A forest disaster unmanned aerial vehicle monitoring and early warning system and method

Abstract

The application discloses a forest disaster unmanned aerial vehicle monitoring and early warning system and method, and belongs to the technical field of forest disaster monitoring. The method comprises the following steps: acquiring vegetation reflectance spectrum collected by a multispectral imager carried by an unmanned aerial vehicle and ground temperature data collected by a thermal infrared imager; calculating a normalized difference vegetation index and a photosynthetically active radiation component from the vegetation reflectance spectrum, and generating a vegetation health index by geometric averaging of the two; constructing a temperature gradient matrix from the ground temperature data, and calculating a temperature field entropy value by taking each element as a state value; performing multivariate multiplication fusion of the vegetation health index and the temperature field entropy value to generate a disaster risk index; fitting an elliptical confidence domain based on two-dimensional distribution of the vegetation health index and the temperature field entropy value of the same region in the same period in history, and taking a risk index exceeding the elliptical confidence domain as a warning trigger condition to send a warning signal. The application realizes early identification of vegetation stress state, and improves disaster warning accuracy and timeliness.

Description

Technical Field

[0001] This invention relates to the field of forest disaster monitoring technology, specifically to a forest disaster unmanned aerial vehicle (UAV) monitoring and early warning system and method. Background Technology

[0002] Forest disaster monitoring and early warning require the timely detection of vegetation physiological anomalies and potential fire precursors. Existing technologies often rely on single remote sensing indicators for judgment, such as using the Normalized Difference Vegetation Index (NDV) to reflect vegetation cover changes, or simply relying on abnormal increases in surface temperature to identify fire points. In practical applications, when forest ecosystems are subjected to drought stress or the early stages of pests and diseases, subtle changes in vegetation spectral characteristics and surface thermal environment often occur simultaneously, but single indicators are insufficient to characterize the complex evolution of disasters. The NDC is sensitive to vegetation density and chlorophyll content, but its changes may lag before fires or in the early stages of pests and diseases and are easily affected by atmospheric and soil background interference. Surface temperature anomaly detection struggles to distinguish between normal diurnal warming, orographic thermal effects, and genuine hazardous heat sources; relying solely on temperature thresholds to trigger early warnings generates numerous false alarms. Attempts to integrate multi-source data typically employ linear weighting or simple threshold discrimination, failing to fully explore the nonlinear correlation between vegetation physiological states and the spatial structure of thermal infrared images, and have limited ability to capture weak signals in the early stages of disasters.

[0003] Existing methods generally use fixed empirical values ​​or manually assigned levels for setting early warning thresholds, which cannot adapt to dynamic changes in different seasons, altitudes, and forest types. In the same region, vegetation indices are naturally lower and temperature fields are naturally higher during the dry season; using fixed thresholds easily leads to misjudging normal seasonal fluctuations as disaster risks. Historical data has not been effectively used to construct adaptive normal dynamic baselines, resulting in insufficient performance and stability of early warning systems under different spatiotemporal conditions.

[0004] The issues that need to be addressed include how to integrate vegetation spectral indices and the spatial heterogeneity of the land surface temperature field to more sensitively reflect the early risks of forest disasters, and how to use historical monitoring data from the same period to construct adaptive confidence boundaries to reduce false alarms due to seasonal fluctuations. Summary of the Invention

[0005] This paper presents a method and system for monitoring and early warning of forest disasters using unmanned aerial vehicles (UAVs). The method generates a vegetation health index by geometrically averaging the normalized difference vegetation index derived from the vegetation reflectance spectrum and the photosynthetically active radiation component. It uses the entropy value of the temperature gradient matrix from surface temperature data to characterize the spatial heterogeneity of the thermal field, and fuses these two in a multivariate multiplication to obtain a disaster risk index. Simultaneously, it uses a two-dimensional distribution of historical vegetation health index and temperature field entropy values ​​to fit an elliptical confidence domain as a dynamic early warning boundary. This addresses the problems of insensitivity of single indicators to disaster precursors and high false alarm rates with fixed thresholds, achieving early warning of forest disasters through multi-source remote sensing fusion.

[0006] To achieve the above objectives, the present invention provides the following technical solution: The present invention provides a method for monitoring and early warning of forest disasters using unmanned aerial vehicles (UAVs), comprising:

[0007] The system acquires vegetation reflectance spectra from a multispectral imager mounted on a drone and surface temperature data from a thermal infrared imager. Based on the vegetation reflectance spectra, it calculates the normalized difference vegetation index and photosynthetically active radiation component. It then performs a geometric mean operation on the normalized difference vegetation index and the photosynthetically active radiation component to generate a vegetation health index. This index comprehensively characterizes the greenness of the vegetation canopy and its photosynthetic capacity, making the single index more sensitive to early physiological changes such as water stress and pest infestation.

[0008] A temperature gradient matrix is ​​constructed based on the surface temperature data. Information entropy is calculated using each element in the temperature gradient matrix as a state value to obtain the temperature field entropy value. The temperature gradient matrix retains the information on abrupt temperature changes in space, while the information entropy quantifies the degree of disorder in the temperature distribution as a whole, so that the contribution of local abnormal high temperature points or cold points to disaster characterization is not smoothly ignored.

[0009] The vegetation health index and the temperature field entropy value are multiplied together to generate a disaster risk index. This fusion method makes the risk index non-linearly amplified when vegetation health deterioration and temperature field disorder occur simultaneously. It can effectively suppress false alarms caused by normal fluctuations of a single factor and highlight the combined disaster-causing situations such as fires and pest outbreaks.

[0010] Based on the two-dimensional distribution of vegetation health index and temperature field entropy value accumulated in the same area during the same historical period, an elliptical confidence region is fitted. The disaster risk index exceeding the elliptical confidence region is used as the early warning trigger condition to issue an early warning signal. The historical statistical boundary is used to replace the fixed threshold, which is adapted to the normal fluctuation range of different seasons and different forest types, thereby improving the timeliness and regional adaptability of the early warning.

[0011] As a preferred technical solution of the present invention, when acquiring vegetation reflectance spectra, a multispectral imager is used to acquire raw radiance data in the visible and near-infrared bands. After radiometric calibration and atmospheric correction, the surface reflectance is obtained. The normalized difference vegetation index is calculated by extracting the red band reflectance and the near-infrared band reflectance from the surface reflectance. At the same time, the blue band reflectance and the red band reflectance are extracted to calculate the photosynthetically active radiation component. Through accurate atmospheric correction and multi-band collaborative extraction, the inversion accuracy of vegetation index and photosynthetically active radiation component is ensured, and the impact of changes in ambient light on the consistency of monitoring results is reduced.

[0012] Preferably, when constructing the temperature gradient matrix, the temperature values ​​of each pixel acquired by the thermal infrared imager are used to form a two-dimensional temperature matrix. The horizontal and vertical temperature differences are calculated for each pixel in the two-dimensional temperature matrix, and the temperature difference vectors in the two directions are synthesized to obtain the gradient magnitude. The gradient magnitudes of all pixels are used to construct the temperature gradient matrix. Using the gradient magnitude instead of the original temperature for entropy calculation can amplify the weight of abnormal structures such as temperature edges and fire fronts in the entropy value, thereby improving the ability to identify early fire lines and hidden hotspots.

[0013] Furthermore, the generation of the disaster risk index also includes acquiring air temperature and humidity data collected by a temperature and humidity sensor mounted on a drone, as well as wind speed data collected by an anemometer; calculating the equilibrium moisture content of combustibles based on the air temperature and humidity; substituting the equilibrium moisture content of combustibles and the wind speed data into a preset lookup table function to obtain a modulation factor; and multiplying the vegetation health index, the temperature field entropy value, and the modulation factor as the disaster risk index. By introducing dynamic modulation of meteorological elements, the risk index can reflect the promoting effect of surface combustible dryness and wind speed on fire spread, resulting in a higher disaster risk index generated under the same vegetation and temperature anomalies in flammable and easily spreading meteorological scenarios, making early warning triggering more reliable.

[0014] For calculating the equilibrium moisture content of combustibles, preferably, an approximate conversion based on thermodynamic equilibrium is used. The air temperature is converted into saturated water vapor pressure, and the air humidity is converted into actual water vapor pressure. The relative humidity is obtained based on the ratio of the two. Then, the relative humidity is substituted into the empirical function of wood equilibrium moisture content to obtain the equilibrium moisture content of combustibles. This method can obtain a reference value of the moisture content of dead branches and fallen leaves with close accuracy to actual measurement without carrying an additional combustible moisture content measuring device, simplifying the payload of the UAV.

[0015] Preferably, when fitting the elliptical confidence region, the historical records of the vegetation health index and the temperature field entropy value collected by the UAV in the same month are accumulated. The mean vector and covariance matrix of the two variables in the historical records are calculated. The radius of the ellipse is determined by taking the mean vector as the center, the direction of the eigenvector of the covariance matrix as the axis, and the critical value of the chi-square distribution at a confidence level of 0.95, thus generating the elliptical confidence region. This confidence region is adapted to the normal seasonal cycle and interannual fluctuation of the forest area, and can suppress false alarms with statistically controllable probability, and identify abnormal composite signals that deviate from the normal ecological spectrum as early as possible.

[0016] As a supplementary technical solution of the present invention, point cloud data of lidar carried by UAV is also acquired. Based on the point cloud data, forest canopy height and canopy closure are extracted. The forest canopy height and canopy closure are input into a shading correction function, and a correction coefficient is output. The correction coefficient is multiplied by the vegetation health index to obtain the corrected vegetation health index. The influence of shading and canopy gap on the vegetation index is compensated by the canopy structure information, so that the vegetation health index can still accurately reflect the overall physiological status of understory vegetation and canopy under closed forest stands and complex terrain conditions.

[0017] As another supplementary technical solution of the present invention, the carbon monoxide concentration and carbon dioxide concentration collected by the electrochemical gas sensor on the drone are also obtained, and the ratio of the carbon monoxide concentration to the carbon dioxide concentration is calculated. When the ratio is within the preset wood burning characteristic ratio range, a fire event is confirmed, and the confirmation signal is superimposed on the warning signal. By distinguishing biomass combustion from background concentration fluctuations through gas ratio characteristics, the existence of open flame or smoldering can be effectively confirmed, reducing the possibility of misjudgment based solely on thermal anomalies and vegetation changes.

[0018] In the implementation of issuing early warning signals, preferably, the early warning level is determined according to the degree to which the disaster risk index exceeds the elliptical confidence region. The early warning lights on the UAV are lit up according to the flashing frequency corresponding to the early warning level. At the same time, the UAV sends its own GPS coordinates and real-time optical images captured by the multispectral imager to the ground control center through a wireless communication link. The multi-level early warning and real-time image feedback enable ground personnel to immediately obtain the location, scope and severity of the fire, providing a basis for rapid dispatch of firefighting forces.

[0019] The present invention also provides a forest disaster drone monitoring and early warning system, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of the above-mentioned forest disaster drone monitoring and early warning method. The system can be integrated into a drone's onboard computer or a ground station, and utilizes the fusion of multi-source remote sensing and meteorological sensor information to achieve automated, highly sensitive, low false alarm monitoring and graded early warning of forest disasters.

[0020] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0021] The vegetation health index is generated by geometrically averaging the normalized difference vegetation index (NDVI) and the photosynthetically active radiation (PADR) component, replacing a single vegetation index or a simple arithmetic combination of the two. Geometric averaging has a suppressive effect on extreme values. When any index fluctuates abnormally due to atmospheric interference, shading, or sensor noise, geometric averaging can weaken the impact of this anomaly on the final vegetation health index, making the representation of vegetation status more robust. PADR is directly related to vegetation photosynthetic efficiency, while the normalized difference vegetation index reflects canopy greenness and density. These two indicators differ in their response mechanisms and sensitive time phases when vegetation is damaged. Geometric averaging can take into account both canopy structure degreening and decreased photosynthetic function as stress signals, revealing the declining trend of vegetation health in advance, even before significant changes occur in a single spectral index.

[0022] By constructing a temperature gradient matrix using surface temperature data and calculating information entropy to obtain the temperature field entropy value, the system captures the degree of spatial heterogeneity of the surface thermal field. In the early stages of forest fires, weak anomalous high-temperature points appear in areas with localized deadwood or dry litter accumulation, forming a temperature gradient with the surrounding normal vegetation. The surface temperature of normal, homogeneous forest stands exhibits a relatively gentle spatial distribution, with a low temperature field entropy value. As local heat sources begin to appear and develop, the temperature gradient increases, the spatial temperature difference information increases, and the temperature field entropy value rises accordingly. The sensitivity of the temperature field entropy value to spatial structure allows the system to detect anomalies through thermal field fragmentation characteristics even before the regional average temperature has significantly increased. Compared to traditional methods that rely solely on the magnitude of temperature rise or the area of ​​high-temperature zones, this method has a stronger ability to detect early, small-scale hotspots.

[0023] A disaster risk index is generated by multiplicatively fusing vegetation health index and temperature field entropy value. This fusion method establishes a multiplicative interaction between the two input variables. When vegetation health index decreases and temperature field entropy value increases, the disaster risk index rapidly amplifies, amplifying the dual anomalous signals of disaster precursors. Conversely, when only one indicator shows a slight anomaly, the product value remains at a low level, preventing an overly high response. Compared to fusion methods such as weighted summation or maximization, multiplicative fusion has higher specificity for the combined conditions of simultaneous vegetation degradation and thermal field disturbance, thereby reducing false alarms caused by fluctuations in a single data source.

[0024] An elliptical confidence region is fitted based on a two-dimensional distribution composed of vegetation health index and temperature field entropy values ​​from the same historical period in the same region. A normal baseline range from multivariate statistics is used as the early warning trigger condition. The elliptical confidence region uses the mean vector and covariance matrix to depict the joint distribution characteristics of the two variables in the same historical period. The critical value of the χ² distribution at the confidence level determines the elliptical boundary, which can adaptively characterize the coupling relationship between vegetation and the thermal field under normal seasonal fluctuations and interannual variations. An early warning is triggered when the disaster risk index exceeds the elliptical confidence region, meaning that the current observed value deviates significantly from the historical normal state in a joint sense, rather than simply due to a single variable exceeding the limit. This method allows the early warning threshold to automatically adjust with season, geographical location, and forest stand type, maintaining high sensitivity to real disaster events while reducing false alarms caused by natural phenomena such as seasonal drought or winter leaf fall. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0026] Figure 1 This is a flowchart of a method for monitoring and early warning of forest disasters using drones;

[0027] Figure 2 This is a flowchart of the temperature field entropy calculation process;

[0028] Figure 3 This is a flowchart of the generation of the elliptical confidence region and shadow correction for vegetation health index and temperature field entropy value;

[0029] Figure 4 This is a schematic diagram showing the distribution of normalized difference vegetation index and photosynthetically active radiation component;

[0030] Figure 5 It is the curve showing the relationship between the equilibrium moisture content of combustibles and the modulation factor corresponding to wind speed, as well as the data points from field fire spread experiments. Detailed Implementation

[0031] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, 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.

[0032] See Figure 1 This invention provides a method for monitoring and early warning of forest disasters using unmanned aerial vehicles (UAVs), comprising:

[0033] The system acquires vegetation reflectance spectra from a multispectral imager mounted on a drone and surface temperature data from a thermal infrared imager.

[0034] The normalized difference vegetation index and photosynthetically active radiation component are calculated based on the vegetation reflectance spectrum. The normalized difference vegetation index and the photosynthetically active radiation component are then geometrically averaged to generate a vegetation health index.

[0035] A temperature gradient matrix is ​​constructed based on the surface temperature data. The information entropy is calculated using each element in the temperature gradient matrix as a state value to obtain the temperature field entropy value.

[0036] The vegetation health index and the temperature field entropy value are fused by multivariate multiplication to generate a disaster risk index.

[0037] Based on the two-dimensional distribution of vegetation health index and temperature field entropy value in the same area during the same historical period, an elliptical confidence region is fitted. The disaster risk index exceeding the elliptical confidence region is used as the early warning trigger condition, and an early warning signal is issued.

[0038] Example 1

[0039] In practice, a multispectral imager is mounted on a drone to acquire raw radiance data in the visible and near-infrared bands. The multispectral imager's band settings include the red, near-infrared, and blue bands. Radiometric calibration is performed on the raw radiance data using the radiometric calibration coefficients set at the factory of the multispectral imager. These coefficients include a radiometric calibration gain coefficient and a radiometric calibration bias coefficient. During radiometric calibration, the raw radiance data of each pixel is multiplied by the corresponding radiometric calibration gain coefficient, and the radiometric calibration bias coefficient is added to obtain the radiance data for each band. Following radiometric calibration, atmospheric correction was performed on the radiance data. This correction employed a method based on the 6S radiative transfer model. The atmospheric parameters input for atmospheric correction included aerosol optical thickness, atmospheric water vapor content, ozone concentration, and observational geometric parameters. Aerosol optical thickness was obtained from the image itself using the dark target method. Atmospheric water vapor content and ozone concentration were acquired from reanalysis meteorological data. Observational geometric parameters included solar zenith angle, observational zenith angle, and relative azimuth angle. The atmospheric correction output was surface reflectance data. Red-band, near-infrared-band, and blue-band surface reflectance were extracted from the surface reflectance data. During extraction, the corresponding pixel value was determined based on the center wavelength of each band of the multispectral imager.

[0040] In practice, the normalized difference vegetation index is calculated by subtracting the surface reflectance in the red band from the surface reflectance in the near-infrared band, obtaining the difference result, summing the surface reflectance in the red band from the surface reflectance in the near-infrared band, obtaining the sum result, and dividing the difference result by the sum result to obtain the normalized difference vegetation index. Calculation of the photosynthetically active radiation (PAR) component: Divide the surface reflectance in the blue band by the surface reflectance in the red band to obtain the blue-red ratio. Substitute this blue-red ratio into a pre-calibrated PAR component conversion relationship. This relationship is a univariate linear regression equation established by ground-measured PAR components and the blue-red ratio. The slope coefficient and intercept coefficient of the univariate linear regression equation are determined by the least squares method. Multiply the blue-red ratio by the slope coefficient and add the intercept coefficient to obtain the PAR component. The slope coefficient is set to a value between -1.2 and -0.8, and the intercept coefficient is set to a value between 1.1 and 1.3. The specific values ​​of the slope coefficient and intercept coefficient are determined based on ground calibration experiments for different vegetation types. After obtaining the normalized difference vegetation index and the PAR component, a geometric mean is performed to generate the vegetation health index. The geometric mean calculation uses the following formula:

[0041]

[0042] in, Indicates the vegetation health index. Represents the normalized difference vegetation index. This represents the photosynthetically active radiation component. Before performing geometric averaging, [the following is a partial translation of the original text, which is incomplete and requires further context]. Perform non-negative processing if If it is less than zero, then... Set it to zero to ensure that the square root operation is valid in the real number field. The value is between 0 and 1, and the result is The value ranges from 0 to 1. The closer the value is to 1, the better the health of the vegetation.

[0043] See Figure 4 In the figure, the horizontal axis represents the Normalized Difference Vegetation Index (NDVI), ranging from 0 to 1, and the vertical axis represents the Photosynthetically Active Radiation Component (FPAR), also ranging from 0 to 1. The legend divides the Vegetation Health Index (VHI) into four intervals, each corresponding to a different symbol and color: blue dots represent samples with a VHI less than 0.25; orange squares represent samples with a VHI between 0.25 (inclusive) and 0.50; green triangles represent samples with a VHI between 0.50 (inclusive) and 0.75; and red diamonds represent samples with a VHI greater than or equal to 0.75.

[0044] The sample points in the figure show a clear distribution pattern: as the NDVI and FPAR values ​​increase, the range of VHI values ​​gradually increases, and the sample points gradually transition from the blue area in the lower left to the red area in the upper right, indicating a positive correlation between NDVI and FPAR. Blue dots are concentrated near NDVI below 0.25 and FPAR below 0.40, representing areas with poor vegetation condition; orange squares are mainly distributed in the range of NDVI approximately 0.20 to 0.45 and FPAR approximately 0.20 to 0.60, representing moderate to low vegetation health; green triangles are scattered in the range of NDVI approximately 0.45 to 0.80 and FPAR approximately 0.55 to 0.75, indicating good vegetation condition; red rhombuses are mainly distributed in areas where NDVI is above 0.70 and FPAR is above 0.70, reflecting good vegetation health.

[0045] Example 2

[0046] In specific implementation, please refer to Figure 2 The thermal infrared imager is mounted on a drone. Its detector array contains several rows and columns of pixels. After scanning the ground surface, the output of each pixel is converted into a surface temperature value through radiometric calibration and atmospheric correction. The surface temperature values ​​of all pixels are arranged in the row and column order of the detector array, forming a two-dimensional temperature matrix. The element in the r-th row and c-th column of the two-dimensional temperature matrix is ​​represented as... , where r is the row index and c is the column index.

[0047] In practice, for each pixel in the two-dimensional temperature matrix, the horizontal and vertical temperature differences are calculated. For pixels located outside the boundary, the horizontal temperature difference is obtained using a center difference method: extracting the temperature values ​​of the pixels adjacent to the right of the first pixel. Temperature value of the pixel to the left Calculate the horizontal temperature difference. , The vertical temperature difference is obtained using the center difference method: extracting the temperature value of the pixel directly below the pixel. Temperature values ​​of the pixels above and adjacent pixels Calculate the temperature difference in the vertical direction. , For pixels located on the left boundary, c=1, and the horizontal temperature difference is calculated using forward differencing: For pixels located on the right boundary, c=C, where C is the total number of columns in the two-dimensional temperature matrix. The horizontal temperature difference is calculated using backward difference. For pixels located at the upper boundary, r=1, and the vertical temperature difference is calculated using forward differencing: For pixels located at the lower boundary, r = R, where R is the total number of rows in the two-dimensional temperature matrix. The temperature difference in the vertical direction is calculated using backward difference. .

[0048] In practical implementation, the horizontal temperature difference value of each pixel is obtained. Temperature difference in the vertical direction Then, vector synthesis is performed to obtain the gradient magnitude. The synthesis method is After calculating the gradient magnitude for all pixels in the two-dimensional temperature matrix, all gradient magnitudes are arranged in their original row and column positions to form a temperature gradient matrix. The number of rows and columns in the temperature gradient matrix is ​​the same as that in the two-dimensional temperature matrix.

[0049] In practice, all gradient magnitude elements in the temperature gradient matrix are obtained, and the minimum and maximum values ​​of the gradient magnitudes are determined. The range between the minimum and maximum values ​​is then evenly divided into 20 intervals. The number of intervals is based on the following principle: when the total number of elements in the temperature gradient matrix is ​​in the thousands to tens of thousands, 20 intervals ensure that each interval contains a sufficient number of samples on average, guaranteeing the statistical stability of the probability estimation while preserving the structural characteristics of the gradient magnitude distribution. The number of gradient magnitude elements falling into each interval is counted, and the probability of the m-th interval is obtained by dividing the number of elements falling into the m-th interval by the total number of elements in the temperature gradient matrix. The calculation formula is: m is the interval index, ranging from 1 to 20. This represents the number of gradient magnitude elements falling within the m-th interval. This represents the total number of elements in the temperature gradient matrix, which is equal to the product of the number of rows and columns of the two-dimensional temperature matrix. Let R be the total number of rows in the two-dimensional temperature matrix, and C be the total number of columns in the two-dimensional temperature matrix. Based on the probabilities of all intervals... Calculate the entropy of the temperature field The calculation formula is:

[0050]

[0051] in, This represents the entropy value of the temperature field, where m is the interval index, ranging from 1 to 20. This represents the probability of the gradient magnitude occurring within the m-th interval. Indicates base 2 The logarithm of, when When the value is 0, it is stipulated that The value is 0. Temperature field entropy value. It reflects the uniformity of the gradient magnitude distribution in the temperature gradient matrix. The larger the value, the more disordered the spatial temperature variation.

[0052] Example 3

[0053] In practice, the drone is equipped with both a temperature and humidity sensor and an anemometer. The temperature and humidity sensor collects air temperature and humidity data at fixed time intervals, and the anemometer collects wind speed data at fixed time intervals. The air temperature is expressed as... Air humidity is expressed as , The measured value of air vapor pressure is directly output from the temperature and humidity sensor, and the unit is hectopascals (hPa). Wind speed data is expressed as... The unit is meters per second.

[0054] In practice, based on air temperature Calculate saturated water vapor pressure , The calculation uses the Magnus approximation formula, and during the calculation process, Substituting the numerical value into the operational expression corresponding to the Magnus approximation formula, we obtain the saturated vapor pressure. , The unit is hectopascals (hPa). (Based on air humidity) and saturated water vapor pressure Calculate relative humidity , pass Divide by get, It is a dimensionless ratio.

[0055] In practical implementation, relative humidity Substituting the empirical function for the equilibrium moisture content of wood, we obtain the equilibrium moisture content of the combustible material. The empirical function for the equilibrium moisture content of wood is expressed as a quadratic equation in one variable, representing the equilibrium moisture content of combustibles. The calculation formula is:

[0056]

[0057] in, Indicates the equilibrium moisture content of the combustible material. Indicates relative humidity. , and These are empirical coefficients. The value range is from -0.0020 to -0.0005, and the empirical coefficient is... The value ranges from 0.10 to 0.25, and the empirical coefficient is... The value ranges from 0.02 to 0.08, and the empirical coefficient is... , and The data acquisition method is as follows: Combustible samples of major tree species are collected from the forest area to be monitored. The combustible samples are placed in a constant temperature and humidity chamber, and different relative humidity gradients are set under a constant temperature of 25 degrees Celsius, covering a range of 20% to 95%. Under each relative humidity gradient, the combustible sample mass is allowed to reach equilibrium, and the moisture content at equilibrium is measured. Several sets of relative humidity-equilibrium moisture content data points are obtained. Least square quadratic polynomial fitting is performed on the data points to obtain empirical coefficients applicable to the forest area to be monitored. , and Equilibrium moisture content of combustible materials The value of is a dimensionless value between 0 and 1.

[0058] In practice, the preset lookup function is stored as a two-dimensional lookup table, with the horizontal axis of the two-dimensional lookup table corresponding to the equilibrium moisture content of the combustible material. The horizontal axis is divided into 20 intervals, covering a range from 0 to 1, with an interval width of 0.05; the vertical axis of the two-dimensional lookup table corresponds to the wind speed data. The vertical axis is divided into 10 intervals, covering a range from 0 to 20 meters per second, with each interval having a width of 2 meters per second. Each cell in the two-dimensional lookup table stores a modulation factor. The value of the modulation factor The value ranges from 0.5 to 1.5, and the modulation factor is... The rule for its value is: when the equilibrium moisture content of the combustible material is... The lower the wind speed data When it is higher, the modulation factor The closer the value is to 1.5, the higher the equilibrium moisture content of the combustible material. Higher and wind speed data The lower the modulation factor The closer the value is to 0.5, the higher the modulation factor of each cell in the two-dimensional lookup table. Specific values ​​were determined using data from field fire spread experiments. When querying the two-dimensional lookup table, the equilibrium moisture content of the combustible material was calculated in real time. and real-time collected wind speed data Determine the equilibrium moisture content of combustible materials Adjacent interval nodes on the horizontal axis and wind speed data The modulation factor is stored in four cells formed by four adjacent nodes on the vertical axis. The modulation factor is obtained by bilinear interpolation of the values. The output value.

[0059] In specific implementation, the vegetation health index and temperature field entropy value are obtained. The vegetation health index is obtained using the method described in Example 1, and the temperature field entropy value is obtained using the method described in Example 2. The vegetation health index and temperature field entropy value are then compared with a modulation factor. The product of these three factors serves as the disaster risk index, which is used to characterize the overall degree of risk of forest disasters occurring in the current region.

[0060] See Figure 5 In the figure, the horizontal axis represents the equilibrium moisture content (FMC) of the combustible material, ranging from 0 to 1, and the vertical axis represents the modulation factor α, ranging from 0.4 to 1.6. The legend indicates five curves showing the variation of the modulation factor α under different wind speeds, corresponding to wind speeds of 0 m / s (solid line), 5 m / s (dashed line), 10 m / s (dotted line), 15 m / s (dotted line), and 20 m / s (dotted-dotted line). The figure also uses gray scatter dots to indicate the data points from the field fire spread experiment.

[0061] Under all wind speed conditions, the modulation factor α decreased with increasing equilibrium moisture content (FMC), indicating that the modulation factor α was larger at lower FMCs and decreased as FMC increased. The variation range of the modulation factor α was approximately between 0.5 and 1.5, which is consistent with the setting range of the modulation factor α in Example 3. The curves for different wind speeds were generally arranged in parallel, with the higher the wind speed, the higher the position of the modulation factor α curve, indicating that under the same FMC conditions, the higher the wind speed, the larger the modulation factor α.

[0062] Example 4

[0063] In specific implementation, please refer to Figure 3 The historical records of vegetation health index and temperature field entropy values ​​collected by drones in the same month are accumulated. The accumulation method of the historical records is as follows: for the forest area to be monitored, a fixed spatial range is defined, and in the same month of several consecutive years, drones are used to collect data from multispectral imagers and thermal infrared imagers along a fixed route. The vegetation health index is calculated in the manner described in Example 1, and the temperature field entropy value is calculated in the manner described in Example 2. Each data collection yields a set of paired values ​​of vegetation health index and temperature field entropy value. All paired values ​​are stored in chronological order as a historical record dataset. Each set of data in the historical record dataset corresponds to a collection time.

[0064] In practice, a mean vector is calculated for the vegetation health index and temperature field entropy values ​​in the historical data dataset. This mean vector contains the mean of the vegetation health index. and the mean entropy of the temperature field , It is obtained by calculating the arithmetic mean of all vegetation health indices in the historical data dataset. The vegetation health index is obtained by calculating the arithmetic mean of all temperature field entropy values ​​in the historical data dataset. The covariance matrix is ​​then calculated; it is a 2x2 matrix, where the first element in the first row and first column represents the variance of the vegetation health index. , By calculating the vegetation health index of all historical data sets and The sum of squares of the differences is then divided by the total number of samples in the historical data set minus one; the elements in the second row and second column of the covariance matrix represent the variance of the temperature field entropy values. , By calculating the entropy values ​​of all temperature fields in the historical data dataset and... The sum of squares of the differences is then divided by the total number of samples in the historical data set minus one; the elements in the first row, second column and the second row, first column of the covariance matrix are both covariances. , By calculating the vegetation health index of each data set in the historical data dataset, The difference multiplied by the temperature field entropy value and The difference is obtained by summing all the products and dividing by the total number of samples in the historical data set minus one.

[0065] In practice, the covariance matrix is ​​decomposed into eigenvectors to obtain the first eigenvector and the second eigenvector. The first eigenvector corresponds to the largest eigenvalue of the covariance matrix, and the second eigenvector corresponds to the second largest eigenvalue of the covariance matrix. The first eigenvector and the second eigenvector are orthogonal to each other. The direction of the first eigenvector is determined to be the major axis of the elliptical confidence region, and the direction of the second eigenvector is determined to be the minor axis of the elliptical confidence region.

[0066] In practice, the radius of the elliptical confidence region is determined by the critical value of the χ² distribution at a confidence level of 0.95. The χ² distribution has 2 degrees of freedom, chosen because the elliptical confidence region is jointly composed of two variables: the vegetation health index and the temperature field entropy. The confidence region for a two-dimensional variable is determined using a χ² distribution with 2 degrees of freedom. The critical value of the χ² distribution with 2 degrees of freedom corresponding to a confidence level of 0.95 is 5.9915, and the radius of the ellipse's major axis is... and the minor axis radius of the ellipse The calculation formula is:

[0067]

[0068] in, This represents the radius of the k-th axis, when k is 1. This represents the radius of the major axis of the ellipse, when k is 2. Represents the radius of the minor axis of the ellipse. This represents the critical value of the χ² distribution with 2 degrees of freedom at a confidence level of 0.95. The value is 5.9915. This represents the k-th eigenvalue of the covariance matrix, where k is 1. This represents the largest eigenvalue when k is 2. This represents the second largest eigenvalue. (The mean vector is used as an example.) Centered on the first eigenvector, with the direction of the first eigenvector as the major axis and the direction of the second eigenvector as the minor axis, the length of the major axis is... The length of the minor axis is An elliptical confidence region is generated on the two-dimensional plane of vegetation health index-temperature field entropy value.

[0069] In practice, point cloud data from a lidar system mounted on a drone is acquired. The lidar emits laser beams to the ground in pulses and receives echo signals. Based on the time difference between the laser pulse emission and echo reception times, and the drone's positioning and orientation data, the three-dimensional coordinates of each laser footpoint are calculated, forming point cloud data. Forest canopy height and canopy closure are then extracted from the point cloud data. The forest canopy height extraction method involves: filtering the point cloud data using ground points to separate ground and non-ground point clouds; constructing a digital elevation model (DEM) and a digital surface model (DSM); subtracting the DEM from the DSM yields the canopy height model; within the canopy height model, a sliding window of a preset size traverses the entire monitoring area, and the maximum pixel value within the sliding window is taken as the forest canopy height. The canopy closure extraction method involves: counting the number of pixels in the canopy height model within the sliding window whose pixel value is greater than a preset height threshold; dividing this number by the total number of pixels within the sliding window yields the canopy closure.

[0070] In practice, forest canopy height and canopy density are input into the shading correction function, which is a linear weighted combination. The expression for the shading correction function is: correction coefficient. ,in This represents the correction factor. Indicates the height of the forest canopy. Indicates canopy closure. Indicates the weight of canopy height. Indicates the weight of canopy closure. Indicates the bias term. Canopy height weight. The value ranges from 0.005 to 0.015, representing the weight of canopy closure. The value range is from 0.2 to 0.5, and the bias term... The value ranges from 0.6 to 1.2, with canopy height weighting. Canopy closure weight and bias terms The method for obtaining the vegetation health index is as follows: Vegetation health index under different combinations of canopy height and canopy closure, and vegetation health index without shading conditions, are measured on the ground. A linear regression model between the correction coefficient and canopy height and canopy closure is established, and the canopy height weight is determined using the least squares method. Canopy closure weight and bias terms The value of . Using a correction factor. Multiplying by the vegetation health index yields the corrected vegetation health index, which is then used in the calculation of the subsequent disaster risk index to reduce the impact of forest canopy shadows on the vegetation health index.

[0071] Example 5

[0072] In practice, the drone is equipped with an electrochemical gas sensor, which includes a carbon monoxide sensing unit and a carbon dioxide sensing unit. The carbon monoxide sensing unit converts the concentration of carbon monoxide molecules in the air into an electrical signal through an electrochemical reaction, while the carbon dioxide sensing unit converts the concentration of carbon dioxide molecules in the air into an electrical signal through the principle of non-dispersive infrared absorption. After amplification and analog-to-digital conversion, the two electrical signals are used to obtain carbon monoxide concentration data and carbon dioxide concentration data, respectively. The carbon monoxide concentration data is in parts per million (ppm) by volume, and the carbon dioxide concentration data is in ppm by volume.

[0073] In practice, the carbon monoxide and carbon dioxide concentration data collected simultaneously are compared by dividing the carbon monoxide concentration data by the carbon dioxide concentration data to obtain the concentration ratio. A preset wood combustion characteristic ratio interval is stored in the processor's built-in read-only memory. The lower boundary of the wood combustion characteristic ratio interval is set to 0.02, and the upper boundary is set to 0.15. These boundaries are determined through laboratory wood combustion experiments. In the laboratory, wood samples of the main tree species in the forest area to be monitored are ignited, and the concentrations of carbon monoxide and carbon dioxide released during combustion are continuously measured using a gas analyzer. The ratio of carbon monoxide to carbon dioxide concentration is calculated, and the range of change in the ratio from open flame to smoldering is statistically analyzed. The lower boundary is taken as the lower 95% confidence interval of the lower limit, and the upper boundary is taken as the upper 95% confidence interval of the upper limit, resulting in a wood combustion characteristic ratio interval of [0.02, 0.15]. When the concentration ratio falls within the wood combustion characteristic ratio interval, the processor sets the fire event confirmation flag to true and generates a fire confirmation signal.

[0074] In practice, the process of superimposing the confirmation signal onto the warning signal is as follows: the processor performs a logical OR operation on the fire confirmation signal and the warning base signal triggered by the disaster risk index exceeding the elliptical confidence domain. When either the warning base signal or the fire confirmation signal is valid, the warning output is triggered.

[0075] In practical implementation, the degree to which the disaster risk index exceeds the elliptical confidence region is used to determine the warning level. The determination method is as follows: The correspondence between vegetation health index and temperature field entropy values ​​on the boundary of the elliptical confidence region is obtained from the historical data dataset. For each set of vegetation health index and temperature field entropy values ​​on the boundary, the same multivariate multiplication fusion method used to generate the disaster risk index is applied, combined with modulation factors calculated from real-time air temperature, air humidity, and wind speed data, to obtain the boundary disaster risk index. The maximum value in the boundary disaster risk index is used as the disaster risk index threshold value. The exceedance ratio of the real-time disaster risk index relative to the disaster risk index threshold value is calculated, and the exceedance ratio is determined by the following formula:

[0076]

[0077] in, Indicates the over-limit ratio. This indicates a real-time disaster risk index. This represents the threshold value for the disaster risk index. When... Less than or equal to hour, The value is zero. When the value is greater than zero and less than or equal to 0.1, the warning level is determined to be Level 1; when... When the value is greater than 0.1 and less than or equal to 0.3, the warning level is determined to be Level II; when When the value is greater than 0.3, the warning level is determined to be Level III.

[0078] In practice, the warning light on the drone is connected to the processor via its general-purpose input / output interface. The warning light uses a high-brightness LED array. The processor controls the duty cycle and period of the pulse width modulation signal according to the warning level, thus outputting the flashing frequency of the warning light. For warning level one, the flashing frequency is set to 1 Hz; for warning level two, the flashing frequency is set to 3 Hz; and for warning level three, the flashing frequency is set to 5 Hz. The warning light flashes periodically on and off according to the corresponding flashing frequency, remaining constantly lit during level three warning conditions.

[0079] In practice, the UAV acquires its own GPS coordinates in real time via a Global Navigation Satellite System (GNSS) receiver. These coordinates include latitude, longitude, and altitude data, updated at a frequency of 5 Hz. A multispectral imager simultaneously captures real-time optical images at the moment the warning is triggered; these images are visible light surface images. The processor frames the GPS coordinates and real-time optical images according to a standard network data packet format and transmits them via a wireless data transmission module onboard the UAV. This wireless transmission module uses an L-band microwave communication link to send the data frames to the ground control center.

[0080] In its implementation, a forest disaster drone monitoring and early warning system includes a memory, a processor, and a computer program stored in the memory and running on the processor. The memory employs a combination of non-volatile flash memory and dynamic random access memory (DRAM). The non-volatile flash memory stores the computer program code and historical data datasets, while the DRAM is used for temporary data storage during program execution. The processor is an embedded multi-core processor. When executing the computer program, the processor sequentially performs the following steps: acquiring vegetation reflectance spectrum and surface temperature data; calculating the vegetation health index; constructing a temperature gradient matrix and calculating the temperature field entropy value; generating a disaster risk index; fitting an elliptical confidence region and determining early warning conditions; and sending an early warning signal and superimposing gas sensor data for confirmation, thus implementing all the steps in the above method.

[0081] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for monitoring and early warning of forest disasters using unmanned aerial vehicles (UAVs), characterized in that, include: Acquire vegetation reflectance spectra from a multispectral imager mounted on a drone and surface temperature data from a thermal infrared imager. The normalized difference vegetation index and photosynthetically active radiation component are calculated based on the vegetation reflectance spectrum. The normalized difference vegetation index and the photosynthetically active radiation component are then geometrically averaged to generate a vegetation health index. A temperature gradient matrix is ​​constructed based on the surface temperature data. The information entropy is calculated by using each element in the temperature gradient matrix as a state value to obtain the temperature field entropy value. The vegetation health index and the temperature field entropy value are fused by multivariate multiplication to generate a disaster risk index; Based on the two-dimensional distribution of vegetation health index and temperature field entropy value in the same area during the same historical period, an elliptical confidence region is fitted. The disaster risk index exceeding the elliptical confidence region is used as the early warning trigger condition, and an early warning signal is issued.

2. The method for monitoring and early warning of forest disasters using unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, Obtaining the vegetation reflectance spectrum includes: Raw radiance data in the visible and near-infrared bands are acquired using a multispectral imager. Surface reflectance is obtained through radiometric calibration and atmospheric correction. Red band reflectance and near-infrared band reflectance are extracted from the surface reflectance to calculate the normalized difference vegetation index. Simultaneously, blue band reflectance and red band reflectance are extracted to calculate the photosynthetically active radiation component.

3. The method for monitoring and early warning of forest disasters using unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, A temperature gradient matrix is ​​constructed based on the surface temperature data, including: The temperature values ​​of each pixel acquired by the thermal infrared imager are used to form a two-dimensional temperature matrix. The horizontal temperature difference and the vertical temperature difference are calculated for each pixel in the two-dimensional temperature matrix. The horizontal temperature difference and the vertical temperature difference are vector-synthesized to obtain the gradient magnitude. The gradient magnitudes of all pixels are used to form the temperature gradient matrix.

4. The method for monitoring and early warning of forest disasters using unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, The vegetation health index and the temperature field entropy value are fused using a multivariate multiplication method to generate a disaster risk index, including: Acquire air temperature and humidity data collected by the temperature and humidity sensors on the drone, as well as wind speed data collected by the anemometer. The equilibrium moisture content of combustibles is calculated based on the air temperature and air humidity. The equilibrium moisture content of combustibles and the wind speed data are then substituted into a preset lookup function to obtain the modulation factor. The product of the vegetation health index, the temperature field entropy value, and the modulation factor is used as the disaster risk index.

5. A method for monitoring and early warning of forest disasters using unmanned aerial vehicles (UAVs) according to claim 4, characterized in that, Calculating the equilibrium moisture content of combustibles based on the air temperature and air humidity includes: An approximate formula based on thermodynamic equilibrium is used to convert the air temperature into saturated water vapor pressure and the air humidity into actual water vapor pressure. The relative humidity is obtained based on the ratio of the actual water vapor pressure to the saturated water vapor pressure. The relative humidity is then substituted into the empirical function of wood equilibrium moisture content to obtain the equilibrium moisture content of the combustible material.

6. A method for monitoring and early warning of forest disasters using unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, The fitted ellipse confidence region includes: The historical records of the vegetation health index and the temperature field entropy value collected by the drone in the same month are accumulated. The mean vector and covariance matrix of the two variables in the historical records are calculated. The radius of the ellipse is determined by taking the mean vector as the center, the direction of the eigenvector of the covariance matrix as the axis, and the critical value of the χ² distribution at a confidence level of 0.95, and the elliptical confidence region is generated.

7. A method for monitoring and early warning of forest disasters using unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, It also includes acquiring point cloud data from a lidar system mounted on a drone, extracting forest canopy height and canopy closure from the point cloud data, inputting the forest canopy height and canopy closure into a shading correction function, outputting a correction coefficient, and multiplying the correction coefficient by the vegetation health index to obtain the corrected vegetation health index.

8. A method for monitoring and early warning of forest disasters using unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, It also includes acquiring the carbon monoxide and carbon dioxide concentrations collected by the electrochemical gas sensor on the drone, calculating the ratio of the carbon monoxide concentration to the carbon dioxide concentration, and confirming a fire event when the ratio is within a preset range of wood burning characteristic ratios, and superimposing the confirmation signal onto the warning signal.

9. A method for monitoring and early warning of forest disasters using unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, The issuance of the warning signal includes: The warning level is determined based on the degree to which the disaster risk index exceeds the elliptical confidence region. The warning lights on the UAV are illuminated at a flashing frequency corresponding to the warning level. At the same time, the UAV transmits its own GPS coordinates and real-time optical images captured by the multispectral imager to the ground control center via a wireless communication link.

10. A forest disaster unmanned aerial vehicle (UAV) monitoring and early warning system, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the forest disaster drone monitoring and early warning method according to any one of claims 1 to 9.

Images