Real-time monitoring method and system for forest fire danger based on multi-source perception cooperation

Abstract

The application discloses a forest fire danger real-time monitoring method and system based on multi-source sensing cooperation, and relates to the technical field of fire danger monitoring.The system comprises a plurality of sensing node groups, a fire point positioning module, an event detection module, a wind field processing module and a cooperative verification module; based on multi-source sensing cooperation, suspected flame feature pixel clusters are extracted by using multi-spectral images, and fire point position estimation values are obtained by geographic back projection; environmental abnormal events are synchronously detected and positions are recorded; a downwind angle is determined by combining wind field data to construct a fan-shaped potential impact area.When the abnormal event position falls into the fan-shaped area or the distance from the fire point is less than a preset close distance threshold, and the event starting time difference is less than the maximum correlation duration, a fire danger confirmation signal is generated.The method judges homology by spatial and temporal double constraints, reduces false positives of single spectrum interference such as sunlight reflection and high-temperature ground, and reduces false positives of non-fire abnormalities of environmental sensors, and improves the reliability of monitoring.

Description

Technical Field

[0001] This invention relates to the field of fire risk monitoring technology, specifically to a method and system for real-time monitoring of forest fire risk based on multi-source sensing collaboration. Background Technology

[0002] Early detection of forest fire risks is crucial for forest resource protection. Existing monitoring methods mainly fall into two categories: one is flame detectors based on multispectral or infrared optical images, which detect fire points by identifying the spectral characteristics of flames, but are prone to false alarms when faced with interference such as sunlight reflection and high-temperature ground surfaces; the other is anomaly detection systems based on environmental sensors (such as temperature and humidity sensors, particulate matter concentration sensors), which indirectly determine fire risks by sensing changes in environmental parameters caused by fires, but non-fire factors such as wind, rain, and human activities can also trigger abnormal alarms.

[0003] Because single sensors inherently suffer from false alarms, existing technologies include simple linkages between optical detection and environmental sensing, such as confirming a fire hazard when both alarms occur simultaneously. However, this simple AND-OR logic lacks a deeper assessment of whether the two events truly originate from the same source—a suspicious fire detected by the optical detector and an anomaly sensed by the environmental sensor may originate from completely different physical sources, and mere temporal coincidence is insufficient to establish a reliable causal relationship. Therefore, a technical solution is urgently needed that can accurately determine whether visual fire points and environmental anomalies share the same origin. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method and system for real-time monitoring of forest fire risk based on multi-source sensing collaboration, which solves the problems mentioned in the background.

[0005] To achieve the above objectives, the technical solution of the present invention is as follows: Firstly, a real-time forest fire risk monitoring method based on multi-source sensing collaboration is applied to a real-time forest fire risk monitoring system based on multi-source sensing collaboration, which includes multispectral flame detectors, wind speed and direction sensors, and environmental sensors. The method includes the following steps: S1. Data Processing: S1.1 Acquire multispectral images collected by a multispectral flame detector, extract suspected flame feature pixel clusters that meet the preset flame spectrum determination conditions from the multispectral images, and perform geographic back projection on the suspected flame feature pixel clusters to obtain the estimated fire point location. Geographic back projection is a coordinate transformation method that converts image pixel coordinates into geodetic latitude and longitude coordinates. Among them, the starting time of the suspected flame feature pixel cluster is the starting acquisition time when the pixel area of ​​the suspected flame feature pixel cluster first exceeds the preset area threshold and continues to increase for multiple consecutive frames. S1.2 Acquire environmental data synchronously collected by environmental sensors, detect abnormal environmental events in real time based on the changing characteristics of the environmental data, and record the location of the sensor node that issued the abnormal environmental event. S1.3. Obtain wind field data collected by wind speed and direction sensors within a preset time window before and after the start of an environmental anomaly event, and determine the downwind angle based on the wind field data; the downwind angle is obtained by adding 180 degrees to the wind direction angle and taking the modulus of 360 degrees.

[0006] S2, Spatiotemporal verification: S2.1 Using the estimated location of the fire point as the diffusion source, construct a fan-shaped potential influence area with a preset angle and a preset maximum influence distance along the downwind direction; Spatial consistency is determined to be satisfied when the location of the sensor node that issued the environmental anomaly falls within the fan-shaped potential influence area, or when the distance between the estimated fire location and the location of the sensor node that issued the environmental anomaly is less than a preset proximity threshold. S2.2 Obtain the start time of the abnormal environmental event and the start time of the suspected flame feature pixel cluster. When the absolute value of the time difference between the two is less than the preset maximum association time, it is determined that the time consistency is satisfied. S3. Fire risk confirmation: A fire hazard confirmation signal is generated when spatial and temporal consistency are satisfied.

[0007] Secondly, a real-time forest fire risk monitoring system based on multi-source sensing and collaboration includes: Multiple sensing node groups are deployed in the monitoring area. Each sensing node group integrates a multispectral flame detector, a wind speed and direction sensor, and at least one type of environmental sensor. The fire location module is used to extract suspected flame feature pixel clusters from the multispectral images obtained by the multispectral flame detector and obtain the estimated fire location through geographic backprojection. The event detection module is used to detect abnormal environmental events in real time from environmental data acquired by environmental sensors and record the location of the sensor node that emitted the event. The wind field processing module is used to acquire wind field data collected by wind speed and direction sensors within a preset time window before and after the onset of an environmental anomaly, and to determine the downwind angle. The collaborative verification module is used to perform spatial consistency verification and temporal consistency verification, and outputs a fire hazard confirmation signal when spatial consistency and temporal consistency are satisfied.

[0008] Furthermore, the spatial consistency verification includes: using the estimated fire point location as the diffusion source, constructing a fan-shaped potential influence area with a preset angle and a preset maximum influence distance along the downwind angle; when the location of the sensor node that issued the environmental anomaly event falls within the fan-shaped potential influence area, or when the distance between the estimated fire point location and the location of the sensor node that issued the event is less than a preset proximity threshold, it is determined that the spatial consistency is satisfied. The time consistency check includes: obtaining the start time of the abnormal event and the start time of the suspected flame feature pixel cluster recorded by the fire point location module. When the absolute value of the time difference between the two is less than the preset maximum association time, it is determined that the time consistency is satisfied.

[0009] Thirdly, a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a method for real-time monitoring of forest fire risk based on multi-source sensing collaboration.

[0010] This invention provides a method and system for real-time monitoring of forest fire risk based on multi-source sensing collaboration. Compared with existing technologies, it has the following advantages: This invention uses the ratio of infrared 4.3μm to 3.8μm bands to extract suspected flame pixels, which can effectively distinguish real flames from typical false alarm sources such as sunlight reflection and high-temperature ground surfaces. By using geographic back projection to convert image coordinates into estimated geographic locations, it provides a unified coordinate system benchmark for subsequent spatial verification and suppresses the false alarm rate of single spectral criteria from the optical perception level.

[0011] It can simultaneously collect environmental data and detect abnormal environmental events, and record the location of the sensor node that issued the event. It can capture environmental disturbances accompanying fire hazards from the dimensions of changes in physical quantities such as temperature, humidity and particulate matter concentration, making up for the insensitivity of relying solely on optical images to flameless or obscured fire points.

[0012] By acquiring wind field data and determining the downwind angle, and using the estimated location of the fire point as the diffusion source, a fan-shaped potential influence area is constructed. The effect of wind field transmission is incorporated into the spatial consistency judgment, making the correlation analysis between environmental anomalies and fire points more consistent with the actual physical process of smoke and heat diffusion.

[0013] When the location of the sensor node that emits an environmental anomaly falls into the fan-shaped potential influence area, or the distance between the two is less than a preset proximity threshold, it is determined that spatial consistency is satisfied. This provides a spatial constraint condition based on wind field guidance and proximity rules for determining whether the optical fire point and the environmental anomaly are from the same source.

[0014] The system obtains the start time of the abnormal environmental event and the start time of the suspected flame feature pixel cluster. When the absolute value of the time difference between the two is less than the preset maximum correlation duration, it is determined that the time consistency is satisfied. The system constrains the causal relationship between the two types of events belonging to the same fire process from the time dimension.

[0015] Fire hazard confirmation signals are generated when spatial and temporal consistency is met. Collaborative decision-making is carried out through cross-verification of multi-source sensing information in the spatiotemporal dimension, reducing false alarms caused by relying on a single sensor or simple logic linkage, and improving the reliability of fire hazard confirmation. Attached Figure Description

[0016] Figure 1 This is a system block diagram of the real-time forest fire risk monitoring system based on multi-source sensing collaboration of the present invention.

[0017] Figure 2 This is a flowchart illustrating the real-time monitoring method for forest fire risk based on multi-source sensing collaboration according to the present invention. Detailed Implementation

[0018] 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.

[0019] Please see Figure 1 and Figure 2 As shown, the embodiments of the present invention provide the following technical solutions: Example 1 This invention provides a real-time forest fire risk monitoring method based on multi-source sensing collaboration. This method is implemented in a monitoring area with multiple sensing node groups deployed. Each sensing node group integrates a multispectral flame detector, wind speed and direction sensor, temperature and humidity sensor, and particulate matter concentration sensor. All sensors within the group share the same spatial coordinates and a synchronization clock. The method is as follows: S1. Data Processing: S1.1 Multispectral Image Processing: Multispectral flame detectors acquire multispectral images at a fixed frequency.

[0020] For each pixel in the multispectral image, the ratio of its radiant intensity in the 4.3μm infrared band to that in the 3.8μm infrared band is calculated and used as a characteristic value for flame spectral determination.

[0021] This step uses the ratio of infrared radiant intensity in the 4.3μm band to that in the 3.8μm band as the characteristic value for flame spectral identification. This is based on the fact that high-temperature CO2 gas produced by hydrocarbon combustion has an extremely strong narrow-band emission peak in the 4.3μm band, while common false alarm targets such as sunlight reflection and hot ground surfaces lack a high-temperature CO2 gas envelope, resulting in a much lower relative radiative response in the 4.3μm band compared to actual flames. By calculating the band intensity ratio, which approximates the radiance ratio, the spectral differences between flame targets and false alarm sources can be effectively amplified. Simultaneously, it helps to offset, to some extent, the multiplicative effects of detection distance, atmospheric transmission, and detector gain variations on absolute radiant intensity, thus achieving robust extraction of suspected flame pixels.

[0022] Extract the preset flame spectral feature range [F] low ,F high ], the flame spectrum determination feature value is located in the flame spectrum feature range [F low ,F high Pixels marked with a spectral similarity are identified as points whose spectral characteristics closely match the CO2 emission characteristics during flame combustion. These pixels are clustered into clusters using connected component labeling, and only pixels with an area exceeding a preset area threshold A are retained. th The pixel clusters are considered as suspected flame feature pixel clusters.

[0023] Among them, F low F is the boundary threshold between background noise and weak fire characteristics. high This serves as the boundary threshold between strong interference sources such as flames and high-temperature ground surfaces. Both can be pre-calibrated based on the detector's spectral response characteristics and historical fire data. The pixel area that is retained will first exceed the preset area threshold A. th Furthermore, the starting acquisition time of the subsequent N consecutive frames that continue to increase is recorded as the starting time of the suspected flame feature pixel cluster, where N is a preset value. In this embodiment, the connected component labeling aggregation method is as follows: traverse all pixels of the image, mark the suspected spectral points as 1, and the rest as 0, to generate a binary mask image; for each pixel with a value of 1 in the mask image, use the 8-connected component labeling algorithm to assign the same unique cluster ID to spatially adjacent labeled pixels; area statistics: count the total number of pixels corresponding to each cluster ID, as the area of ​​the pixel cluster; filter out pixels with an area lower than the threshold A. th The remaining clusters of valid suspected flame feature pixels are retained.

[0024] In this embodiment, the determination method for subsequent consecutive frames to continue growing is to accumulate and cache the determination of the most recent N frames through a sliding window. Only when all N frames meet the growth condition are the frames included in the window backtracked to determine that they constitute continuous growth.

[0025] Geographic backprojection was performed using the centroid location of the suspected flame feature pixel cluster and the detector's imaging geometry parameters to obtain the estimated fire location value P. F In this embodiment, the geographic backprojection method is as follows: Obtain the pixel set of the suspected flame feature pixel cluster, calculate the arithmetic mean of the row coordinates of each pixel in the pixel set as the centroid row coordinates, and the arithmetic mean of the column coordinates of each pixel as the centroid column coordinates, denoted as the centroid image coordinates, i.e., the centroid position.

[0026] Read the distortion coefficients obtained from the detector's factory calibration or field calibration, including radial and tangential distortion coefficients. Based on the distortion model, calculate the row-direction distortion offset and column-direction distortion offset at the centroid image coordinates, respectively. Subtract the corresponding distortion offset from the centroid image coordinates to obtain the coordinates of the distortion-free ideal image point.

[0027] The detector imaging geometric parameters, including the horizontal equivalent focal length, the vertical equivalent focal length, and the principal point coordinates in pixels, are obtained. Based on the inverse transformation of perspective projection, the offset between the ideal image point coordinates and the principal point coordinates is divided by the equivalent focal length in the corresponding direction, and then a normalized unit component is added to form a direction vector in the detector image space coordinate system. This direction vector points from the projection center of the detector to the ground object point corresponding to the centroid.

[0028] The extrinsic parameters of the detector in the world coordinate system are obtained. These parameters include the three-dimensional position coordinates of the detector's projection center and a rotation matrix representing the detector's attitude. The rotation matrix can be derived from the detector's pitch, roll, and yaw angles using trigonometric functions. Using this rotation matrix, the image space direction vector is rotated to the world coordinate system, resulting in the ray direction vector in the world coordinate system. A ray equation is then constructed, extending from the detector's projection center along the ray direction vector. The coordinates of points on the ray are represented in first-order parametric form, with positive values ​​indicating forward extension from the projection center.

[0029] Obtain digital elevation model data of the field-of-view coverage area of the detector, or preset a reference elevation value representing the local average ground elevation. Under the setting that the vertical axis of the world coordinate system points to the zenith, the surface elevation can be expressed as a horizontal plane whose vertical coordinate is equal to the elevation value. Substitute the ray equation into the elevation plane equation to solve for the value of the ray extension parameter: when the vertical axis component of the ray direction vector is not zero, divide the difference between the elevation value and the vertical coordinate of the detector projection center by the vertical axis component to obtain the extension parameter; when the vertical axis component is zero, an alternative method can be used, for example, projecting the ray onto the horizontal plane, solving for its intersection with the circle centered at the projection center of the detector and with a preset maximum effective detection distance as the radius, and taking the intersection along the ray direction as an approximate solution of the fire point position, wherein the maximum effective detection distance is preset according to the specification parameters of the multi-spectral flame detector, such as 1500 meters. Substitute the extension parameter back into the ray equation to obtain the three-dimensional position coordinates of the fire point in the world coordinate system. When there is no intersection between the ray and the elevation surface, the following judgment rules are adopted: if the detector has a downward pitch angle and the field of view covers the ground surface, the ground projection point of the detector projection center is taken as the approximate fire point coordinate; if the detector has an upward elevation angle and there is no ground intersection, the current frame of collaborative verification is directly abandoned, and recalculation is performed for the next frame of image.

[0030] Adopt a predefined WGS84 ellipsoid model and the corresponding coordinate system transformation relationship to convert the three-dimensional position coordinates of the fire point into the geodetic coordinate form of longitude, latitude and elevation, and take the longitude value and latitude value as the estimated fire point position.

[0031] S1.2, Abnormal event detection: The temperature and humidity sensor and the particulate matter concentration sensor collect environmental data synchronously, wherein the environmental data includes the temperature value and current humidity value collected by the temperature and humidity sensor, and the particulate matter concentration value collected by the particulate matter concentration sensor.

[0032] Through , calculate the combined temperature and humidity anomaly index Y TH : wherein T d and H d are the current temperature value and current humidity value respectively, T0 and H0 are the average value of temperature values and the average value of humidity values of the temperature and humidity sensor within a preset first specified period respectively, σ T and σ H are the standard deviation of temperature values and the standard deviation of humidity values respectively, α and β are preset weight coefficients satisfying α+β=1, which are used to adjust the contribution proportion of temperature change and humidity change to the anomaly index, with default values for example α=β=0.5. They can also be adjusted according to the strength of correlation between temperature and humidity and fire risk in historical fire data. In the formula, only when the temperature rises abnormally, that is T d >T0, and the humidity drops abnormally, that is H d <H0, the corresponding terms contribute to Y THIncrease.

[0033] pass Calculate the particulate matter concentration jump index Y PM Among them, PM d The current particulate matter concentration value, PM p This represents the average particulate matter concentration value of the sensor over a preset second specified period. A preset temperature and humidity combination threshold TH is extracted. th and particulate matter threshold PM th When Y TH Exceeding the preset temperature and humidity combination threshold TH th When Y is detected, an abnormal temperature and humidity change event is determined; when Y PM PM exceeding the preset particulate matter threshold th When this occurs, an abnormal sudden change in particulate matter concentration is determined.

[0034] Among them, abnormal temperature and humidity changes and abnormal particulate matter concentration changes are considered environmental anomalies, and the start time of an environmental anomaly is the moment when the corresponding index first exceeds the corresponding threshold. The location P of the sensor node that emitted the environmental anomaly is also recorded. Y .

[0035] Since temperature and humidity sensors and particulate matter concentration sensors within the same sensing node group share the same spatial coordinates, the sensor node position P Y These are the coordinates of the sensing node group.

[0036] S1.3 Wind field data determination: Wind field data collected within a preset time window before and after the onset of an environmental anomaly is obtained by wind speed and direction sensors on at least one sensing node group closest to the estimated location of the fire point.

[0037] The start time of the abnormal environmental event is denoted as t0, and the preset time window is [t0-Δt1, t0+Δt2]. The wind speed and direction sensor acquisition time that is closest to t0 within the preset time window is denoted as t0. k And obtain the acquisition time t. k Wind speed S(t) k ) and wind direction angle A(t) k (This is used as wind field data.)

[0038] If two acquisition times have the same time interval as t0, then the earlier acquisition time is taken as t. k , where k represents the number of the selected acquisition time in the wind speed and direction sensor sampling sequence.

[0039] Among them, the downwind angle is determined by A(t) kThe wind direction angle is calculated as (1) + 180°. In this embodiment, the wind direction angle adopts the meteorological standard and is defined as the direction from which the wind blows. The downwind angle, i.e. the diffusion direction, is the wind direction angle plus 180 degrees.

[0040] S2, Spatiotemporal verification: S2.1 Spatial Consistency Verification: Based on the estimated fire location P F As a diffusion source, a fan-shaped potential influence area with a preset angle and a preset maximum influence distance of 500 meters is constructed along the downwind direction. In this embodiment, the preset angle is ±30° and the maximum influence distance is 500 meters.

[0041] When the sensor node location P that issued the environmental anomaly event Y If a value falls into the sector, it is determined that the spatial consistency is satisfied.

[0042] Furthermore, when P F With P Y When the distance between them is less than the preset proximity threshold of 50 meters, regardless of the wind direction, it is directly determined that the spatial consistency is met.

[0043] S2.2 Time Consistency Verification: Obtain the start time of the detected environmental anomaly event and denot it as SA.

[0044] When both abnormal temperature and humidity events and abnormal particulate matter concentration events occur simultaneously, the earliest start time of the two events is taken as S. A ; Obtain the start time of the suspected flame feature pixel cluster and denote it as S. F .

[0045] Extract the preset maximum association duration S max When |S A -S F | Less than the maximum association duration S max If the time consistency is satisfied, then it is determined that the time consistency is met.

[0046] S3. Fire risk confirmation: When an environmental sensor and a flame detector from the same sensing node group trigger an abnormal environmental event and extract a cluster of suspected flame feature pixels, respectively, and both satisfy spatial and temporal consistency, a fire hazard confirmation signal is generated and an early warning terminal is triggered.

[0047] Example 2 Based on Embodiment 1, this embodiment provides an alternative spatial consistency verification method, which uses a dynamic potential influence region distribution instead of a fixed sector region for judgment. The specific method is as follows: Wind field data at different sampling times within a preset sliding time window is obtained from wind field data from wind speed and direction sensors. Using wind field data sequences recorded in the vicinity of the fire location estimate PF within a preset sliding time window, a dynamic potential impact area distribution is constructed with the wind direction angle in the wind field data sequences as the independent variable. .

[0048] The method for constructing the dynamic potential impact area distribution X is as follows: Obtain the wind speed S at each sampling time in the wind field data sequence. j and wind angle A j And through U j =A j The downwind angle U at each sampling time was calculated by adding 180°. j Where j=1,2,......,M, M represents the number of sampling times in the wind field data sequence; For each sampling moment within the sliding time window, the downwind angle U at that moment is used. j A Gaussian kernel function with wind direction angle γ as the independent variable is generated, centered on the distribution. ;in, The preset angle extension parameter is used to represent the width of the influence of a single wind field observation on the surrounding direction. Its value can be set between 10° and 30°, or preset according to the variance of the wind field data within the sliding time window.

[0049] Then, the wind speed S at each sampling time... j As the weighting coefficients of the Gaussian kernel function at that sampling moment, the Gaussian kernel functions at all sampling moments within the sliding time window are linearly superimposed, that is, by... The distribution of the unnormalized dynamic potential impact area was obtained, among which, This represents the original probability density value at the wind direction angle γ; Distribution of Unnormalized Dynamic Potential Impact Areas Normalization is performed by integrating X0(γ) over the global range [0°, 360°) and then... Calculate the normalization constant and then... Obtain the wind direction angle Distribution of the dynamic potential influence area of ​​the independent variable This leads to the distribution of the dynamic potential impact area. The integral value in all possible directions is 1. dγ represents the differential element of the wind direction angle in radians. If the entire circumference [0°, 360°] is divided into countless extremely small intervals, the width of each interval is dγ.

[0050] When performing spatial consistency checks, calculate from P FPoint to P Y The geographical azimuth angle is denoted as θ. Substitute θ into the dynamic potential impact area distribution... Obtain the probability density value corresponding to that azimuth angle. ). judge Whether it exceeds the preset collaborative probability threshold.

[0051] when If the probability exceeds a preset coordination probability threshold, spatial consistency is determined to be satisfied. The preset coordination probability threshold can be set based on the typical range of values ​​for the probability density function under no-fire conditions.

[0052] Furthermore, when the estimated fire location P F Location P of the sensor node that issued the environmental anomaly event Y When the distance between them is less than the preset proximity threshold of 50 meters, spatial consistency is directly determined to be satisfied regardless of the geographical azimuth intensity.

[0053] Example 3 Based on Example 1, this embodiment further proposes: a combined temperature and humidity threshold TH. th Particulate matter threshold (PM) th and flame spectral characteristic range [F low , F high The periodic updates are selected to be daily in this embodiment, and the thresholds remain constant throughout the day.

[0054] For temperature and humidity sensors and particulate matter concentration sensors, a daily baseline update time window is preset, for example, from 03:00 to 04:00 AM daily, which is typically a low-risk period for forest fires. Within this baseline update time window, temperature, humidity, and particulate matter concentration data are collected, and the mean value within this window is calculated and denoted as T. base H base PM base And the standard deviation, denoted as σT base σH base σPM base .

[0055] pass The baseline value of the temperature and humidity combination anomaly index was calculated. The baseline value for particulate matter concentration is directly taken as the window mean (PM). base ; pass and Calculate the temperature and humidity combination threshold TH for the day. th and particulate matter threshold PM thWherein, k1 and k2 are preset sensitivity coefficients. In this embodiment, k1 is 3 and k2 is 5.

[0056] For the flame spectral characteristic range [F low , F high The adjustment involves acquiring multispectral images within the same time window and filtering out pixels with a pixel area smaller than A. th For non-fire background pixel clusters, the distribution of their flame spectrum determination feature values ​​is statistically analyzed, and the p-percentile of this distribution is taken as F. high The dynamically adjusted value is taken as the q percentile of the distribution. low The dynamic adjustment value is used to adapt to changes in the spectral reflectance of vegetation in different seasons. In this embodiment, p is set to 99.5 and q is set to 0.5.

[0057] Example 4 This embodiment also provides a real-time forest fire risk monitoring system based on multi-source sensing collaboration, including multiple sensing node groups, an event detection module, a wind field processing module, and a collaborative verification module. This system is used to execute the real-time forest fire risk monitoring methods based on multi-source sensing collaboration described in Embodiments 1 to 3. The execution method is as follows: all sensing node groups, synchronized by a unified clock, acquire multispectral images, environmental data, and wind field data in parallel; the fire location module processes the image data and outputs P... F and S F The event detection module processes environmental data and outputs P. Y and S A The fire location module and the event detection module perform parallel operations, and the results are sent to the collaborative verification module. The collaborative verification module receives P... F and S A Then, a request is sent to the wind farm processing module, which then retrieves the distance P based on this request. F The wind field data of the nearest node within the SA neighborhood window is used to calculate the downwind angle or dynamic distribution. The collaborative verification module then integrates the spatial and temporal verification results and outputs a fire hazard confirmation signal.

[0058] Furthermore, any content not described in detail in this specification is existing technology known to those skilled in the art.

[0059] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0060] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

[0061] The various embodiments of this application have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A method for real-time monitoring of forest fire risk based on multi-source sensing collaboration, characterized in that, The method includes the following steps: S1. Data Processing: S1.1 Acquire multispectral images, extract suspected flame feature pixel clusters that meet the preset flame spectrum determination conditions from the multispectral images, and obtain the estimated fire point location through geographic back projection; S1.2 Acquire environmental data, detect abnormal environmental events in real time based on the changing characteristics of the environmental data, and record the location of the sensor node that issued the abnormal environmental event; S1.3 Obtain wind field data and determine the downwind angle based on the wind field data; S2, Spatiotemporal verification: S2.

1. Using the estimated fire location as the diffusion source, construct a fan-shaped potential influence area with a preset angle and a preset maximum influence distance along the downwind direction. When the location of the sensor node that issued the environmental anomaly event falls within the fan-shaped potential influence area, or when the distance between the estimated fire location and the location of the sensor node that issued the environmental anomaly event is less than a preset proximity threshold, it is determined that spatial consistency is satisfied. S2.2 Obtain the start time of the abnormal environmental event and the start time of the suspected flame feature pixel cluster. When the absolute value of the time difference between the two is less than the preset maximum association time, it is determined that the time consistency is satisfied. S3. Fire risk confirmation: A fire hazard confirmation signal is generated when spatial and temporal consistency are satisfied.

2. The method for real-time monitoring of forest fire risk based on multi-source sensing collaboration according to claim 1, characterized in that, Determining whether space consistency is satisfied can also be achieved in the following ways: Wind field data at different sampling times within a preset sliding time window are obtained from the wind field data to form a wind field data sequence, which includes wind speed and wind direction angle. Using the wind field data sequence recorded in the vicinity of the estimated fire location within the preset sliding time window, with the downwind angle at each sampling time as the distribution center and wind speed as the weighting coefficient, a dynamic potential impact area distribution with wind direction angle as the independent variable is constructed by superimposing and normalizing using a Gaussian kernel function. The geographical azimuth angle from the estimated fire location to the location of the sensor node that emitted the environmental anomaly is calculated, and it is determined whether the probability density value of this geographical azimuth angle in the dynamic potential impact area distribution exceeds a preset cooperative probability threshold. When the probability density value exceeds the preset cooperative probability threshold, spatial consistency is determined to be satisfied.

3. The method for real-time monitoring of forest fire risk based on multi-source sensing collaboration according to claim 1, characterized in that, The methods for detecting abnormal environmental events are as follows: Using temperature and humidity values ​​from environmental data, calculate a temperature and humidity combination anomaly index. When the temperature and humidity combination anomaly index exceeds a preset temperature and humidity combination threshold, a temperature and humidity abrupt change event is determined to have occurred. And / or using particulate matter concentration values ​​from environmental data, calculate the current particulate matter concentration value relative to a short-term historical baseline jump index. When the jump index exceeds a preset particulate matter jump threshold, a particulate matter concentration abrupt change event is determined to have occurred. The start time of an environmental anomaly event refers to the moment when the anomaly index first exceeds the corresponding preset threshold.

4. The method for real-time monitoring of forest fire risk based on multi-source sensing collaboration according to claim 1, characterized in that, The extraction method for suspected flame feature pixel clusters is as follows: For each pixel in the multispectral image, the ratio of its radiant intensity in the 4.3μm infrared band to that in the 3.8μm infrared band is calculated and used as a characteristic value for flame spectrum determination. Pixels whose flame spectrum feature values ​​fall within a preset flame spectrum feature range are marked as suspected spectral points, and then aggregated into pixel clusters by connecting component marking. Clusters of pixels whose pixel area exceeds a preset area threshold are retained as suspected flame feature pixel clusters.

5. The method for real-time monitoring of forest fire risk based on multi-source sensing collaboration according to claim 4, characterized in that, The start time of a suspected flame feature pixel cluster refers to the first frame acquisition time when the pixel area of ​​the suspected flame feature pixel cluster first exceeds the preset area threshold and continues to increase for multiple subsequent frames.

6. The method for real-time monitoring of forest fire risk based on multi-source sensing collaboration according to claim 1, characterized in that, Multispectral images are acquired through a multispectral flame detector, environmental data is acquired synchronously through environmental sensors, and wind field data is acquired through wind speed and direction sensors within a preset time window before and after the onset of an environmental anomaly.

7. The method for real-time monitoring of forest fire risk based on multi-source sensing collaboration according to claim 6, characterized in that, in, Environmental sensors include temperature and humidity sensors and particulate matter concentration sensors.

8. The method for real-time monitoring of forest fire risk based on multi-source sensing collaboration according to claim 6, characterized in that, Wind field data are collected by wind speed and direction sensors on at least one sensing node group closest to the estimated location of the fire point.

9. A real-time forest fire risk monitoring system based on multi-source sensing collaboration, the system being used to execute the real-time forest fire risk monitoring method based on multi-source sensing collaboration as described in any one of claims 1-8, characterized in that, The system includes: Multiple sensing node groups are arranged in the monitoring area. Each sensing node group consists of a multispectral flame detector, a wind speed and direction sensor, and at least one type of environmental sensor. The fire location module is used to extract suspected flame feature pixel clusters from the multispectral images obtained by the multispectral flame detector and obtain the estimated fire location through geographic backprojection. The event detection module is used to detect abnormal environmental events in real time from environmental data acquired by environmental sensors and record the location of the sensor node that issued the abnormal environmental event. The wind field processing module is used to acquire wind field data collected by wind speed and direction sensors within a preset time window before and after the onset of an environmental anomaly, and to determine the downwind angle. The collaborative verification module is used to perform spatial consistency verification and temporal consistency verification, that is, to determine whether spatial consistency and temporal consistency are satisfied. When spatial consistency and temporal consistency are satisfied, a fire hazard confirmation signal is output.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the method as claimed in any one of claims 1 to 8.

Images