Fire source early warning system, method and equipment based on unmanned aerial vehicle monitoring and storage medium

Abstract

The invention discloses a fire source early warning system, method and equipment based on unmanned aerial vehicle monitoring and a storage medium, and relates to the technical field of safety management, and the system comprises an unmanned aerial vehicle data acquisition module, a secondary grid division module, a small grid risk judgment module, a region comprehensive evaluation module and a grading disposal linkage module. The unmanned aerial vehicle collects multi-source data, and the secondary grid division module dynamically divides regional comprehensive grids and differentiated sub-grids adaptive to equipment density and terrain complexity based on a linear iterative clustering algorithm; the small grid risk judgment module calculates three indexes of equipment thermal anomaly, vegetation flammability and coupling spreading, determines an optimal threshold value through weighting between-class variance and quantifies the risk; and the regional comprehensive evaluation module adopts a PLSR algorithm to accurately calculate a comprehensive risk coefficient and performs graded early warning. According to the invention, dynamic and accurate monitoring of fire source risks is realized, early warning timeliness and pertinence are improved, disposal resource allocation is optimized, and equipment damage, economic loss and personnel safety risks caused by fire disasters are effectively reduced.

Description

Technical Field

[0001] This invention relates to the field of safety management technology, and more specifically to a fire source early warning system, method, device and storage medium based on drone monitoring. Background Technology

[0002] With the rapid development of the new energy industry, photovoltaic power stations, as the core carrier of clean energy, have gradually expanded to areas with complex terrain such as mountains and hills. Mountainous photovoltaic power stations, due to their open terrain and abundant solar resources, have become an important deployment scenario for photovoltaic projects. However, the special environment of these power stations also brings significant fire safety hazards: on the one hand, the rugged terrain of mountainous areas, the dispersed photovoltaic module arrays, and the fact that they are mostly laid along slopes, with long cable lines and dense joints, make them prone to fires caused by equipment aging, lightning strikes, module overheating, and short circuits; on the other hand, the surrounding area of ​​the power station is often accompanied by flammable vegetation such as weeds and shrubs, which are prone to spontaneous combustion during the dry season, and the fire can spread rapidly along the slopes. Furthermore, the inconvenient transportation and scarce fire lanes in mountainous areas make fire fighting far more difficult than in photovoltaic power stations in plains areas. Once the fire gets out of control, it will directly destroy a large number of photovoltaic modules, inverters, and other core equipment, causing huge economic losses, and may also cause ecological damage to the mountainous area and even threaten the safety of people in surrounding villages.

[0003] With the integration of drone technology and sensing technology, drone-based fire monitoring systems for mountain photovoltaic power plants have been initially promoted. Existing systems typically carry visible light cameras, infrared thermal imagers, and temperature sensors. Utilizing the flexibility and mobility of drones, they enable aerial inspections of dispersed photovoltaic modules, cable lines, and surrounding vegetation. Through image recognition and temperature threshold judgment algorithms, they can pinpoint the location of fire sources, preliminarily identify the size of the fire, and transmit early warning information to a backend management platform. This approach addresses, to some extent, the insufficient coverage and delayed response issues of traditional monitoring methods.

[0004] However, for the special scenario of mountain photovoltaic power stations, the existing drone-based fire source early warning system still has key technical shortcomings. The existing system's response plan does not take into account the core characteristics of mountain photovoltaic power stations, such as whether the fire source is close to the dense photovoltaic array area, inverter room, cable trench and other key equipment areas, whether the fire is spreading along the slope, whether it threatens the surrounding flammable vegetation, and the difficulty of access in mountainous terrain and the distribution of fire lanes; it only adopts a uniform alarm push process, which leads to the misallocation of response resources.

[0005] Therefore, how to achieve targeted fire suppression early warning around the fire source is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] In view of this, the present invention provides a fire source early warning system, method, device and storage medium based on drone monitoring to solve the problems existing in the background art.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: A fire source early warning system based on drone monitoring includes: a drone data acquisition module, a secondary grid division module, a small grid risk assessment module, a regional comprehensive assessment module, and a graded response linkage module. The drone data acquisition module collects multi-source data via multi-source sensors carried by the drone, and outputs a multi-source data set after preprocessing. The secondary grid division module, based on the multi-source data set, divides the region into a comprehensive grid using a linear iterative clustering algorithm, and then divides the comprehensive grid into differentiated sub-grids, outputting a comprehensive grid set and a differentiated sub-grid set. The small grid risk assessment module quantifies the risk level of the differentiated sub-grids in the differentiated sub-grid set, outputting a risk level set. The regional comprehensive assessment module calculates the comprehensive risk coefficient of the small grid risk data within the parent superpixel using the PLSR algorithm, assesses the level of the comprehensive risk coefficient based on a preset risk threshold, and issues a corresponding intensity warning based on the assessment level.

[0008] Preferably, the multi-source data collected by the UAV data acquisition module includes: Visible light image data, For infrared thermal imaging data, It is five-band multispectral data. GPS coordinate data, For terrain elevation data, This is meteorological data.

[0009] Preferably, the secondary mesh generation module specifically includes: The regional integrated grid division unit establishes a six-dimensional feature vector based on visible light image data, GPS coordinate data, and terrain elevation data. Initial cluster centers are evenly distributed according to a preset number of parent superpixels. Within a 3×3 pixel range around each initial cluster center, the gradient of neighboring pixels is calculated. The final cluster center is reselected based on the device distribution density. Within a 2S×2S neighborhood of the final cluster center, the comprehensive distance of the pixels is calculated based on color, space, and scene adaptation distance. Pixels are distributed based on the comprehensive distance to obtain the integrated grid of each region. The differentiated sub-mesh partitioning module presets the number of sub-superpixels and dynamically adjusts the number of sub-superpixels according to the device density and terrain complexity within each region's integrated mesh. After calculating the sub-cluster center step size, it follows the pixel allocation logic of the region integrated mesh partitioning unit to obtain the differentiated sub-mesh.

[0010] Preferably, the regional integrated grid division unit specifically includes: Fusion of visible light image data Lab color space, GPS coordinate data The coordinates (x, y) and The slope grade forms a six-dimensional feature vector. ;in, l For brightness, a Channel A b For channel B, Slope grade; initial cluster center step size Where N is the total number of pixels in the monitoring area, and K is the preset number of parent superpixels; Calculate the gradient value within the initial 3×3 pixel neighborhood. Combining the device distribution density weights, the location of the device distribution density weight with the smallest gradient and within the preset device density range is selected as the final cluster center; Overall distance ; in , , m is the maximum color distance constant. i As the final cluster center identifier, j As the identifier of the pixels in the neighborhood of the cluster center, For color distance, For spatial distance, Adjust the distance to suit the scene.

[0011] Preferably, dynamically adjusting the number of sub-superpixels based on the device density and terrain complexity within the integrated grid of each region specifically includes: ; in, For equipment density, For terrain complexity; The preset number of sub-superpixels, To adjust the number of sub-superpixels, For the corresponding weights.

[0012] Preferably, the small grid risk assessment module specifically includes: The label calculation unit is determined, and the equipment thermal anomaly index, vegetation flammability index, and coupling spread index of each differentiated subgrid are calculated as follows: Equipment thermal anomaly index ; in, This is the infrared temperature value. This refers to the mid-to-far infrared reflectance value. Values ​​in the visible light band. The higher the value, the more severe the thermal anomaly. Vegetation flammability index ; in, Values ​​in the near-infrared band. This is the value in the shortwave infrared band. The higher the value, the more flammable the vegetation; Coupling Spread Index ; in, For slope, Let d be the angle between the slope and the direction of fire spread, d be the distance from the center of the fire source, and h be the elevation difference. , As weight, The wind speed influence coefficient, For wind speed, This is the elevation difference suppression coefficient. The higher the value, the greater the risk of spread; The optimal threshold calculation unit iterates through all candidate threshold values ​​within the range of each index value in the label calculation unit; for each candidate threshold value, all sub-superpixels are divided into foreground and background, and the pixel ratio of foreground and background and the mean of foreground and background regions are calculated; device association weights are introduced to calculate the weighted inter-class variance; the weighted inter-class variance corresponding to each candidate threshold value is calculated one by one, and the candidate value with the largest weighted inter-class variance is selected as the optimal threshold for each index; The risk assessment unit calculates a risk score by weighting the indices of the assessment label calculation unit with the optimal thresholds corresponding to the indices of the optimal threshold calculation unit.

[0013] Preferably, the regional comprehensive assessment module includes: The matrix processing unit constructs an explanatory variable matrix using the equipment thermal anomaly index, vegetation flammability index, coupling spread index, risk score, and equipment importance level; and constructs a response variable matrix using the equipment damage loss coefficient and personnel safety risk coefficient; and performs standardization processing on the explanatory variable matrix and the response variable matrix. Component screening unit, initializing load vector and weight vector The load vector is initialized as a standardized explanatory variable matrix. The orientation of the first principal component, the weight vector is initialized based on the loading vector; the scores of the explanatory variables are calculated. and response variable score ,in Weights for response variables, based on the standardized response variable matrix. and Determine the maximum correlation; update the loads and weights. ; Calculate the residual matrix , Using E and F as the new and Iterate through the above steps until the residual variance is less than a preset threshold, and finally determine the three PLS components. The state vector update unit uses the scores of the three PLS components as the base dimension, combined with a meteorological dynamic correction term. Constructing the region state vector The PLS component weights are calculated based on the proportion of explained variance, using the following formula: ,in For the first z The proportion of variance explained by each PLS component The total explained variance proportion; new data collected by the UAV is received at fixed time intervals, and the weighted PLS component score is calculated. Update the weather correction items to The updated state vector is obtained. ; The regional comprehensive risk coefficient calculation unit is based on the updated state vector and combines the equipment importance factor. and meteorological weight Calculate the comprehensive risk coefficient of each region's integrated grid using the following formula: .

[0014] A fire source early warning method based on drone monitoring includes: collecting multi-source data using a drone equipped with multi-source sensors, and outputting a multi-source data set after preprocessing; dividing the multi-source data set into a regional comprehensive grid using a linear iterative clustering algorithm, and then dividing the regional comprehensive grid into differentiated sub-grids, outputting a regional comprehensive grid set and a differentiated sub-grid set; quantifying the risk level of the differentiated sub-grids in the differentiated sub-grid set, and outputting a risk level set; calculating the comprehensive risk coefficient of the small grid risk data within the parent superpixel using the PLSR algorithm, evaluating the level of the comprehensive risk coefficient based on a preset risk threshold, and issuing a corresponding intensity warning based on the evaluation level.

[0015] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the steps of a fire source early warning system based on drone monitoring.

[0016] 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 fire source early warning system based on UAV monitoring.

[0017] As can be seen from the above technical solution, compared with the prior art, the present invention discloses a fire source early warning system, method, device and storage medium based on UAV monitoring. It adapts to the complex terrain and equipment distribution characteristics of mountain photovoltaic power stations by dynamically dividing the grid into two levels, constructs a multi-dimensional risk index by combining multi-source sensor data, and realizes dynamic assessment of regional comprehensive risk by using the PLSR algorithm. At the same time, it generates graded disposal strategies by linking equipment importance, meteorological and terrain factors. It effectively solves the problems of poor adaptability, one-sided risk assessment and mismatch of disposal resources in the existing early warning system, significantly improves the accuracy and timeliness of fire source early warning, and minimizes the equipment damage, economic loss and personnel safety risks caused by fire. Attached Figure Description

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

[0019] Figure 1 The system flowchart provided for this invention; Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] This invention discloses a fire source early warning system based on drone monitoring, such as... Figure 1 As shown, the system includes: a UAV data acquisition module, a secondary grid division module, a small grid risk assessment module, a regional comprehensive assessment module, and a tiered response linkage module. The UAV data acquisition module collects multi-source data using multi-source sensors carried by the UAV, and outputs a multi-source data set after preprocessing. The secondary grid division module, based on the multi-source data set, divides the region into a comprehensive grid using a linear iterative clustering algorithm, and then divides the comprehensive grid into differentiated sub-grids, outputting a comprehensive grid set and a differentiated sub-grid set. The small grid risk assessment module quantifies the risk level of the differentiated sub-grids in the differentiated sub-grid set and outputs a risk level set. The regional comprehensive assessment module calculates the comprehensive risk coefficient of the small grid risk data within the parent superpixel using the PLSR algorithm, assesses the level of the comprehensive risk coefficient based on a preset risk threshold, and issues a corresponding intensity warning based on the assessment level.

[0022] Preferably, the multi-source data collected by the UAV data acquisition module includes: Visible light image data, For infrared thermal imaging data, It is five-band multispectral data. GPS coordinate data, For terrain elevation data, This is meteorological data.

[0023] Preferably, the secondary mesh generation module specifically includes: The regional integrated grid division unit establishes a six-dimensional feature vector based on visible light image data, GPS coordinate data, and terrain elevation data. Initial cluster centers are evenly distributed according to a preset number of parent superpixels. Within a 3×3 pixel range around each initial cluster center, the gradient of neighboring pixels is calculated. The final cluster center is reselected based on the device distribution density. Within a 2S×2S neighborhood of the final cluster center, the comprehensive distance of the pixels is calculated based on color, space, and scene adaptation distance. Pixels are distributed based on the comprehensive distance to obtain the integrated grid of each region. The differentiated sub-mesh partitioning module presets the number of sub-superpixels and dynamically adjusts the number of sub-superpixels according to the device density and terrain complexity within each region's integrated mesh. After calculating the sub-cluster center step size, it follows the pixel allocation logic of the region integrated mesh partitioning unit to obtain the differentiated sub-mesh.

[0024] Preferably, the regional integrated grid division unit specifically includes: Fusion of visible light image data Lab color space, GPS coordinate data The coordinates (x, y) and The slope grade forms a six-dimensional feature vector. ;in, l For brightness, a Channel A b For channel B, Slope grade; initial cluster center step size Where N is the total number of pixels in the monitoring area, and K is the preset number of parent superpixels; Calculate the gradient value within the initial 3×3 pixel neighborhood. Combining the device distribution density weights, the location of the device distribution density weight with the smallest gradient and within the preset device density range is selected as the final cluster center; Overall distance ; in , , m is the maximum color distance constant.i As the final cluster center identifier, j As the identifier of the pixels in the neighborhood of the cluster center, For color distance, For spatial distance, Adjust the distance to suit the scene.

[0025] Preferably, dynamically adjusting the number of sub-superpixels based on the device density and terrain complexity within the integrated grid of each region specifically includes: ; in, For equipment density, For terrain complexity; The preset number of sub-superpixels, To adjust the number of sub-superpixels, For the corresponding weights.

[0026] Preferably, the small grid risk assessment module specifically includes: The label calculation unit determines the equipment thermal anomaly index, vegetation flammability index, and coupling spread index for each differentiated subgrid. The data required for the calculation are all from visible light image data. Infrared thermal imaging data Five-band multispectral data GPS coordinate data Topographic elevation data and meteorological data The details are as follows: Equipment thermal anomaly index ; in, This is the infrared temperature value. This refers to the mid-to-far infrared reflectance value. Values ​​in the visible light band. The higher the value, the more severe the thermal anomaly. Vegetation flammability index ; in, Values ​​in the near-infrared band. This is the value in the shortwave infrared band. The higher the value, the more flammable the vegetation; Coupling Spread Index ; in, For slope, Let d be the angle between the slope and the direction of fire spread, d be the distance from the center of the fire source, and h be the elevation difference. , As weight, The wind speed influence coefficient, For wind speed, This is the elevation difference suppression coefficient. The higher the value, the greater the risk of spread; The optimal threshold calculation unit iterates through all candidate threshold values ​​within the range of each index value in the label calculation unit; for each candidate threshold value, all sub-superpixels are divided into foreground and background, and the pixel ratio of foreground and background and the mean of foreground and background regions are calculated; device association weights are introduced to calculate the weighted inter-class variance; the weighted inter-class variance corresponding to each candidate threshold value is calculated one by one, and the candidate value with the largest weighted inter-class variance is selected as the optimal threshold for each index; The risk assessment unit calculates a risk score by weighting the indices of the assessment label calculation unit with the optimal thresholds corresponding to the indices of the optimal threshold calculation unit.

[0027] The PLSR algorithm is used to determine the PLS components, which are weighted linear combinations of the original explanatory and response variables, and are able to capture the most important variances in the dataset. Preferably, the regional comprehensive assessment module includes: The matrix processing unit constructs an explanatory variable matrix using the equipment thermal anomaly index, vegetation flammability index, coupling spread index, risk score, and equipment importance level; and constructs a response variable matrix using the equipment damage loss coefficient and personnel safety risk coefficient; and performs standardization processing on the explanatory variable matrix and the response variable matrix. Component screening unit, initializing load vector and weight vector The load vector is initialized as a standardized explanatory variable matrix. The orientation of the first principal component, the weight vector is initialized based on the loading vector; the scores of the explanatory variables are calculated. and response variable score ,in Weights for response variables, based on the standardized response variable matrix. and Determine the maximum correlation; update the loads and weights. ; Calculate the residual matrix , Using E and F as the new and Iterate through the above steps until the residual variance is less than a preset threshold, and finally determine the three PLS components. The state vector update unit uses the scores of the three PLS components as the base dimension, combined with a meteorological dynamic correction term. Constructing the region state vector The PLS component weights are calculated based on the proportion of explained variance, using the following formula: ,in For the first z The proportion of variance explained by each PLS component The total explained variance proportion; new data collected by the UAV is received at fixed time intervals, and the weighted PLS component score is calculated. Update the weather correction items to The updated state vector is obtained. ; The regional comprehensive risk coefficient calculation unit is based on the updated state vector and combines the equipment importance factor. and meteorological weight Calculate the comprehensive risk coefficient of each region's integrated grid using the following formula: .

[0028] This application utilizes Partial Least Squares Regression (PLSR) analysis to transform multi-source sensor data (visible light images, infrared thermal imaging, multispectral data, GPS coordinates, etc.) collected by UAVs into a regional integrated state vector. This vector accurately reflects the fire risk status of the monitored area by capturing key risk indicators such as equipment thermal anomalies, vegetation flammability, and coupled spread, as well as the complex relationships between them. Using the PLS components and their scores determined by the PLSR algorithm, a multi-dimensional regional state vector is constructed by integrating equipment importance levels and meteorological dynamic correction terms. The weight of each component is calculated by explaining the variance ratio to ensure that the state vector accurately reflects the importance of each risk influencing factor. New data collected by the UAV is received at fixed time intervals, and the state vector is continuously updated to reflect the dynamic changes in fire risk in the monitored area in real time.

[0029] A fire source early warning method based on drone monitoring includes: collecting multi-source data using a drone equipped with multi-source sensors, and outputting a multi-source data set after preprocessing; dividing the multi-source data set into a regional comprehensive grid using a linear iterative clustering algorithm, and then dividing the regional comprehensive grid into differentiated sub-grids, outputting a regional comprehensive grid set and a differentiated sub-grid set; quantifying the risk level of the differentiated sub-grids in the differentiated sub-grid set, and outputting a risk level set; calculating the comprehensive risk coefficient of the small grid risk data within the parent superpixel using the PLSR algorithm, evaluating the level of the comprehensive risk coefficient based on a preset risk threshold, and issuing a corresponding intensity warning based on the evaluation level.

[0030] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the steps of a fire source early warning system based on drone monitoring.

[0031] 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 fire source early warning system based on UAV monitoring.

[0032] Based on preset risk thresholds, a comprehensive risk coefficient is assessed, and corresponding intensity warnings are issued according to the assessment level, including: Low-risk scenarios ρ <0.8): Autonomous operation and maintenance: pushes precise GPS coordinates of the fire source, risk index details and vegetation distribution data; micro drones carrying dry powder extinguishing agent extinguish the fire, continuously monitor for 15 minutes, update risk data every 3 minutes, and adjust the monitoring frequency according to wind speed.

[0033] Medium-risk scenarios (0.8≤ ρ <1.5): Operation and maintenance fire protection linkage: The photovoltaic module circuits on the spread path are cut off through the power station operation and maintenance system (precise power cut-off based on cable trench coordinates); the coordinates of the fire source, navigation information of mountain fire lanes, real-time status vectors and terrain elevation data are pushed to the fire department, and the regional risk is updated every 5 minutes, and rescue route suggestions are adjusted according to the slope.

[0034] High-risk scenarios ρ ≥1.5): Highest level emergency response: Cut off the main power supply to the area and simultaneously send evacuation notices to the operation and maintenance center, emergency management department and surrounding villages; dispatch more than 3 drones to build a monitoring network and transmit the dynamics of the fire spread in real time.

[0035] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. The methods disclosed in the embodiments are described simply because they correspond to the methods disclosed in the embodiments; relevant parts can be found in the method section.

[0036] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those 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 invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A fire source early warning system based on drone monitoring, characterized in that, include: The system includes a UAV data acquisition module, a secondary grid division module, a small grid risk assessment module, a regional comprehensive assessment module, and a hierarchical response linkage module. The UAV data acquisition module uses a UAV to carry multi-source sensors to collect multi-source data, and outputs a multi-source data set after preprocessing. The secondary grid partitioning module, based on the multi-source data set, divides the region into a comprehensive grid according to a linear iterative clustering algorithm, and then divides the region into differentiated sub-grids within the comprehensive grid, outputting a set of comprehensive grids and a set of differentiated sub-grids; The small grid risk assessment module quantifies the risk level of the differentiated subgrids in the differentiated subgrid set and outputs a risk level set. The regional comprehensive assessment module uses the PLSR algorithm to calculate the comprehensive risk coefficient of small grid risk data within the parent superpixel, evaluates the level of the comprehensive risk coefficient based on a preset risk threshold, and issues a corresponding intensity warning according to the evaluation level.

2. The fire source early warning system based on UAV monitoring according to claim 1, characterized in that, The multi-source data collected by the UAV data acquisition module includes: Visible light image data, For infrared thermal imaging data, It is five-band multispectral data. GPS coordinate data, For terrain elevation data, This is meteorological data.

3. A fire source early warning system based on UAV monitoring according to claim 2, characterized in that, The secondary grid partitioning module specifically includes: The regional integrated grid division unit establishes a six-dimensional feature vector based on visible light image data, GPS coordinate data, and terrain elevation data. Initial cluster centers are evenly distributed according to a preset number of parent superpixels. Within a 3×3 pixel range around each initial cluster center, the gradient of neighboring pixels is calculated. The final cluster center is reselected based on the device distribution density. Within a 2S×2S neighborhood of the final cluster center, the comprehensive distance of the pixels is calculated based on color, space, and scene adaptation distance. Pixels are distributed based on the comprehensive distance to obtain the integrated grid of each region. The differentiated sub-mesh partitioning module presets the number of sub-superpixels and dynamically adjusts the number of sub-superpixels according to the device density and terrain complexity within each region's integrated mesh. After calculating the sub-cluster center step size, it follows the pixel allocation logic of the region integrated mesh partitioning unit to obtain the differentiated sub-mesh.

4. A fire source early warning system based on UAV monitoring according to claim 3, characterized in that, The regional integrated grid division unit specifically includes: Fusion of visible light image data Lab color space, GPS coordinate data The coordinates (x, y) and The slope grade forms a six-dimensional feature vector. ;in, l For brightness, a Channel A b For channel B, Slope grade; initial cluster center step size Where N is the total number of pixels in the monitoring area, and K is the preset number of parent superpixels; Calculate the gradient value within the initial 3×3 pixel neighborhood. Combining the device distribution density weights, the location of the device distribution density weight with the smallest gradient and within the preset device density range is selected as the final cluster center; Overall distance ; in , , m is the maximum color distance constant. i As the final cluster center identifier, j As the identifier of the pixels in the neighborhood of the cluster center, For color distance, For spatial distance, Adjust the distance to suit the scene.

5. A fire source early warning system based on UAV monitoring according to claim 3, characterized in that, The number of sub-superpixels is dynamically adjusted based on the device density and terrain complexity within each region's integrated grid, specifically including: ； in, For equipment density, For terrain complexity; The preset number of sub-superpixels, To adjust the number of sub-superpixels, For the corresponding weights.

6. A fire source early warning system based on UAV monitoring according to claim 1, characterized in that, The small grid risk assessment module specifically includes: The label calculation unit is determined, and the equipment thermal anomaly index, vegetation flammability index, and coupling spread index of each differentiated subgrid are calculated as follows: Equipment thermal anomaly index ； in, This is the infrared temperature value. This refers to the mid-to-far infrared reflectance value. Values ​​in the visible light band. The higher the value, the more severe the thermal anomaly. Vegetation flammability index ； in, Values ​​in the near-infrared band. This is the value in the shortwave infrared band. The higher the value, the more flammable the vegetation; Coupling Spread Index ; in, For slope, Let d be the angle between the slope and the direction of fire spread, d be the distance from the center of the fire source, and h be the elevation difference. , As weight, The wind speed influence coefficient, For wind speed, This is the elevation difference suppression coefficient. The higher the value, the greater the risk of spread; The optimal threshold calculation unit iterates through all candidate threshold values ​​within the range of each index value in the label calculation unit; for each candidate threshold value, all sub-superpixels are divided into foreground and background, and the pixel ratio of foreground and background and the mean of foreground and background regions are calculated; device association weights are introduced to calculate the weighted inter-class variance; the weighted inter-class variance corresponding to each candidate threshold value is calculated one by one, and the candidate value with the largest weighted inter-class variance is selected as the optimal threshold for each index; The risk assessment unit calculates a risk score by weighting the indices of the assessment label calculation unit with the optimal thresholds corresponding to the indices of the optimal threshold calculation unit.

7. A fire source early warning system based on UAV monitoring according to claim 6, characterized in that, The regional comprehensive assessment module includes: The matrix processing unit constructs an explanatory variable matrix using the equipment thermal anomaly index, vegetation flammability index, coupling spread index, risk score, and equipment importance level; and constructs a response variable matrix using the equipment damage loss coefficient and personnel safety risk coefficient; and performs standardization processing on the explanatory variable matrix and the response variable matrix. Component screening unit, initializing load vector and weight vector The load vector is initialized as a standardized explanatory variable matrix. The orientation of the first principal component, the weight vector is initialized based on the loading vector; the scores of the explanatory variables are calculated. and response variable score ,in Weights for response variables, based on the standardized response variable matrix. and Determine the maximum correlation; update the loads and weights. ; Calculate the residual matrix , Using E and F as the new and Iterate through the above steps until the residual variance is less than a preset threshold, and finally determine the three PLS components. The state vector update unit uses the scores of the three PLS components as the base dimension, combined with a meteorological dynamic correction term. Constructing the region state vector The PLS component weights are calculated based on the proportion of explained variance, using the following formula: ,in For the first z The proportion of variance explained by each PLS component The total explained variance proportion; new data collected by the UAV is received at fixed time intervals, and the weighted PLS component score is calculated. Update the weather correction items to The updated state vector is obtained. ; The regional comprehensive risk coefficient calculation unit is based on the updated state vector and combines the equipment importance factor. and meteorological weight Calculate the comprehensive risk coefficient of each region's integrated grid using the following formula: .

8. A fire source early warning method based on drone monitoring, applied to a fire source early warning system based on drone monitoring as described in any one of claims 1-7, characterized in that, include: The drone carries multiple sensors to collect multi-source data, and outputs a multi-source data set after preprocessing. Based on the multi-source data set, a regional integrated grid is divided according to the linear iterative clustering algorithm, and then a differentiated sub-grid is divided within the regional integrated grid, outputting a regional integrated grid set and a differentiated sub-grid set; The risk level of the differentiated subgrids in the differentiated subgrid set is quantified, and a risk level set is output. The PLSR algorithm is used to calculate the comprehensive risk coefficient of small grid risk data within the parent superpixel. Based on the preset risk threshold, the comprehensive risk coefficient is used to assess the level, and a corresponding intensity warning is issued according to the assessment level.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the system according to any one of claims 1 to 7.

10. 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 system according to any one of claims 1 to 7.

Images