the Euclidean distance. Matching within a frame.

[0038] Furthermore, a motion trajectory is generated for each continuously appearing initial target, and the trajectory data includes the target's position, timestamp, and target image patch in each frame; If the duration of a motion trajectory exceeds 10 seconds and its spatial movement range exceeds 5% of the image width, the current trajectory is considered a valid behavior trajectory. From the original sequence, the image subsequence from the first appearance of the current target to its departure or until the tripping moment is extracted and packaged into target behavior image fragments.

[0039] The tripping event and the corresponding target behavior image fragment are pushed to the operator's workbench interface in the form of a work order. The interface displays the following side by side: tripping event details; target behavior image fragment, in which the initial target is highlighted; and an editable rule assumption form.

[0040] Operators observe the behavior of the initial target in the image sequence. They describe the current object's category, behavior, and environment in a form. Based on their expertise and trip type, they process historical trip events and construct the impact characteristics of different high-risk objects. The goal is to transform difficult-to-quantify operational experience into calculable and searchable structured rules through repeated comparison and correction with specific historical image evidence.

[0041] It is important to note how to transform the hard-to-express experiences of operations and maintenance personnel based on historical events into searchable and computable structured knowledge; this provides a data-driven and reliable prior knowledge foundation for subsequent automated identification and evaluation.

[0042] In this embodiment of the invention, S200 involves spatially labeling high-risk objects based on the aforementioned impact characteristics to construct a high-risk object resource database, including the following steps S201-S204: S201. Select images that meet the aforementioned influence characteristics as samples, and align the sample images with the pre-constructed 3D scene model. The specific operation steps are as follows: Read the impact characteristics of different high-risk objects. Each high-risk object has corresponding risk characteristics, including: object category, object behavior description, environmental conditions, and a risk triggering condition expressed as image pixel distance. Taking a crane as an example, its characteristic is that the visual projection distance of the crane's boom extension behavior is less than a certain pixel when the wind speed is greater than a specified speed.

[0043] Images containing all high-risk objects are selected from historical data as candidate samples. For each candidate sample image, the region of suspected high-risk object in the image is detected, and the minimum image pixel distance between the suspected region and the wire outline in the image is estimated. It should be noted that the minimum image pixel distance is obtained by extracting a discrete set of edge points from the conductor contour data. The set of edge points consists of two-dimensional coordinates. The boundary of the suspected high-risk object region is obtained. For the automatically detected region, its minimum bounding rectangle is calculated to obtain another set of two-dimensional boundary point coordinates. Each point in the conductor edge point set is traversed. For each conductor point, the Euclidean distance to each point in the object region boundary point set is calculated. The Euclidean distance is calculated based on the difference between the horizontal and vertical coordinates of the two points, which is obtained by taking the square root of the sum of the squares. Among all the calculated distance values, the minimum value is found and recorded, which is the minimum image pixel distance.

[0044] Specifically, the minimum image pixel distance is compared with the image pixel distance threshold Z corresponding to the influence characteristics of high-risk objects: In response to a minimum image pixel distance being less than or equal to 120% of the image pixel distance threshold, i.e., 1.2Z, the current object is marked as a high-risk potential sample and automatically imported into the resource library construction workflow. It should be noted that this ensures that limited annotation resources are concentrated on scenarios that are already close to or meet the risk conditions in two-dimensional images, thereby improving the efficiency and relevance of library construction.

[0045] S202. Adjust the 3D placeholder frame of the object in the 3D scene model to coincide with the 2D outline of the object in the sample image. The specific operation steps are as follows: For each imported high-risk potential sample image, it is back-projected onto the 3D digital ground model according to the principle of perspective projection, and spatial annotation is performed. However, in reality, due to monocular vision and errors in the ground model and the actual parking state of the object, the initial 3D placeholder box deviates from the true 3D position and pose of the object in the image. Therefore, the original 2D image, the current view rendering of the 3D scene model, and the initial 3D placeholder frame are displayed synchronously. By dragging and dropping with the mouse, the current placeholder can be moved and rotated along the ground in the 3D scene. The length, width, and height of the placeholder can be adjusted, as well as its pitch and tilt angles. The 2D projection of the 3D placeholder in the current camera viewpoint can be pixel-level aligned with the 2D outline polygon of the object manually annotated in the original image. Once the two objects overlap, record the final three-dimensional space coordinates of the eight vertices, the center point coordinates, the length, width, height dimensions, and Euler angles of the current three-dimensional placeholder frame to determine the spatial position, size, and orientation of the high-risk object in a unified coordinate system.

[0046] S203. Calculate the minimum three-dimensional spatial distance between the three-dimensional coordinates of the overlapping object and the three-dimensional coordinates of the guide wire, and construct a high-risk object resource database. The specific operation steps are as follows: Based on the obtained precise 3D spatial occupant model of the high-risk object and the existing 3D spatial curve model of the guide wire in the 3D scene, the shortest distance from the object's feature points to the guide wire is calculated. Specifically: Feature points are sampled from the surface of the object's 3D placeholder frame. For each point on the 3D curve of the guide wire, the 3D Euclidean distance from it to each feature point of the object is calculated. All point pairs are traversed, and the minimum distance value is found to be the shortest distance from the object's feature point to the guide wire. At the same time, the shortest feature point on the object and the shortest point on the guide wire are recorded. Establish a mapping relationship between the image pixel distance threshold Z and the calculated minimum distance value; More specifically, the ratio of the distance between the image pixels that cause risk determination in the current object sample to the minimum distance value is used as a scaling factor, and the current scaling factor, the calculated minimum distance value, and the corresponding rule ID are associated and stored.

[0047] S204. In response to the results of all spatial annotations, construct a high-risk object resource database, including: identification information, object information, 3D spatial information, risk distance information, rule association information, and environmental information; Specifically, the identification information includes object ID, source image ID, acquisition time, and tower number; the object information includes object category and two-dimensional coordinates; the three-dimensional spatial information includes the aligned three-dimensional placeholder frame and the three-dimensional coordinates of the nearest feature point on the object; the risk distance information includes the calculated minimum distance value from the object to the conductor and the three-dimensional coordinates of the nearest point on the conductor; the rule association information is a scaling factor; and the environmental information includes weather conditions and lighting conditions.

[0048] It is important to explain how to establish an accurate correspondence between empirical rules based on two-dimensional image pixels and real-world three-dimensional geographic coordinates and spatial distances, so that the recognition results not only match the appearance but also have a reasonable physical location.

[0049] In an embodiment of the present invention, step S300 involves identifying high-risk objects from video data and identifying changes in conductor status from image data, including the following steps S301-S310: Reference Figure 2 S301. Foreground extraction is performed on the real-time monitoring image to obtain candidate regions, and features of the candidate regions are extracted. The specific operation steps are as follows: For any input real-time image, a Gaussian mixture model is used to model the background and extract all moving foreground regions; Meanwhile, by selecting the Gaussian component with the largest weight from the Gaussian mixture model, reading the mean of the current Gaussian component as the predicted brightness value on the background reference image of the current pixel, and subtracting the brightness value of each pixel in the current input frame (if it is a color image, it is usually converted to the brightness channel first) from the predicted brightness value at the corresponding position in the background reference image pixel by pixel and taking the absolute value, the absolute difference map between the current frame and the background model is calculated. In response to static regions where the absolute difference value is greater than or equal to a preset brightness change threshold, these regions are also marked as change regions to be analyzed. It should be noted that the preset brightness change threshold is set as the sum of the average absolute difference between the brightness value of each pixel and its corresponding background model prediction value within a statistical continuous time period and twice the standard deviation.

[0050] All moving foreground regions and the changing regions to be analyzed are merged to generate an initial set of candidate connected regions, and the pixel area of ​​each candidate connected region is calculated.

[0051] S302. Compare the features with those in the high-risk object resource database to generate a preliminary category hypothesis. The specific steps are as follows: For each candidate region, calculate the minimum bounding rectangle and obtain the aspect ratio based on the ratio of its width to its height; Calculate the seven Hu invariant moments of the contour of the candidate connected region to form the contour feature vector; In response to the current candidate connected region being tracked in consecutive frames, the average motion velocity and motion direction over the past 5 frames are calculated; In one optional embodiment, short-term tracking can be performed using centroid-based Kalman filter prediction and Hungarian algorithm matching for tracking in consecutive frames.

[0052] Furthermore, the aspect ratio, contour feature vector, and average motion velocity of each candidate connected region are compared with the features of all sample records in the high-risk object resource database. The comparison uses a weighted Euclidean distance metric. The difference between aspect ratio and average motion speed is calculated after normalization based on standard deviation. For the contour feature vector, the inverse variance of each Hu moment obtained from the resource library is used as the weight to calculate the weighted shape distance; For each candidate connected region, find the K nearest samples in the resource library and calculate the distribution of the object categories to which these K samples belong; Based on this distribution, at least one primary category is generated for candidate connected regions, including object category and weight corresponding to object category, wherein the weight corresponding to object category is the number of samples belonging to the current object category among the nearest neighbors of K samples divided by K. At the same time, the range of physical parameters corresponding to the current object category is extracted from these K samples, including: aspect ratio range, area range, and motion mode; S303. For each of the primary category hypotheses, perform the first-level verification, the second-level verification, and the third-level verification in sequence. The specific operation steps are as follows: For each candidate connected region and its corresponding primary class hypothesis, a three-level sequential validation is performed. Each level of validation generates a sub-score, and the confidence level of the current primary class hypothesis is dynamically updated. Specifically, the first level of verification determines whether the measured features of the candidate connected region fall within the range of physical parameters of the assumed object category: The first condition is to determine whether the aspect ratio is within the stated aspect ratio range; The second condition is to determine whether the pixel area of ​​each candidate connected region is within the stated area range; The third condition is to determine whether the motion pattern of the candidate connected region matches the motion pattern. For example, a pile driver should have a vibration mode, while a truck is usually in translation or stationary mode. If all conditions are met, the first level of validation passes, and the first sub-score is... =1.0; if any condition is not met, then the first subfraction... =0.2.

[0053] Furthermore, the second level of verification is contour matching verification based on the projection of the 3D model: Using the existing 3D models of the corresponding object categories in the resource library, and using the calibration parameters of the current camera, the 3D models of the object categories in the resource library are projected onto the current image plane; By iteratively adjusting the current 3D model's 2D position on the horizontal ground and its rotation angle θ around the vertical axis, the matching degree between the 2D contour of the model projection and the actual 2D contour of the candidate connected region is maximized. It should be noted that the matching degree is calculated by measuring the bidirectional distance between the projected contour of the model and the actual contour B of the candidate connected region: For each sampling point on contour M, calculate the shortest Euclidean distance from the sampling point to contour B and calculate the average value to obtain d(M->B). Similarly, calculate d(B->M) and define the contour matching score. It should be noted that d(B->M) represents the one-way average distance from contour B to contour M, that is, for each sampling point on contour B, the shortest Euclidean distance from the sampling point to contour M is calculated and the average value is obtained. It should also be noted that the contour matching score is defined as the average of the shortest distances from each point on the projected contour of the 3D model to the actual object contour, and the average of the shortest distances from each point on the actual object contour to the projected contour of the model. These two averages are added together and divided by 2 to obtain the bidirectional average distance value. This bidirectional average distance value is then normalized by dividing it by the equivalent diameter of the projected contour of the model to obtain the normalized distance value. Finally, the number 1 is divided by the sum of 1 and the normalized distance value, and the result is the contour matching score. The maximum contour matching score is between [0,1], with a better match being closer to 1. The maximum contour matching score obtained by continuously optimizing the projection position is the second sub-score. If the maximum contour matching score obtained through optimization is lower than the preset matching threshold, the second-level verification is considered to have failed. =0, where the preset matching threshold is set to 0.6. The preset matching threshold is determined based on a large number of experiments. If the value is lower than this, the contour matching is considered to be too low.

[0054] Furthermore, the third level of verification is the verification of the rationality of the spatial location and altitude: Only after passing the second-level verification can the third-level verification be performed. This involves using the coordinates in the projected 3D model to verify whether the height values ​​of these points are within the allowable range of the Ground Elevation Model (DEM). If they pass, the third sub-score is obtained. =1, otherwise 0.

[0055] S304. Calculate the fusion confidence score of each category for the candidate region based on the validation results at each level. The specific steps are as follows: For each object category in the candidate connected region, the corresponding fusion confidence score is... for: ; in, , , Assigned to respectively , as well as The weighting coefficients satisfy the condition that the sum of the three is 1. In this embodiment, the weighting coefficients are set to... =0.3, =0.5, =0.2; These are the weights set in the primary category hypothesis; It is a distance deviation penalty item; This is the penalty coefficient, set to 2.0. The aim is to identify anomalies where the contour matching may be good by chance, but the spatial location does not conform to the distance-size relationship of this type of object.

[0056] It should be noted that the distance deviation penalty term is calculated as follows: from the samples of object categories in the resource library, the statistical relationship between the horizontal distance of the object to the camera and the pixel area of ​​the object in the image under similar viewpoints is statistically determined. Using the pixel area of ​​the current candidate connected region, the expected horizontal distance is predicted through statistical relationships. At the same time, based on the continuously optimized projection position of the second level and combined with the camera extrinsic parameters, the actual horizontal distance is calculated. Specifically, the ratio of the absolute value of the actual horizontal distance minus the predicted expected horizontal distance to the predicted expected horizontal distance represents the relative deviation between the actual and predicted distances.

[0057] Furthermore, in response to a fusion confidence level greater than the global decision threshold, regions that meet this condition are identified as high-risk objects, where the global decision threshold is set to 0.65.

[0058] S305. When a candidate region whose fusion confidence exceeds the global determination threshold is identified as a high-risk object, a suspected threat region is defined centered on the high-risk object. The specific operation steps are as follows: For each object that is ultimately determined to be high-risk, the basic safety distance is read from the corresponding impact characteristics; Using the current three-dimensional position of the high-risk object as the center and the basic safety distance as the initial radius, a spherical warning zone is generated in three-dimensional space; Projecting the spherical warning area onto the current two-dimensional image plane yields a closed region, which is the suspected threat area from the current perspective. The suspected threat area is displayed as a semi-transparent, bright color block overlaid on the monitoring screen. At the same time, the system records all information about the current high-risk object.

[0059] In summary, this method addresses how to identify specific categories of high-risk objects, significantly reduces the false recognition rate, and outputs suspected threat areas with three-dimensional geographic coordinates, providing precise targets for subsequent analysis.

[0060] S306. Maintain the data of the conductor to obtain the pixel-level center line. The specific operation steps are as follows: Real-time maintenance of conductor data: For each camera view in the monitoring, during the initial calibration (selecting a working period with no wind and no external interference), through automatic identification, determine two fixed anchor points for each phase conductor in the image: the pixel coordinates of the conductor suspension point in the image; Using these two anchor points and the shape of the conductor sag under the camera, a pixel-level center line representing the current phase conductor in the reference state is generated through interpolation calculation. At the same time, the apparent width of the conductor at certain pixel intervals along the center line in the reference state is recorded. The apparent width of the conductor is obtained by edge detection in the direction perpendicular to the center line at that point.

[0061] S307. Based on pixel-level centerlines, a non-interference-resistant strip search strategy is used to extract wires and obtain sub-pixel precision centerlines. The specific operation steps are as follows: Specifically, guided filtering is performed on the image within a strip area of ​​twice the apparent width of the guideline pixels above and below the pixel-level centerline as the guiding center. Furthermore, the gray-scale centroid method is used to extract the sub-pixel-level centerline. For each column of pixels within the strip region, calculate the gray value on the cross section perpendicular to the local direction of the conductor. Treating the gray value as mass, the gray centroid position of the cross section is the sub-pixel coordinate of the conductor center in that column. The grayscale centroid position is the ratio of the sum of the products of each pixel position and its grayscale value to the sum of all pixel grayscale values. By calculating column by column, the sub-pixel precision center line of the current frame conductor is obtained. At the same time, at each sub-pixel position of the calculated center line, the apparent width of the current conductor is also obtained by edge detection of its normal cross-section using the Canny operator.

[0062] S308. Spatially align the subpixel precision centerline with the pixel-level centerline. In an optional embodiment, the spatial alignment may be based on a fixed anchor point. Furthermore, calculate the state parameters: average lateral offset and average width change rate; In one embodiment of the present invention, the average lateral offset can be calculated as follows: within the effective length of the conductor, N points are uniformly sampled. For each sampled point, the pixel distance between the corresponding point on the sub-pixel precision center line and the corresponding point on the pixel-level center line in the normal direction to the local pixel-level center line is calculated. The absolute value of this distance for all sampled points is taken and then averaged to obtain the average lateral offset.

[0063] In one embodiment of the present invention, the average width change rate can be calculated by taking the percentage of the apparent width of the current conductor minus the apparent width of the conductor to the apparent width of the conductor at the same N sampling points, and then averaging the absolute values ​​to obtain the average width change rate.

[0064] S309. Calculate the unilateral visual tension index of the current guide segment based on the sub-pixel precision centerline and real-time data. The specific operation steps are as follows: Based on the identified suspected threat areas, find the conductor segment in the image that is spatially closest to this area, and analyze the local morphology of the conductor on both sides of this conductor segment. The specific steps are as follows: select an analysis point on the conductor segment that is spatially closest to the suspected threat area, and divide the conductor segment into a left segment and a right segment with the analysis point as the boundary. Take a continuous sequence of sub-pixel center points on the left segment, fit a second-order polynomial curve using the least squares method, and calculate the left curvature value of the currently fitted curve at the analysis point. The curvature value reflects the degree of curvature of the curve; a straight line has zero curvature, and the greater the curvature, the larger the absolute value of the curvature. Similarly, calculate the right curvature value of the right segment at the analysis point.

[0065] When an object approaches from either side and exerts a pulling or pushing force on the guide wire, the shape of the guide wire at the analysis point will change, resulting in an imbalance of curvature on both sides of the analysis point. In this case, the unilateral visual tension index T is defined as follows: ; in, Let be the left curvature value at the analysis point. Let be the right curvature value at the analysis point. For a minimal constant (e.g.) To prevent division by zero, the T value is not affected by the overall curvature of the conductor, and only the relative change is considered; This represents the average lateral offset. The average width change rate, , It is a small positive integer, and can take the value 0.1 to avoid zero values; This is a baseline constant, which can take the value 1, to ensure that the logarithmic parameter is greater than 0; It should be noted that the T-index integrates local morphological asymmetry (curvature difference) and overall deformation reliability (environmental factors), and does not directly rely on pixel-level micro-movements that are difficult to capture. Instead, it senses the potential force on one side by analyzing the balance of the local geometry of the conductor.

[0066] S310. Compare the unilateral visual tension index with the dynamic threshold to determine whether there is a change in the state of the conductor. The specific operation steps are as follows: The calculated unilateral visual tension index T will be compared with the dynamic threshold. Comparison, The base value was set to 0.15, which was obtained by statistically averaging the T-index under historical no-threat scenarios; Dynamic threshold The adjustment is made dynamically based on the average width change rate, by adding 1 to the product of one-tenth of the average width change rate of the conductor and the base value. Response to T greater than If the condition is met, it is determined that the conductor has a state change in the corresponding suspected threat area, and a Boolean value of true is output. At the same time, the current unilateral visual tension index and the corresponding suspected threat area are recorded, and a conductor state vector is output. .

[0067] In this embodiment of the invention, step S400, based on the identification result, determines whether to issue a high-risk object threat warning for the transmission line corridor, including the following steps S401-S408: Reference Figure 3 The following steps are taken to calculate the shortest spatial distance from the high-risk object to the conductor, query the corresponding impact characteristics of the high-risk object, and obtain the basic warning distance: S401. Receive high-risk object identification, including: the object category of the high-risk object; the current location coordinates of the high-risk object in the three-dimensional geographic coordinate system, denoted as... ; A state identifier describing the behavior of an object, denoted as ; Simultaneously, the system receives the conductor status identification results, including: a real-time three-dimensional set of the conductor's location points, denoted as... ; Lateral offset of the conductor relative to the reference state; Abnormal micro-motion flag, denoted as a Boolean value, a true Boolean value indicates that an abnormal micro-motion has been detected.

[0068] S402. Based on a three-dimensional geographic coordinate system, calculate... arrive Shortest spatial distance, traversal For each 3D point in the array, calculate the relationship between that point and... The Euclidean distance between the objects and the conductors is taken as the minimum of all distances.

[0069] S403. Based on the object category of the high-risk object, query the corresponding impact characteristics of the high-risk object, retrieve the entries corresponding to the current high-risk object category, and extract the basic warning distance parameters. This represents the minimum safe distance that a high-risk object must maintain under standard static conditions.

[0070] S404. Determine the object state factor based on the behavioral state identifier of high-risk objects. The specific operation steps are as follows: State identifiers for parsing object behavior Calculate the object state factor according to predefined mapping rules. The mapping rule is if If the high-risk object is indicated to be in the extended boom position, then... Assign the value 1.2; if If the object is in motion, then Assign the value 1.1; if If the indication is other or default, then Assign the value 1.0.

[0071] S405. Based on the results of the conductor state change judgment, calculate the conductor state factor. The specific operation steps are as follows: Calculate the conductor state factor based on the conductor's lateral offset and abnormal micro-motion indicators. : Set the lateral offset of the reference conductor The value is 0.5 meters, specifically determined based on the maximum permissible wind deflection commonly found in transmission line design specifications. The normalized offset is calculated as the ratio of the lateral offset of the conductor to the lateral offset of the reference conductor. Set the abnormal micro-motion weight coefficient The value is 0.5, where the abnormal micro-motion weighting coefficient is determined based on experimental observation data and represents the additional contribution ratio to the threat level when abnormal micro-motions are present.

[0072] S406. Obtain the real-time wind speed, and calculate the environmental factors based on the real-time wind speed and the reference wind speed. The specific operation steps are as follows: The system obtains real-time wind speed data W from meteorological sensors installed on poles or corridors via a data interface, and sets a reference wind speed. The wind speed is set to 10 m / s, and a wind speed influence coefficient is also set. The wind speed influence coefficient is 0.2, which is derived from a simplified model of wind pressure and offset, and the conductor state factor is calculated comprehensively. .

[0073] S407. Using the ratio of the basic warning distance to the shortest spatial distance as a base, multiply it by the object state factor, the conductor state factor, and the environmental factor to obtain the comprehensive threat index. The specific operation steps are as follows: Taking into account distance, object behavior, conductor response, and environmental conditions, a product model is used to reflect the multi-factor coupling amplification effect and calculate the comprehensive threat index. : ; in, This represents the actual distance between the object and the wire; The value is 1 when the Boolean value of the abnormal micro-motion flag M is true, and 0 when the Boolean value is false.

[0074] S408. Compare the calculated comprehensive threat index H with the first threshold and the second threshold to determine the threat level L; The first threshold is set to 1, and the second threshold is set to 1.5. It should be noted that the first and second thresholds were optimized by retrospectively analyzing a large amount of historical early warning event data, while ensuring a low false alarm rate. The judgment logic is as follows: if the comprehensive threat index is less than the first threshold, then the threat level is judged to be safe. If the first threshold is less than or equal to the comprehensive threat index and less than the second threshold, the threat level is determined to be a yellow alert. If the overall threat index is greater than or equal to the second threshold, the threat level is determined to be a red alert.

[0075] It should be noted that the solution lies in overcoming the limitations of the traditional single distance threshold method, shifting the focus from whether the distance is exceeded to the extent of the risk, thereby significantly improving the accuracy and situational adaptability of early warnings, and effectively balancing false alarms and missed alarms.

[0076] In an embodiment of the present invention, step S500 involves performing time-domain analysis on threat warning data to generate a risk prediction report, including the following steps S501-S504: S501. Based on historical risk warnings of the transmission line corridor, obtain continuously updated data streams, as follows: Each warning record contains the following fields: warning trigger time, high-risk object that triggered the warning, threat level, wind speed at the time of the warning event, and weather conditions.

[0077] S502. Delineate the grid for historical risk warning data, calculate the density of warning events within each grid, and merge adjacent grids as risk hotspot areas. The specific operation steps are as follows: Extract the set of all threat level (yellow alert) events within a specified historical period (e.g., the past 30 days) from historical risk warnings. Each event includes its geographic coordinates. Spatial clustering is performed using a grid- and density-based method. Specifically: S5021. Divide the geographical area of ​​the entire transmission line corridor into a two-dimensional grid of a fixed size. In an optional embodiment, the fixed size may be 50 meters. 50 meters.

[0078] S5022. Count the number of warning events falling into each two-dimensional grid, and use it as the original event density of the current two-dimensional grid.

[0079] S5023. Set the spatial density threshold. It should be noted that the spatial density threshold is initially set by sorting the original event densities of all historical grids and taking the 90th percentile value.

[0080] S5024. Identify all grids whose original event density exceeds the spatial density threshold and mark them as high-density grids.

[0081] S5025. Merge adjacent high-density grids that share edges or corners to form connected risk hotspot regions, and calculate the geometric center and region boundary of each hotspot region.

[0082] S503. For each identified risk hotspot area, conduct periodic and trend analyses. The specific steps are as follows: S5031. Periodic analysis specifically involves re-aggregating event sequences along two or more time dimensions: aggregation by hour to generate a 24-hour distribution, aggregation by day of the week to generate a distribution from Monday to Sunday, and aggregation by month and day to identify specific dates.

[0083] For each dimension of the distribution, statistical characteristics are calculated: peak periods and average frequency of events. For example, it is calculated that for any risk hotspot area on a given day, the event frequency is significantly higher from Monday to Friday than on weekends.

[0084] If, in a certain dimension of the distribution, there are three or more consecutive time units where the event frequency exceeds 150% of the average frequency of the current dimension, and the current pattern repeats at least three times in four consecutive cycles, then it is determined that there is an activity peak in the current risk hotspot area, and the peak period is recorded.

[0085] S5032. Trend Analysis Specifically, it involves calculating the total number of events and the average threat level in current risk hotspot areas on a weekly basis.

[0086] Furthermore, linear regression analysis was performed on the total number of events and the average threat level sequence for the most recent N weeks, respectively, and the slope of the regression line for each sequence was calculated to represent the frequency trend slope and the threat level trend slope. It should be noted that the slope of the regression line is constructed by creating a two-dimensional sequence from the data of the most recent N weeks. For the frequency sequence, the independent variable is the week number and the dependent variable is the frequency of events in that week; for the rank sequence, the independent variable uses the same week number and the dependent variable is the average threat rank in that week. Calculate the arithmetic mean of all values ​​in the independent variable sequence and the arithmetic mean of all values ​​in the dependent variable sequence, respectively. The independent variable deviation is calculated by subtracting the arithmetic mean of the independent variables from the independent variables. Similarly, the dependent variable deviation is calculated by calculating the product of the independent variable deviation and the dependent variable deviation for each data point in turn. All N product results are added together to obtain the total product sum. Calculate the square of the independent variable deviation for each data point in turn, and sum all N squared results to obtain the sum of squares of the independent variable deviations.

[0087] The ratio of the total sum of products to the sum of squares of the deviations of the independent variables is used as the slope of the linear regression.

[0088] Furthermore, if the slope of the frequency trend is greater than the frequency threshold, then the frequency of warning events is considered to be on the rise. If the slope of the risk level trend is greater than the risk level threshold, the risk severity is considered to be on the rise.

[0089] Among them, the frequency threshold and the grade threshold are determined by statistical analysis of data from historically stable regions, with the 75th percentile of the trend slope being used as the default value.

[0090] S504. Perform spatial analysis on the aforementioned risk hotspot areas, infer the causes of the risks, and generate a risk prediction report. The specific steps are as follows: The polygonal boundaries of each risk hotspot area are spatially overlaid with external geographic information system (GIS) data layers for analysis; these external GIS data layers include, but are not limited to, urban planning maps, construction permit areas, road networks, and vegetation growth areas. S5041. Calculate the spatial relationship between the current risk hotspot area and various GIS elements, specifically: For areal features: calculate the geometric intersection between the risk hotspot area and the current GIS areal feature polygon. After the calculation, calculate the area of ​​the intersection part, and then divide the current intersection area by the total area of ​​the risk hotspot area itself to obtain an overlap area ratio.

[0091] For linear features: Calculate the geometric center point of the polygon in the risk hotspot area, and calculate the perpendicular distance from the center point to the specified linear feature, which is the shortest distance from the point to the line.

[0092] For point features: Use the ray method to judge each relevant GIS point and check whether its coordinates fall within the polygon boundary of the risk hotspot area.

[0093] S5042. Establish an association rule base for matching, and match all spatial relationships in the current risk hotspot area with the impact characteristics of high-risk objects one by one.

[0094] Based on S5032, it should be noted that in response to regions where the slope of the frequency trend is greater than the frequency threshold, the total number of events in the most recent 4 weeks is used to calculate the weighted moving average as the predicted frequency for the next week, and the weighting coefficient increases as time approaches. For areas with peak activity periods, the current peak period will be marked as a high-risk period for the next cycle. In response to regions where the frequency trend slope is greater than twice the frequency threshold, the activity expansion direction is determined by analyzing the main direction of the event coordinate sequence. Along the activity expansion direction, the boundary of the risk hotspot area is extended outward by a buffer distance, wherein the buffer distance is manually defined. All the obtained data is pushed to the operation and maintenance terminal for storage, for future use.

[0095] Example 3: This example also provides an electronic device applicable to a method for detecting the state of high-risk objects in a power transmission line corridor, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the method for detecting the state of high-risk objects in a power transmission line corridor as proposed in the above examples.

[0096] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements a method for detecting the status of high-risk objects in a power transmission line corridor as proposed in the above embodiments.

[0097] The storage medium proposed in this embodiment belongs to the same inventive concept as the method for detecting the state of high-risk objects in a power transmission line corridor proposed in the above embodiments. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0098] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0099] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for detecting the status of high-risk objects in a power transmission line corridor, characterized in that: include, Analysis of the impact characteristics of high-risk objects based on image and video data of the power transmission line corridor; Based on the aforementioned impact characteristics, high-risk objects are spatially labeled to construct a high-risk object resource database; High-risk object identification is performed on video data, and conductor status change identification is performed on image data; Based on the identification results, determine whether to issue a high-risk object threat warning for the power transmission line corridor; Perform time-domain analysis on threat warning data to generate risk prediction reports.

2. The method for detecting the status of high-risk objects in a power transmission line corridor as described in claim 1, characterized in that, The analysis of the impact characteristics of high-risk objects includes: Obtain the first and second databases of the transmission line corridor; Retrieve suspected causative image sequences from the first database; Preprocessing of suspected causative image sequences; Moving target detection and trajectory tracking are performed on the preprocessed suspected causative image sequence to extract target behavior image fragments; The system receives descriptive information from target behavior image fragments as input, representing the impact characteristics of high-risk objects.

3. The method for detecting the status of high-risk objects in a power transmission line corridor as described in claim 2, characterized in that, The construction of the high-risk object resource database includes: Images that meet the aforementioned influence characteristics are selected as samples, and the sample images are aligned with a pre-constructed 3D scene model; Adjust the 3D placeholder frame of the object in the 3D scene model to match the 2D outline of the object in the sample image; Calculate the minimum three-dimensional spatial distance between the three-dimensional coordinates of the overlapping object and the three-dimensional coordinates of the guide wire, and construct a high-risk object resource database.

4. The method for detecting the status of high-risk objects in a power transmission line corridor as described in claim 3, characterized in that, The high-risk object identification includes: Foreground extraction is performed on real-time monitoring images to obtain candidate regions, and features of the candidate regions are extracted. The features are compared with those in the high-risk object resource database to generate a preliminary category hypothesis. For each of the primary category hypotheses, perform Level 1, Level 2, and Level 3 verifications sequentially. The fusion confidence level of each candidate region corresponding to each category is calculated based on the validation results at each level. When a candidate region whose fusion confidence exceeds the global determination threshold is identified as a high-risk object, a suspected threat region is defined with the high-risk object as the center.

5. The method for detecting the status of high-risk objects in a power transmission line corridor as described in claim 4, characterized in that, The identification of changes in conductor state includes: Maintain the data of the conductor to obtain the pixel-level centerline; Based on pixel-level centerlines, a non-interference-resistant strip search strategy is used to extract wires and obtain sub-pixel precision centerlines. The unilateral visual tension index of the current conductor segment is calculated based on the subpixel precision centerline and real-time data. The unilateral visual tension index is compared with a dynamic threshold to determine whether there is a change in the state of the conductor.

6. The method for detecting the status of high-risk objects in a power transmission line corridor as described in claim 5, characterized in that, The identification results include: Calculate the shortest spatial distance from the high-risk object to the conductor, query the corresponding impact characteristics of the high-risk object, and obtain the basic warning distance; Determine the object state factor based on the behavioral state identifier of high-risk objects; Calculate the conductor state factor based on the results of conductor state change judgment; Obtain real-time wind speed and calculate environmental factors based on the real-time wind speed and reference wind speed; The comprehensive threat index is obtained by multiplying the ratio of the basic warning distance to the shortest spatial distance as the base, and then multiplying it by the object state factor, the conductor state factor, and the environmental factor.

7. The method for detecting the status of high-risk objects in a power transmission line corridor as described in claim 6, characterized in that, The determination of whether to issue a high-risk object threat warning for the transmission line corridor includes: The comprehensive threat index is compared with the first threshold and the second threshold; When the comprehensive threat index is less than the first threshold, it is determined to be safe; When the comprehensive threat index is greater than or equal to the first threshold and less than the second threshold, it is determined to be a yellow warning level; When the comprehensive threat index is greater than or equal to the second threshold, it is determined to be a red alert level.

8. The method for detecting the status of high-risk objects in a power transmission line corridor as described in claim 7, characterized in that, The time-domain analysis includes: Based on historical risk warnings of transmission line corridors, obtain continuously updated data streams; Define a grid for historical risk warning data, calculate the density of warning events within each grid, and merge adjacent grids as risk hotspot areas; For each identified risk hotspot area, periodic and trend analyses are conducted. Spatial analysis is performed on the aforementioned risk hotspot areas to infer the causes of the risks and generate a risk prediction report.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method for detecting the state of high-risk objects in a power transmission line corridor as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the method for detecting the state of high-risk objects in a power transmission line corridor as described in any one of claims 1 to 8.

Images