Risk comprehensive assessment system and method based on image acquisition and identification analysis

Abstract

The invention discloses a risk comprehensive assessment system and method based on image acquisition and identification analysis, and relates to the field of power transmission line safety monitoring. Comprising an image acquisition module, a data preprocessing module, a risk identification module, a multi-dimensional risk assessment module and an early warning linkage module. The image acquisition module acquires a multi-view real-time image; the preprocessing module carries out median filtering denoising and Laplacian sharpening; the risk identification module uses a risk identification model to detect water and soil loss, equipment abnormity and environmental risks; the multi-dimensional evaluation module fuses an image recognition result, machine account data, meteorological factors and historical maintenance records, and dynamically adjusts weights through an angular momentum model and trigonometric series to calculate a comprehensive risk score; and the early warning linkage module performs hierarchical response according to the score. According to the scheme, multi-source data fusion, accurate identification and dynamic early warning are realized, and the accuracy and timeliness of risk prevention and control of the power transmission line are improved.

Description

Technical Field

[0001] The present invention relates to the field of power transmission line safety monitoring, and in particular to a comprehensive risk assessment system and method based on image acquisition and recognition analysis. Background Art

[0002] With the rapid development of the national economy, the scale of social electricity consumption and the scale of power lines are also increasing. The phenomenon of crossing transmission line networks is also increasing. Transmission lines often cross railways, highways, etc., which also puts higher requirements on line operation and maintenance.

[0003] Soil erosion and defects in the foundation of transmission lines are important reasons for the destruction of the crossing point structure. At the current stage, we have not fully grasped the information on various influencing factors of the "three spans and two adjacent" of the line. When soil erosion and major defects in the tower body occur, we cannot quickly discover and further assess the relevant risks, and cannot prepare corresponding fault handling plans in advance; when the line is shut down due to faults, we cannot promptly understand the impact on the operation of the crossed lines, roads and railways or nearby oil and gas stations; when the line fault causes the electric railway to stop operating, we have not established contact with the relevant electric railway dispatchers. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a risk comprehensive assessment system and method with greater timeliness and reliability.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0006] A comprehensive risk assessment system based on image acquisition and recognition analysis, the key of which is that it includes an image acquisition module, a data preprocessing module, a risk identification module, a multi-dimensional risk assessment module and an early warning linkage module;

[0007] The image acquisition module collects real-time image information and transmits it to the data preprocessing module. After processing, the risk identification module identifies the risk, and after the risk is determined, the multi-dimensional risk assessment module assesses it. Finally, the early warning linkage module takes action based on the risk assessment results.

[0008] The image acquisition module acquires multi-view real-time image data of the "three spans and two adjacent" points of the transmission line by combining multiple data sources including drones, fixed cameras and manual photography;

[0009] The image preprocessing module preprocesses the collected image, including median filtering denoising and Laplace operator sharpening processing;

[0010] The risk identification module constructs a risk identification model, performs target detection and positioning on the image, and obtains risk identification results through comparative analysis;

[0011] The multi-dimensional risk assessment module generates a comprehensive risk score by integrating image recognition risks, ledger data, meteorological factors and historical maintenance records, and then performing weighted calculations;

[0012] The early warning linkage module determines the risk level through the comprehensive risk score and takes alarm action.

[0013] Preferably, the risk identification model includes a soil and water loss detection model and an equipment anomaly and environmental risk detection model;

[0014] The soil and water loss detection model uses the ResNet backbone network to extract image features, uses the proposal network to generate proposal boxes, and realizes target detection and positioning of soil and water loss areas through ROIPooling and classifiers;

[0015] The equipment anomaly and environmental risk detection model adopts the CSPDarkNet backbone network, PAN-FPN feature fusion and deformable convolution to achieve rapid positioning and identification of shock-absorbing hammer distance anomalies and external breaking machine intrusion targets.

[0016] Preferably, the multi-dimensional risk assessment module scores the image recognition results, ledger data, meteorological factors and historical maintenance records, and assigns corresponding weights for weighted calculation to obtain a comprehensive risk score. The early warning linkage module acts on the comprehensive risk score according to the score classification. When the early warning threshold is exceeded, an alarm action is triggered, and the external system is connected to push the alarm information.

[0017] Preferably, the ledger data includes the tower form, basic building form, slope angle and geological exploration results; the meteorological factors include wind speed, rainfall and lightning activity frequency; the historical maintenance records are scored by combining time and maintenance frequency.

[0018] A comprehensive risk assessment method based on image acquisition and recognition analysis, the key of which is to include the following steps:

[0019] (1) Image collection stage: Multi-source equipment is used to collect images of the “three spans and two adjacent” points of the transmission line;

[0020] (2) Image preprocessing stage: the image is subjected to median filtering denoising and Laplace sharpening preprocessing in sequence;

[0021] (3) Risk identification stage: The pre-processed images are input into the soil and water loss detection model to detect soil and water loss areas and generate risk levels; the pre-processed images are input into the equipment anomaly and environmental risk detection model to detect equipment anomalies and environmental risks and generate detection results;

[0022] (4) Comprehensive scoring stage: integrating image detection results, ledger data, meteorological data and historical maintenance records to calculate the comprehensive risk score;

[0023] (5) Early warning stage: Implement graded early warning according to risk scores and coordinate with relevant departments for pre-emptive control.

[0024] Preferably, the median filter denoising is performed by sorting the grayscale values ​​of the pixel area through an odd-sized template and taking the median value, and the formula is:

[0025] g(x,y)=Med{f(xk,yl)∣(k,l)∈W}

[0026] Where W is the domain window, f(xk,yl) is the pixel value in the domain;

[0027] The Laplace sharpening enhances the edge by calculating the two-dimensional second-order partial differential. The specific formula is:

[0028] ▽2f=f(x+1,y)+f(x-1,y)+f(x,y+1)+f(x,y-1)-4f(x,y).

[0029] Preferably, the fusion image recognition risk, ledger data, meteorological factors and historical maintenance records are weighted to calculate a comprehensive risk score, where:

[0030] Construct a risk numerical calculation model based on angular momentum,

[0031] L k (t)=‖r k (t)‖·‖p k (t)‖·cosθ k

[0032] Where ‖r2(t)‖ is the modulus of the position vector, which is also the scalar representation of the spatial risk value;

[0033] ‖p2(t)‖ is the momentum vector modulus and is also the scalar representation of the dynamic trend value;

[0034] θ k : Evaluate the synergistic effect of different risk factors and calculate the correlation between ledger data and meteorological factors through cosine similarity.

[0035] The trigonometric series is used to dynamically adjust the weights of meteorological factors, and the corresponding weights are automatically increased in scenarios such as typhoon season and rainy season. The specific formula is as follows:

[0036]

[0037] Basic weight w 30 Associated with the historical mean of meteorological factors; high-frequency terms (the larger n is) correspond to short-term severe convective weather, and the weight correction term ∑An cos(·) is directly added to the base weight.

[0038] A correction factor is set in the calculation of the comprehensive score. When the image recognition confidence exceeds the threshold, high-risk weather or equipment damage occurs, the weight of the corresponding dimension is increased. The comprehensive score formula is:

[0039]

[0040] L k (t) is the cumulative value of each dimension, W i′ is the dynamic weight, λ k is the correction factor.

[0041] Preferably, the early warning linkage module responds in a graded manner based on the risk score and the speed of change, and issues an alarm when the risk change in a single dimension exceeds a threshold but the comprehensive risk does not exceed a threshold.

[0042] The beneficial effects of adopting the above technical solution are:

[0043] The image acquisition module of the present invention integrates multi-source data from drones, fixed cameras and manual photography to achieve full-time, all-weather and multi-perspective monitoring coverage of the "three spans and two adjacent" points of the transmission line, thereby improving the timeliness and reliability of the data.

[0044] The present invention adopts a comprehensive risk assessment with multi-dimensional dynamic weight adjustment, which makes the monitoring of emergencies more sensitive and adaptively adjusts the impact of meteorological factors on the line according to the season, thereby improving the reliability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0046] Figure 1 This is a schematic diagram of the structure of a comprehensive risk assessment system based on image acquisition and recognition analysis proposed by the present invention;

[0047] Figure 2 It is a flow chart of a soil and water loss detection model constructed in a comprehensive risk assessment system based on image acquisition and recognition analysis proposed by the present invention. DETAILED DESCRIPTION

[0048] To make the above-mentioned objectives, features, and advantages of the present invention more clearly understood, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings and specific implementation methods of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0049] A comprehensive risk assessment system based on image acquisition and recognition analysis, such as Figure 1 This system includes image acquisition module, data preprocessing module, risk identification module, multi-dimensional risk assessment module and early warning linkage module;

[0050] The image acquisition module receives image information through external devices and transmits it to the data preprocessing module. After the image data is clarified by the data preprocessing module, the risk identification module performs risk identification and transmits the risk information to the multi-dimensional risk assessment module. The risk numerical assessment is performed based on the multi-dimensional data, and the early warning linkage module performs graded actions based on the risk numerical value.

[0051] A comprehensive risk assessment method based on image acquisition and recognition analysis,

[0052] (1) Image collection stage: Multi-source equipment is used to collect images of the “three spans and two adjacent” points of the transmission line;

[0053] (2) Image preprocessing stage: the image is subjected to median filtering denoising and Laplace sharpening preprocessing in sequence;

[0054] (3) Risk identification stage: The pre-processed images are input into the soil and water loss detection model to detect soil and water loss areas and generate risk levels; the pre-processed images are input into the equipment anomaly and environmental risk detection model to detect equipment anomalies and environmental risks and generate detection results;

[0055] (4) Comprehensive scoring stage: integrating image detection results, ledger data, meteorological data and historical maintenance records to calculate the comprehensive risk score;

[0056] (5) Early warning stage: Implement graded early warning according to risk scores and coordinate with relevant departments for pre-emptive control.

[0057] The image acquisition module obtains multi-perspective real-time image data of the "three spans and two adjacent" points by combining multi-source data such as drones, fixed cameras, and manual photography, covering monitoring needs in different time periods and weather conditions.

[0058] The data preprocessing module performs median filtering denoising and Laplace operator sharpening on the collected images in sequence, providing high-quality image information for subsequent image processing.

[0059] Median filter denoising: Use an odd-sized template (such as 3×3, 5×5) to sort the grayscale values ​​of the pixel neighborhood and take the median. The formula is:

[0060] g(x,y)=Med{f(xk,yl)∣(k,l)∈W}

[0061] Where W is the domain window, f(xk, yl) is the pixel value in the domain; at the same time, in order to more easily find the center value, the total number of pixels contained in the W domain is generally an odd value.

[0062] Laplace sharpening enhances edges by calculating the two-dimensional second-order partial differential. The specific formula is:

[0063] ▽2f=f(x+1,y)+f(x-1,y)+f(x,y+1)+f(x,y-1)-4f(x,y);

[0064] This method emphasizes sudden changes in grayscale in the image and deemphasizes slowly changing areas of the image. Therefore, some gradient light gray edge lines will become the background color of the image outline. Then, the original image and the sharpened image are superimposed, maintaining the original image while still showing the enhanced edge image. This achieves image sharpening, enhances the contrast between the target and the background, and improves the accuracy of subsequent target detection. This solves the problem of partially blurred samples in image data.

[0065] The risk identification module builds a dual-model architecture to achieve accurate detection, including a soil and water loss detection model and an equipment anomaly and environmental risk detection model.

[0066] like Figure 2 Soil and water loss detection model: Based on the ResNet backbone network, image features are extracted, candidate regions are generated through the proposal network (RPN), and target detection and positioning of soil and water loss areas are achieved through ROIPooling and classifiers. It is suitable for monitoring slowly changing terrain environments. The specific process is as follows:

[0067] The model uses ResNet as the backbone network, extracting features from the input image through convolution or pooling operations to obtain feature extraction results. A region proposal network (RPN) is introduced to propose boxes that match the number of anchor boxes at each feature map location in the original image and determine whether the proposed box area is the foreground. In the RPN, a 3×3 convolution first performs a sliding scan on the feature map to obtain semantic information about the corresponding receptive region in the original image. Then, two 1×1 convolutions are used to obtain category regression information and position regression information for the foreground and background of each proposed box, respectively.

[0068] The loss function of FasterR-CNN consists of two parts: the category loss function and the position loss function between the RPN proposal box and the real box. The first part of the loss function is:

[0069]

[0070] in, is the loss function for the proposed box category, is the position loss function, i is the order number of the proposed box, Ncls The same size as the batch size during batch training, N reg The number of proposed boxes is consistent with that proposed by RPN, and λ is the loss weight of the proposed box position. This model can achieve high-precision detection and identification of regional soil erosion.

[0071] At the same time, a multi-resolution topological analysis of regional soil and water loss characteristics based on image recognition was conducted. In a multi-scale, complex space, an anisotropic diffusion analysis model was established to perform information enhancement and multi-module detection on regional soil and water loss remote sensing images, reconstructing a geographic vector field dataset. Using the differences between the original image and the filtered image, edge images were statistically analyzed in blocks, and the grayscale contrast between adjacent regions was calculated. The multi-resolution spectrum parameters of regional soil and water loss characteristics at the (p+q) order are:

[0072]

[0073] According to the spectral parameters of the local area of ​​the target, the distribution of the fuzzy edge feature parameters of the original image and the filtered image is:

[0074]

[0075] By analyzing the spectral characteristics of water body remote sensing SAR images, multi-resolution registration and topological design of regional soil and water loss characteristics are achieved.

[0076] The global directional change information is reorganized and local visualization features are reorganized by combining the regional soil and water loss remote sensing image detection method. The geometric pixel distribution set expression of the water body of the river is obtained as follows:

[0077]

[0078] Where Tm is the pixel intensity value; τ mk is the contrast edge of the connected area; v(t) is the river water parameter, w nk is the characteristic parameter of regional soil erosion in the connected area; K(m) is the large-scale geographic space pixel point; M is the edge pixel set of the connected area.

[0079] Through different resolution texture expressions and scale detection, the distribution probability within each subinterval β and the global distribution probability are obtained as follows:

[0080] minWH=min{w(C),h(V)}

[0081]

[0082] Where w(C) is the intensity information entropy; h(V) is the probability statistical interval function; Area(CC) is the comprehensive abstract characteristic quantity of regional soil and water loss scalar information; Area(pic) is the statistical probability parameter distribution set.

[0083] Based on a certain statistical probability model, remote sensing information enhancement technology is used to enhance the neighborhood of composite information entropy. At this time, the probability statistical interval of soil and water loss feature identification is:

[0084]

[0085] in is the maximum composite information entropy of soil and water loss; X i Identify points of remote sensing features in geographic space; is the minimum composite information entropy of soil and water loss.

[0086] Based on the above analysis, a statistical probability model is established, and the composite information entropy of the geographic vector field is introduced. Through the analysis of the overall change characteristics, the scalar attribute distribution set of regional soil and water loss is obtained as follows:

[0087]

[0088] where x a is the scalar attribute of regional soil erosion; v a It is a comprehensive abstract parameter of scalar information; is the multi-frequency noise vector attribute; It is a multi-resolution dense texture information. Through large-scale dense vector field remote sensing image data detection, through the global direction change information reconstruction, combined with the regional soil and water loss remote sensing image detection method, local visualization feature reconstruction is performed.

[0089] Regional water and soil erosion feature recognition output: Combined with geometric primitives and texture visualization methods, remote sensing feature detection of regional water and soil erosion characteristics and regional information reconstruction are performed. Multi-frequency noise is used as input texture to obtain the grayscale pixel value of the regional water and soil erosion remote sensing image.

[0090]

[0091] Where: θ kl is the data sampling parameter of the uniform grid of remote sensing image; θ kl is the phase characteristics of the texture field distribution of the regional soil erosion vector field; x1-(tk),…,x n -(tk) is the output texture value of the regional soil and water loss sampling point; is the composite information entropy of the geographic vector field. Taking the vector direction distribution probability as the input feature quantity and the field strength as the intensity information entropy, the statistical probability function is:

[0092]

[0093] in: is the composite information entropy of the geographic vector field of regional soil erosion; is the soil and water loss entropy synthesis coefficient; is the composite parameter of geographic vector field; is the distribution probability of the soil and water loss vector direction. Based on the above analysis, the composite information entropy of the geographic vector field is extracted and the regional soil and water loss characteristics are identified through remote sensing image analysis.

[0094] Equipment Anomaly and Environmental Risk Detection Model: Taking the anti-vibration hammer distance detection as an example, this model utilizes the CSPDarkNet backbone network, PAN-FPN feature fusion, and deformable convolution to rapidly locate and identify targets such as anti-vibration hammer distance anomalies and external mechanical intrusion, improving detection capabilities for small and deformed targets. It also utilizes the convolutional neural network from the YOLOv8 series of models. YOLOv8 consists of a Backbone network and a Head network. The Backbone network uses the CSPDarkNet architecture for feature extraction, while the Head network utilizes PAN-FPN feature fusion to enhance detection capabilities for targets of varying scales. The Detect module utilizes a decoupled head design to improve detection accuracy. Furthermore, it employs an Anchor-Free object detection method to enhance the flexibility and accuracy of detecting irregularly shaped targets.

[0095] To address the poor performance of traditional convolution in handling irregular defects, deformable convolution was introduced. By adding a two-dimensional offset to the positions of traditional convolutional networks, the sampling points can be freely deformed, improving the network's adaptability to irregular object shapes. The deformable convolution computational process is as follows: first, an additional convolution operation is performed to calculate the offset of each element in each convolution window. The convolution kernel then adjusts the sampling point position on the input image based on the offset value. Finally, the feature map of the adjusted sampling point position is convolved to obtain the final output.

[0096] The multi-dimensional risk assessment module builds a risk numerical calculation based on angular momentum and a dynamic weight adjustment analysis model based on trigonometric series.

[0097] First, we conduct scalar angular momentum modeling for risk numerical calculation. Combined with the characteristics of risk accumulation in transmission lines, we regard transmission line risk as a comprehensive reflection of "spatial location" and "change trend". The cumulative risk value is calculated through the correlation between the three:

[0098] Spatial Risk Value: Calculates the inherent risk of the area where the tower is located based on its geographic location and environmental parameters. In practice, towers located on steep slopes (α > 45°) and at higher altitudes have significantly higher spatial risk values ​​than towers located in plain areas.

[0099] Dynamic trend value: Analyze the time variation of historical data, pay more attention to the increase in the number of maintenance in a short period of time, and compare recent data with the long-term average frequency to increase the importance of new data.

[0100] Factor correlation: Evaluate the synergistic effect of different risk factors. When heavy rain weather and high-gradient slopes are combined, the risk of soil erosion increases by more than 50% compared with a single factor. The correlation between the two is judged to be "highly correlated", which is more conducive to three-dimensional risk analysis.

[0101] Comprehensive cumulative value: Multiply the spatial risk value, dynamic trend value, and factor correlation to obtain the cumulative risk value for each dimension, including image recognition, meteorological factors, records, and historical maintenance. Specifically, if a tower has a spatial risk value of 80 points, a dynamic trend value of 70 points, and a correlation with heavy rain of 90%, the cumulative value for the meteorological dimension can be expressed as 80 × 70 × 90% = 5040.

[0102] The specific formula is:

[0103] L k (t)=‖r k (t)‖·‖p k (t)‖·cosθ k

[0104] Where ‖r2(t)‖ is the modulus of the position vector, which is also the scalar representation of the spatial risk value; ‖p2(t)‖ is the modulus of the momentum vector, which is also the scalar representation of the dynamic trend value; θ k : Evaluate the synergistic effect of different risk factors and calculate the correlation between ledger data and meteorological factors through cosine similarity.

[0105] In different dimensions, they have different practical meanings.

[0106] 1) Image recognition risk dimension

[0107] Spatial position eigenvalues: Based on the area of ​​the risk region detected by the image, including the proportion of the soil erosion area and the equipment and environmental risk detection model, the eigenvalues ​​change when the soil erosion area, the distance to the shock absorber, and the area or distance of the mechanical intrusion limit change;

[0108] Trend change characteristic value: Compare the risk area or distance change rate between two consecutive detections, and the characteristic value will change accordingly;

[0109] Factor-related characteristic value: If the detection area also has a high vegetation coverage rate (to inhibit soil and water loss), the related characteristic value will be reduced according to the vegetation coverage rate.

[0110] 2) Meteorological risk dimension

[0111] Spatial location characteristic value: based on the meteorological sensitivity level of the area where the transmission line is located;

[0112] Trend change characteristic value: the increase in real-time wind speed, rainfall and other meteorological factors compared with the historical average;

[0113] Factor correlation eigenvalue: Synchronicity between wind speed and rainfall. In actual situations, when a typhoon is accompanied by heavy rain, the correlation eigenvalue increases accordingly).

[0114] 3) Ledger and maintenance dimensions

[0115] Ledger data: related to the historical average normal ledger data of the current line,

[0116] Trend change characteristic value: associated with the recent rate of change of ledger data;

[0117] The associated characteristic values ​​are associated with historical fault events;

[0118] Maintenance records: associated with the historical average maintenance frequency;

[0119] Trend change characteristic value: Take "maintenance frequency in the past three months" and decrease if it decreases and increase if it increases;

[0120] The correlation characteristic value is associated with the average failure rate of similar equipment.

[0121] Then, based on the seasonal periodicity of meteorological factors, a dynamic weight adjustment based on trigonometric series was established to automatically adjust their importance in risk assessment. Scalar weighting was used instead of vector operations to achieve a three-dimensional risk identification effect:

[0122] Basic weight: By default, meteorological factors (wind speed, rainfall, etc.) have a fixed weight in the comprehensive assessment. This value is set based on the average impact of meteorological factors in historical accidents.

[0123] Cycle modulation weight:

[0124] Rainy season: The system identifies through historical data that rainfall during this period is closely related to soil erosion, and automatically increases the rainfall weight by 25%. At the same time, it reduces the weight of routine inspections to ensure that the total weight remains unchanged, giving priority to risks caused by weather.

[0125] Typhoon season: When a typhoon warning is detected, the wind speed weight is temporarily increased and the "high-frequency risk response mode" is triggered, which means real-time monitoring of wind speed data rather than conventional frequency monitoring.

[0126] The specific formula is:

[0127]

[0128] Basic weight w 30Associated with the historical mean of meteorological factors; high-frequency terms (the larger n is) correspond to short-term severe convective weather, and the weight correction term ∑A n cos(·) is directly added to the base weight.

[0129] Finally, the cumulative risk values ​​of the four dimensions of image recognition, ledger data, meteorological factors, and historical maintenance records are multiplied by the dynamically adjusted weights and summed. After normalization, a comprehensive risk score of 0-100 is obtained:

[0130]

[0131] λ k Correction factor for emergency adjustments

[0132] Image recognition dimension: If the confidence level of soil erosion and environmental factor anomalies exceeds a high threshold, the correction factor is increased;

[0133] Meteorological factor dimension: When high-risk weather occurs, the correction factor increases;

[0134] Ledger and historical maintenance record dimension: When equipment damage is found during maintenance, the correction factor is increased.

[0135] The early warning linkage module makes real-time judgments and graded responses to risk trends.

[0136] First, we conduct a two-dimensional analysis of the risks. Based on the current risk score and the rate of change of the score, we build a two-dimensional analysis framework and classify the risk status into four categories:

[0137] Quadrant 1 (high score + rapid increase): If the comprehensive score is 85 points and increases by 15 points per hour, it is judged as a red alert, indicating that the risk is out of control and an immediate power outage and maintenance and activation of the emergency plan are required.

[0138] Second quadrant (low score + rapid increase): If the comprehensive score is 40 points but increases by 20 points per hour, it is judged as an orange warning, indicating that the risk may rebound and intensified monitoring and deployment of emergency supplies are required within 4 hours.

[0139] The fourth quadrant (high score + slow decline): If the comprehensive score is 75 points but decreases by 5 points per hour, it is judged as a yellow warning and requires a routine inspection once a day to continuously monitor the residual risk.

[0140] Quadrant III (low score + slow decline): If the overall score is 30 points and decreases by 3 points per hour, it is judged to be in a green state, the risk is within a controllable range, and normal operation and maintenance are maintained.

[0141] Furthermore, the system monitors the rate of change of the cumulative risk value of each dimension in real time. Even if the comprehensive score does not reach the warning threshold, if a dimension suddenly accelerates (for example, the tower foundation settlement rate increases by 30% within 2 hours), an alert upgrade will be triggered:

[0142] Detection logic: Compare the risk accumulation values ​​at adjacent time points (such as the current value and the value 2 hours ago). If the change exceeds the historical average fluctuation range (such as the preset value of 20%), it is judged as a "trend mutation".

[0143] Application example: The cumulative risk value of the inventory data of a tower (reflecting the stability of the foundation) was 50 points at 8:00 a.m. and suddenly rose to 70 points at 10:00 a.m., a change of 40%, exceeding the preset threshold of 20%. The system immediately issued an orange warning and arranged personnel for on-site inspection in advance to avoid waiting for the comprehensive score to exceed the standard before responding.

[0144] The above description of the disclosed embodiments will enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to be embodied in the widest possible manner consistent with the principles and novel features disclosed herein.

Claims

1. A comprehensive risk assessment system based on image acquisition and recognition analysis, characterized in that: It includes image acquisition module, data preprocessing module, risk identification module, multi-dimensional risk assessment module and early warning linkage module; The image acquisition module collects real-time image information and transmits it to the data preprocessing module. After processing, the risk identification module identifies the risk, and after the risk is determined, the multi-dimensional risk assessment module assesses it. Finally, the early warning linkage module takes action based on the risk assessment results. The image acquisition module acquires multi-view real-time image data of the "three spans and two adjacent" points of the transmission line by combining multiple data sources such as drones, fixed cameras, and manual photography. The image preprocessing module preprocesses the collected image, including median filtering denoising and Laplace operator sharpening processing; The risk identification module constructs a risk identification model, performs target detection and positioning on the image, and obtains risk identification results through comparative analysis; The multi-dimensional risk assessment module generates a comprehensive risk score by integrating image recognition risks, ledger data, meteorological factors and historical maintenance records, and then performing weighted calculations; The early warning linkage module determines the risk level through the comprehensive risk score and takes alarm action.

2. The comprehensive risk assessment system based on image acquisition and recognition analysis according to claim 1 is characterized in that: The risk identification model includes a soil and water loss detection model and an equipment anomaly and environmental risk detection model; The soil and water loss detection model uses the ResNet backbone network to extract image features, uses the proposal network to generate proposal boxes, and realizes target detection and positioning of soil and water loss areas through ROIPooling and classifiers; The equipment anomaly and environmental risk detection model adopts the CSPDarkNet backbone network, PAN-FPN feature fusion and deformable convolution to achieve rapid positioning and identification of shock-absorbing hammer distance anomalies and external breaking machine intrusion targets.

3. The comprehensive risk assessment system based on image acquisition and recognition analysis according to claim 1 is characterized in that: The multi-dimensional risk assessment module scores image recognition results, ledger data, meteorological factors, and historical maintenance records, and assigns corresponding weights for weighted calculation to obtain a comprehensive risk score. The early warning linkage module acts on the comprehensive risk score according to the score classification. When the early warning threshold is exceeded, an alarm action is triggered, and the external system is connected to push the alarm information.

4. The comprehensive risk assessment system based on image acquisition and recognition analysis according to claim 1 is characterized in that: The ledger data includes the tower type, basic building type, slope angle and geological exploration results; the meteorological factors include wind speed, rainfall and lightning activity frequency; the historical maintenance records are scored by combining time and maintenance frequency.

5. A comprehensive risk assessment method based on image acquisition and recognition analysis, the method being accomplished by means of a system according to any one of claims 1 to 4, characterized in that: The following steps are involved: (1) Image collection stage: Multi-source equipment is used to collect images of the "three spans and two adjacent" points of the transmission line; (2) Image preprocessing stage: the image is subjected to median filtering denoising and Laplace sharpening preprocessing in sequence; (3) Risk identification stage: the preprocessed images are input into the soil erosion detection model to detect the soil erosion areas and generate risk levels; The pre-processed image is input into the device anomaly and environmental risk detection model to detect device anomalies and environmental risks and generate detection results; (4) Comprehensive scoring stage: integrating image detection results, ledger data, meteorological data and historical maintenance records to calculate the comprehensive risk score; (5) Early warning stage: Implement graded early warning according to risk scores and coordinate with relevant departments for pre-emptive control.

6. The comprehensive risk assessment method based on image acquisition and recognition analysis according to claim 5 is characterized in that: The median filter denoising method uses an odd-sized template to sort the grayscale values ​​of the pixel area and take the median value. The formula is: g(x,y)=Med{f(xk,yl)∣(k,l)∈W} Where W is the domain window, f(xk,yl) is the pixel value in the domain; The Laplace sharpening enhances the edge by calculating the two-dimensional second-order partial differential. The specific formula is: ▽2f=f(x+1,y)+f(x-1,y)+f(x,y+1)+f(x,y-1)-4f(x,y).

7. The comprehensive risk assessment method based on image acquisition and recognition analysis according to claim 5 is characterized in that: The fusion image recognition risk, ledger data, meteorological factors and historical maintenance records are weighted to calculate the comprehensive risk score, where: Construct a risk numerical calculation model based on angular momentum, L k (t)=‖r k (t)‖·‖p k (t)‖·cosθ k Where ‖r2(t)‖ is the modulus of the position vector, which is also the scalar representation of the spatial risk value; ‖p2(t)‖ is the momentum vector modulus and is also the scalar representation of the dynamic trend value; θ k : Evaluate the synergy of different risk factors and calculate the correlation between ledger data and meteorological factors through cosine similarity; The trigonometric series is used to dynamically adjust the weights of meteorological factors, and the corresponding weights are automatically increased in scenarios such as typhoon season and rainy season. The specific formula is as follows: Basic weight w 30 Associated with the historical mean of meteorological factors; high-frequency terms (the larger n is) correspond to short-term severe convective weather, and the weight correction term ∑A n cos(·) is directly added to the base weight; A correction factor is set in the calculation of the comprehensive score. When the image recognition confidence exceeds the threshold, high-risk weather or equipment damage occurs, the weight of the corresponding dimension is increased. The comprehensive score formula is: L k (t) is the cumulative value of each dimension, W i′ is the dynamic weight, λ k is the correction factor.

8. The comprehensive risk assessment method based on image acquisition and recognition analysis according to claim 5 is characterized in that: The early warning linkage module responds in a graded manner based on the risk score and the speed of change, and issues an alarm when the risk change in a single dimension exceeds a threshold but the overall risk does not exceed a threshold.

Images