Unmanned aerial vehicle cluster automatic dispatching rescue system based on fire alarm dynamic matching

Abstract

This invention discloses an automatic dispatch and rescue system for drone swarms based on dynamic matching of fire alarms, specifically relating to the field of fire rescue technology. The system includes: a smoke and heat fusion sensing module that identifies thermal anomaly areas and outputs type labels, thermal radiation intensity, and smoke obstruction degree, with the obstruction degree dynamically adjusting the identification threshold; a fire prediction module that calculates a predicted fire chain propagation map; a task airspace re-segmentation module that adjusts the drone deployment order based on the prediction map and drone nest location; a fire chain suppression module that performs pre-deployment and fire extinguishing deployment, providing feedback on changes in area status; and a swarm self-stabilization feedback module that adjusts the sensing sampling frequency based on status changes and bandwidth occupancy. This invention improves the robustness of fire source identification by distinguishing between obscured and exposed hot areas through smoke and heat fusion and adaptively adjusting the threshold; it achieves pre-deployment of flame-retardant measures to suppress chain spread through nonlinear fire chain propagation prediction; and it achieves adaptive stabilization of system resources by dynamically adjusting the sensing sampling frequency based on monitoring bandwidth and fire status.

Description

Technical Field

[0001] This invention relates to the field of fire rescue technology, and more specifically, to an automatic dispatch and rescue system for drone swarms based on dynamic matching of fire alarms. Background Technology

[0002] With the maturity of drone technology, the fire and rescue field has begun to try to use drones for fire reconnaissance and initial fire suppression. Some areas have deployed automated drone nests, which, together with infrared thermal imaging cameras, can automatically identify fire points and dispatch them nearby.

[0003] However, in forest fires or high-rise building fires, the fire is often affected by wind, resulting in "flying fires"—burning debris is blown far away, igniting new fire points and forming a chain reaction of fire spread. Most existing drone firefighting systems are based on single-point fire source models and lack the ability to predict the propagation path of flying fires. When multiple flying fire zones appear almost simultaneously, they can often only passively respond to already burning points, unable to proactively deploy resources in potential landing areas, causing the allocation of firefighting resources to lag behind the speed of fire spread. Furthermore, dense smoke and tree canopies obscuring heat sources reduce the reliability of infrared identification, making it easy to miss hidden fire sources. Therefore, this invention proposes an automatic dispatch and rescue system for drone swarms based on dynamic fire alarm matching to address the above problems. Summary of the Invention

[0004] To achieve the above objectives, the present invention provides the following technical solution: The automatic dispatch and rescue system for drone swarms based on dynamic matching of fire alarms includes: The smoke and heat fusion sensing module receives multi-source data from infrared sensors, millimeter-wave radar, and spectral sensors, identifies all thermal anomaly areas, and outputs type labels and smoke and heat dynamic characteristics of each thermal anomaly area. The thermal anomaly areas include shielded thermal areas and directly visible exposed thermal areas. The type labels are used to distinguish between shielded and exposed thermal areas. The smoke and heat dynamic characteristics include the location coordinates, thermal radiation intensity values, and smoke obstruction degree of each thermal anomaly area. The smoke obstruction degree is used to dynamically adjust the first and third preset thresholds during the identification process. The fire prediction module receives the location coordinates and thermal radiation intensity values ​​of the thermal anomaly area, as well as the wind speed vector field and fuel humidity from real-time meteorological data. It obtains a fire chain propagation prediction map through nonlinear fire transition calculation. The fire chain propagation prediction map includes the location of potential fire landing areas, the expected trigger time, and the propagation probability of each landing area. The mission airspace reassignment module receives the fire chain propagation prediction map and the location of each UAV nest, and generates a mission queue adjustment command. The mission queue adjustment command is used to redistribute the UAV launch order and flight path of each UAV nest. The fire chain suppression module receives task queue adjustment instructions, controls the drone to perform forward deployment and fire suppression dropping, and updates the status of the thermal anomaly area after each dropping, providing feedback on the changes in the status of the thermal anomaly area. The cluster self-stabilizing feedback module receives the status changes of the thermal anomaly area and the communication bandwidth occupancy rate of each UAV nest, generates adjustment parameters, and feeds them back to the smoke and heat fusion sensing module. The adjustment parameters are used to adjust the sensing sampling frequency of the smoke and heat fusion sensing module.

[0005] In a preferred embodiment, the smoke and heat fusion sensing module identifies all thermal anomaly regions and outputs type labels in the following specific way: The thermal radiation intensity distribution map obtained by the infrared sensor and the penetration echo map obtained by the millimeter-wave radar are aligned at the pixel level. For continuous areas where the thermal radiation intensity exceeds the first preset threshold after alignment, a thermal anomaly candidate area is determined. Within the thermal anomaly candidate area, if the millimeter-wave radar echo intensity is lower than the second preset threshold and the spatial gradient change rate of the infrared thermal radiation intensity exceeds the third preset threshold, the thermal anomaly area is marked as a shielded thermal area; otherwise, it is marked as an exposed thermal area.

[0006] In a preferred embodiment, the smoke and heat fusion sensing module outputs the degree of smoke obstruction in the dynamic features of smoke and heat and dynamically adjusts the recognition threshold as follows: If the thermal anomaly area is labeled as an exposed thermal area, the smoke obscuration level is set to zero; if the thermal anomaly area is labeled as a shielded thermal area, the smoke obscuration level is calculated using the following steps: The first step is to collect the average brightness value of the marked shaded heat area in the visible light band and record it as the first brightness. The second step is to collect the average brightness value of the surrounding area without thermal anomalies and smoke in the visible light band, and record it as the second brightness. The third step is to determine whether the second brightness is greater than zero. If the second brightness is greater than zero, the first brightness is divided by the second brightness to obtain the brightness attenuation ratio. If the second brightness is equal to zero, the brightness attenuation ratio is directly set to zero. The fourth step is to subtract the brightness attenuation ratio from the constant to obtain the value of the smoke obstruction level. The fifth step is to multiply the smoke obscuration value by a preset adjustment coefficient to obtain the threshold correction amount. The first preset threshold is subtracted from the threshold correction amount to obtain the updated first preset threshold. The third preset threshold is added to the threshold correction amount to obtain the updated third preset threshold, which is used for thermal anomaly area identification in subsequent cycles.

[0007] In a preferred embodiment, the specific method for obtaining the predicted fire chain propagation map through nonlinear fire transition calculation in the fire prediction module is as follows: each identified thermal anomaly area is regarded as a propagation source, and the wind speed vector field and fuel humidity distribution map in real-time meteorological data are used as the environmental field. The propagation probability of each potential fire landing area is calculated using the following steps: The first step is to divide the downwind direction of each propagation source into multiple fan-shaped segments according to the wind speed, with each segment corresponding to a distance step size. The second step is to extract the average fuel moisture value within each sector segment and substitute it into the propagation probability function. The propagation probability function is an exponential decay function with fuel moisture as the independent variable, and output the segment propagation probability of that sector. The third step is to record the segment propagation probability of the first segment as the cumulative propagation probability of the first segment, and multiply the segment propagation probability of the Nth segment by the cumulative propagation probability of the Nth minus one segment to obtain the cumulative propagation probability of the Nth segment, where N is an integer greater than one. The fourth step is to mark the geographical locations covered by all segments whose cumulative propagation probability exceeds the propagation probability threshold as potential fire landing areas, and output the cumulative propagation probability of the segment as the propagation probability of the landing area.

[0008] In a preferred embodiment, the estimated trigger time in the fire prediction module is calculated as follows: Calculate the straight-line distance between each potential fireball impact zone and its corresponding propagation source, and record it as the distance value; Extract the wind speed component in the direction of propagation from real-time meteorological data and record it as the wind speed component value. Determine whether the absolute value of the wind speed component is greater than the preset minimum wind speed value. If the absolute value of the wind speed component is less than or equal to the preset minimum wind speed value, set the expected trigger time to a preset maximum time value. If the absolute value of the wind speed component is greater than the preset minimum wind speed value, continue to execute the subsequent steps. Divide the distance value by the wind speed component value to obtain the baseline flight time; The predicted trigger time is obtained by multiplying the baseline flight time by the fire transition velocity correction factor; the fire transition velocity correction factor is obtained by looking up the table based on the thermal radiation intensity value of the propagation source and the average fuel humidity along the propagation path.

[0009] In a preferred embodiment, the task space reassignment module generates task queue adjustment instructions in the following manner: According to the propagation probability of each potential fire landing zone in the fire chain propagation prediction map, they are sorted from largest to smallest. The landing zone with the highest propagation probability is assigned the nearest available drone nest. When the same drone nest is selected by multiple landing zones at the same time, the drones of the nest are arranged in order of propagation probability from largest to smallest. The nearest available drone nest is selected by comparing the straight-line distance between the location of each drone nest and the location of the potential fire landing zone.

[0010] In a preferred embodiment, the specific method by which the UAV is controlled to perform pre-deployment in the fire chain suppression module is as follows: When the fire chain propagation prediction map indicates that the propagation probability of a potential fire landing area exceeds a preset risk threshold, a drone carrying flame retardant is dispatched to fly over the landing area in advance before the landing area is actually ignited. The drone sprays flame retardant with the center of the landing area as the center and with a preset radius to form a flame-retardant isolation zone.

[0011] In a preferred embodiment, the status change of the thermal anomaly region fed back by the fire chain suppression module includes a suppressed region identifier, an unsuppressed region identifier, and a newly appearing region identifier; wherein the suppressed region identifier indicates that the thermal radiation intensity value of the thermal anomaly region has decreased by more than a preset decrease threshold within two consecutive sampling cycles, the unsuppressed region identifier indicates that the thermal radiation intensity value of the thermal anomaly region has not decreased by more than the preset decrease threshold, and the newly appearing region identifier indicates that a thermal anomaly region that did not exist in the previous sampling cycle has appeared in the current sampling cycle.

[0012] In a preferred embodiment, the specific method by which the cluster self-stabilizing feedback module generates adjustment parameters and feeds them back to the smoke and heat fusion sensing module is as follows: Real-time monitoring of the communication bandwidth occupancy rate of each UAV nest and the status changes of the thermal anomaly area, and setting a first threshold and a second threshold, wherein the value of the first threshold is greater than the value of the second threshold; When a new area identifier appears, the sensing sampling frequency in the adjustment parameters will be increased to twice the current value; When the communication bandwidth utilization rate exceeds the first threshold for three consecutive sampling periods, the sensing sampling frequency in the adjustment parameters will be reduced to half of the current value. When the communication bandwidth utilization rate is below the second threshold for three consecutive sampling periods and no new area identifier appears, the sensing sampling frequency in the adjustment parameters will be restored to a preset reference frequency. In other cases, the current sensing sampling frequency remains unchanged.

[0013] The technical effects and advantages of this invention are as follows: This invention significantly improves the accuracy and robustness of fire source identification in complex environments through a smoke and heat fusion sensing module. This module simultaneously receives multi-source data from infrared sensors, millimeter-wave radar, and spectral sensors, aligning the thermal radiation intensity distribution map with the penetration echo map at the pixel level. Continuous areas with thermal radiation intensity exceeding a first preset threshold are identified as candidate thermal anomaly areas. Based on this, by determining whether the millimeter-wave radar echo intensity is below a second preset threshold and whether the rate of change of the infrared thermal radiation spatial gradient exceeds a third preset threshold, it can clearly distinguish between concealed thermal areas obscured by smoke or tree canopies and directly visible exposed thermal areas, outputting corresponding type labels. The degree of smoke obscuration in the smoke and heat dynamic characteristics is used to dynamically adjust the first and third preset thresholds: when the smoke is dense, the first preset threshold is lowered to compensate for thermal radiation attenuation, while the third preset threshold is raised to avoid misjudging the smoke edge gradient as the flame edge. This closed-loop adaptive mechanism effectively solves the problem of missed detection of concealed fire sources due to dense smoke and vegetation obstruction, enabling the system to accurately perceive fires even under adverse visual conditions.

[0014] This invention achieves nonlinear prediction of the probability of fire chain propagation through a fire prediction module, elevating fire suppression lead time from "passive response" to "proactive pre-deployment." The fire prediction module treats each identified thermal anomaly area as a propagation source, using real-time wind speed vector field and fuel humidity distribution map as the environmental field. It divides the area into fan-shaped segments along the downwind direction, substituting the average fuel humidity within each segment into an exponential decay function to obtain the segment's propagation probability, and then calculates the cumulative propagation probability through a multiplicative method. Segments with a cumulative propagation probability exceeding the propagation probability threshold are marked as potential fire landing zones, and the estimated trigger time is calculated based on distance and wind speed components. Based on this, the fire chain suppression module schedules drones carrying flame retardants to fly to the landing zone before the estimated trigger time and spray flame retardants, forming a firebreak. This mechanism transforms traditional post-fire suppression into pre-fire prevention, particularly suitable for chain spread caused by fire jumping under strong wind conditions, effectively suppressing the sudden change in fire intensity from a single source to multiple sources.

[0015] This invention dynamically adjusts the sensing sampling frequency through a cluster self-stabilizing feedback module, achieving automatic matching of system resources with fire risk and avoiding task avalanche and communication congestion. The cluster self-stabilizing feedback module monitors the communication bandwidth utilization of each UAV nest and the status changes of thermal anomaly areas in real time. Through a multi-condition hierarchical adjustment mechanism, the system can adaptively switch from a stable monitoring mode to an emergency response mode, increasing data density during high-risk periods and proactively reducing energy consumption during resource-constrained periods, thereby maintaining the overall stability and reliability of operation. Attached Figure Description

[0016] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings; Figure 1This is a schematic diagram of the automatic dispatch and rescue system for drone swarms based on dynamic matching of fire alarms in this invention. Detailed Implementation

[0017] 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 of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0018] Reference Figure 1 The following examples were obtained: Example 1: An automatic dispatch and rescue system for drone swarms based on dynamic matching of fire alarms, comprising: The smoke and heat fusion sensing module receives multi-source data from infrared sensors, millimeter-wave radar, and spectral sensors, identifies all thermal anomaly areas, and outputs type labels and smoke and heat dynamic characteristics of each thermal anomaly area. The thermal anomaly areas include shielded thermal areas and directly visible exposed thermal areas. The type labels are used to distinguish between shielded and exposed thermal areas. The smoke and heat dynamic characteristics include the location coordinates, thermal radiation intensity values, and smoke obstruction degree of each thermal anomaly area. The smoke obstruction degree is used to dynamically adjust the first and third preset thresholds during the identification process. The fire prediction module receives the location coordinates and thermal radiation intensity values ​​of the thermal anomaly area, as well as the wind speed vector field and fuel humidity from real-time meteorological data. It obtains a fire chain propagation prediction map through nonlinear fire transition calculation. The fire chain propagation prediction map includes the location of potential fire landing areas, the expected trigger time, and the propagation probability of each landing area. The mission airspace reassignment module receives the fire chain propagation prediction map and the location of each UAV nest, and generates a mission queue adjustment command. The mission queue adjustment command is used to redistribute the UAV launch order and flight path of each UAV nest. The fire chain suppression module receives task queue adjustment instructions, controls the drone to perform forward deployment and fire suppression dropping, and updates the status of the thermal anomaly area after each dropping, providing feedback on the changes in the status of the thermal anomaly area. The cluster self-stabilizing feedback module receives the status changes of the thermal anomaly area and the communication bandwidth occupancy rate of each UAV nest, generates adjustment parameters, and feeds them back to the smoke and heat fusion sensing module. The adjustment parameters are used to adjust the sensing sampling frequency of the smoke and heat fusion sensing module.

[0019] In the smoke and heat fusion sensing module, the specific method for identifying all thermal anomaly areas and outputting type labels is as follows: The thermal radiation intensity distribution map acquired by the infrared sensor is aligned pixel-level with the penetration echo map acquired by the millimeter-wave radar. The thermal radiation intensity distribution map is a two-dimensional image captured by the infrared sensor, and each pixel value represents the thermal radiation intensity emitted by the object at that location, in degrees Celsius or Kelvin.

[0020] A penetration echo map is a map of the echo intensity received after a millimeter-wave radar emits electromagnetic waves. This map can penetrate smoke and some vegetation, reflecting the reflection characteristics of objects behind it. The unit of echo intensity is usually decibels (dB) and milliwatts (mW). Pixel-level alignment refers to mapping two images one-to-one in spatial coordinates, so that each pixel location simultaneously possesses both thermal radiation intensity and millimeter-wave echo intensity values. For continuous areas where the aligned thermal radiation intensity exceeds a first preset threshold, a candidate thermal anomaly region is identified.

[0021] The first preset threshold is set based on the statistical results of the maximum background thermal radiation when there is no fire in the local area, plus a redundancy. The typical value is 50 degrees Celsius higher than the ambient temperature. For example, when the ambient temperature is 25 degrees Celsius, the first preset threshold is 75 degrees Celsius.

[0022] A continuous region refers to a connected block composed of adjacent pixels. Adjacency determination uses either a four-neighbor or eight-neighbor rule, with eight-neighbor typically used to avoid missed detections. Within this thermal anomaly candidate region, if the millimeter-wave radar echo intensity is lower than a second preset threshold and the spatial gradient change rate of the infrared thermal radiation intensity exceeds a third preset threshold, then the thermal anomaly region is marked as a shielded thermal region; otherwise, it is marked as an exposed thermal region.

[0023] The second preset threshold is set based on the typical echo intensity of millimeter-wave radar in an unobstructed open area, minus an attenuation tolerance. A typical value is negative 1 / 2 BEmW. When the echo is below this value, it indicates the presence of strongly absorbing or scattering substances such as dense smoke or a dense tree canopy ahead. The spatial gradient rate of change refers to the degree of drastic change in thermal radiation intensity in the horizontal direction. It is calculated by summing the absolute values ​​of the differences in thermal radiation intensity between each point and its adjacent points within the candidate area, divided by the area of ​​the region.

[0024] The third preset threshold is set based on experimental statistics showing that the edge of a flame typically exhibits drastic changes in thermal radiation, while the thermal radiation gradient in smoke-covered areas is relatively gentle. A typical value is five degrees Celsius per meter. For example, suppose a forest area has an infrared image showing a continuous region with a thermal radiation intensity of 120 degrees Celsius, exceeding the first preset threshold of 75 degrees Celsius, and is thus identified as a candidate area for thermal anomalies. If the millimeter-wave radar echo intensity in this area is -15 dB / mW, lower than the second preset threshold of -10 dB / mW, and the spatial gradient change rate of the infrared thermal radiation intensity is 8 degrees Celsius per meter, exceeding the third preset threshold of 5 degrees Celsius per meter, then this area is marked as a shielded thermal zone, indicating that the fire source is obscured by tree canopies or dense smoke. If, within the same candidate thermal anomaly zone, the millimeter-wave radar echo intensity is -5 dB / mW (above the threshold) and the gradient change rate is 3 degrees Celsius per meter (below the threshold), then it is marked as an exposed thermal zone, indicating that the fire source is directly visible and unobstructed.

[0025] In the smoke and heat fusion sensing module, the method for outputting the degree of smoke obstruction in the dynamic features of smoke and heat and dynamically adjusting the recognition threshold is as follows: If the thermal anomaly area is labeled as an exposed thermal area, the smoke obstruction level is set to zero. An exposed thermal area indicates that the fire source is directly visible without obstruction, therefore there is no brightness reduction caused by smoke, and the smoke obstruction level is directly zero, requiring no further calculation. If the thermal anomaly area is labeled as a shielded thermal area, the smoke obstruction level is calculated using the following steps: A shaded hot zone indicates that the fire source is obscured by tree canopy or dense smoke. In this case, the brightness of the fire source area in the visible light image will be reduced due to smoke scattering and absorption. The degree of shading can be quantified by comparing the brightness ratio of the smoke-covered area with that of the smoke-free reference area.

[0026] The first step is to collect the average brightness value of the marked shaded hot area in the visible light band and record it as the first brightness. The visible light band refers to the combined brightness value of the red, green and blue channels obtained by a regular optical camera, such as the average value after converting the image to grayscale, with the unit being candela per square meter or dimensionless grayscale value.

[0027] The second step is to collect the average brightness value in the visible light band of the surrounding areas without thermal anomalies and without smoke reference areas, and record it as the second brightness. The areas without thermal anomalies refer to areas that have not been identified as thermal anomaly candidate areas, and the smoke-free reference areas are usually selected with clear skies or ground backgrounds under the same lighting conditions. For example, within 100 meters outside the shaded heat zone, take an area with the same vegetation type but without fire or smoke.

[0028] The third step is to determine whether the second brightness is greater than zero. If the second brightness is greater than zero, the first brightness is divided by the second brightness to obtain the brightness attenuation ratio. If the second brightness is equal to zero, the brightness attenuation ratio is directly set to zero. The second brightness being zero may occur in a completely dark environment or when the sensor is faulty. In this case, the division is meaningless, so the attenuation ratio is directly set to zero to indicate that it cannot be calculated. In actual processing, a default degree of occlusion will also be recorded.

[0029] The fourth step is to subtract the brightness attenuation ratio from the constant 1 to obtain the smoke obstruction level. The brightness attenuation ratio is between zero and one. The smaller the ratio, the more severe the obstruction. Therefore, subtracting the ratio from 1 gives the obstruction level. The larger the value, the denser the smoke.

[0030] The fifth step is to multiply the smoke obscuration value by a preset adjustment coefficient to obtain the threshold correction amount. The preset adjustment coefficient is calibrated based on experimental data, and the typical value range is from 0.5 to 2.0. For example, when the value is 1.0, the correction amount is directly equal to the obscuration value, and when the value is 0.5, the correction amount is halved.

[0031] Subtracting the threshold correction amount from the first preset threshold yields the updated first preset threshold. The first preset threshold is the threshold for judging thermal radiation intensity. Smoke attenuates infrared radiation, causing a decrease in the actual thermal radiation intensity of the fire source. Therefore, the threshold needs to be lowered to improve detection sensitivity. The physical meaning of subtraction is to compensate for smoke attenuation.

[0032] The third preset threshold is added to the threshold correction amount to obtain the updated third preset threshold, which is used for thermal anomaly region identification in subsequent cycles. The third preset threshold is the threshold of the rate of change of the spatial gradient of infrared thermal radiation. Smoke diffusion will generate its own temperature gradient (such as the temperature transition zone at the edge of smoke), which will cause a high gradient in non-fire source areas. Therefore, it is necessary to increase the threshold to avoid misjudging the edge of smoke as the edge of flame. The physical meaning of addition is to filter out the false gradient introduced by smoke.

[0033] The reason for only dynamically adjusting the first and third preset thresholds and not the second preset threshold is that the second preset threshold corresponds to the millimeter-wave radar echo intensity. Millimeter waves have a long wavelength and can penetrate smoke and tree canopies. Their echo intensity is generally not affected by smoke obstruction, so there is no need to adjust it.

[0034] For example: Suppose the first brightness of a shaded hot zone is 40 and the second brightness is 200. The brightness attenuation ratio is 0.2, and the smoke obstruction level is 1.0 minus 0.2, which equals 0.8. If the preset adjustment coefficient is 1.0, the threshold correction is 0.8. If the original first preset threshold was 75 degrees Celsius, it will be updated to 74.2 degrees Celsius; if the original third preset threshold was 5 degrees Celsius per meter, it will be updated to 5.8 degrees Celsius per meter. If the smoke is denser, the first brightness drops to 10, while the second brightness remains at 200. The obstruction level is then 0.95, the first threshold drops to 74.05 degrees Celsius, the third threshold rises to 5.95 degrees Celsius per meter, and so on.

[0035] Nonlinear fire transition refers to the abrupt behavior of a fire source jumping from surface fire to crown fire or crossing firebreaks via flying embers. This behavior cannot be described by a linear diffusion model, so a special algorithm is needed to calculate the probability of flying ember chain propagation. Each identified thermal anomaly area is regarded as a propagation source. The propagation source is the confirmed location of the fire source, such as the location coordinates of the exposed or shaded thermal area output by the previous module. Each propagation source independently generates a flying ember landing area prediction.

[0036] The wind speed vector field and fuel moisture distribution map in real-time meteorological data are used as the environmental field. The wind speed vector field includes the wind direction and wind speed at each grid point, in angle and meters per second, respectively. The fuel moisture distribution map is a vegetation moisture content map obtained by interpolation from satellite or ground monitoring stations, in percentage.

[0037] The following steps are used to calculate the propagation probability of each potential fireball impact zone: The first step involves dividing the leeward direction of each propagation source into multiple fan-shaped segments based on wind speed, with each segment corresponding to a distance step size. The leeward direction refers to the direction indicated by the wind speed vector; for example, if the wind direction is 30 degrees east of north, the leeward direction is 30 degrees east of north. The angle of the fan-shaped segments is set based on experience or experimentation, typically 30 degrees, ensuring the coverage area is neither too wide (leading to false positives) nor too narrow (leading to false negatives). The distance step size is determined based on the drone nest deployment interval and fire suppression response time, typically 50 meters, meaning each 50-meter segment is divided. The maximum predicted distance is usually between 500 and 2000 meters. Wind speed affects the maximum distance the fire jumps; higher wind speeds require more segments. The total number of segments equals the maximum predicted distance divided by the distance step size, rounded down.

[0038] The second step involves extracting the average fuel moisture value for each sector and substituting it into the propagation probability function. This propagation probability function is an exponentially decaying function with fuel moisture as the independent variable, outputting the sector propagation probability. The average fuel moisture value refers to the arithmetic mean of the fuel moisture at all grid points within the geographical area covered by the sector. Lower fuel moisture results in a higher probability of a stray spark igniting a new fire point.

[0039] The exponential decay function takes the form: the segment propagation probability equals the baseline probability multiplied by the negative slope of the natural constant e multiplied by the fuel moisture power, where the baseline probability is the maximum probability without fuel moisture limitations, typically taking a value of 0.8, and the slope is determined based on experimental data, typically taking a value of 0.05. For example, when the fuel moisture is 20%, the segment propagation probability is 0.8 multiplied by e to the power of -0.05, which equals 0.8 multiplied by 0.3679, approximately 0.294. When the fuel moisture is zero, the segment propagation probability is 0.8, and when the fuel moisture is 100%, it is close to 0.8 multiplied by 0.0067, approximately 0.005.

[0040] The third step is to record the segment propagation probability of the first segment as the cumulative propagation probability of the first segment. Then, multiply the segment propagation probability of the Nth segment by the cumulative propagation probability of the Nth (minus one)th segment to obtain the cumulative propagation probability of the Nth segment, where N is an integer greater than one. The physical meaning of this multiplicative recursion is that after the fire starts from the propagation source, it must successively traverse each segment to reach a more distant segment. Therefore, the probability of reaching the Nth segment is equal to the probability of successfully traversing the first N (minus one) segments multiplied by the probability of successfully igniting within the Nth segment.

[0041] For example: Suppose the segment propagation probability of the first segment is 0.294, then the cumulative propagation probability of the first segment is 0.294. If the segment propagation probability of the second segment is 0.200, then the cumulative propagation probability of the second segment is 0.294 multiplied by 0.200, which equals 0.0588. If the segment propagation probability of the third segment is 0.150, then the cumulative propagation probability of the third segment is 0.0588 multiplied by 0.150, which equals 0.00882, and so on. The cumulative probability decreases rapidly with increasing distance.

[0042] The fourth step is to mark the geographical locations covered by all segments whose cumulative propagation probability exceeds the propagation probability threshold as potential fire landing areas, and output the cumulative propagation probability of the segment as the propagation probability of the landing area.

[0043] The propagation probability threshold is set based on the minimum tolerable response probability of the fire suppression system. According to the statistical results of actual fire drills, a typical value is 0.05, meaning that it is necessary to deploy fire suppression resources in advance for sections where the cumulative probability exceeds 5%. For example, if the cumulative probability of the third section mentioned above is 0.00882, which is less than 0.05, it is not marked as a potential flying fire zone; if the cumulative probability of the second section is 0.0588, which is greater than 0.05, the fan-shaped area covered by that section is marked as a potential flying fire zone with a propagation probability of 0.0588.

[0044] Each potential fireball impact zone includes the geographical boundary of that segment (e.g., a fan-shaped area with a radius ranging from fifty to one hundred meters and an angle of thirty degrees centered on the propagation source), which is used by subsequent modules to allocate UAV nests for pre-deployment. The reason for using the above-mentioned multiplicative exponential decay model to calculate the propagation probability is that fireball jumping is a multi-stage dependent process. The sparks generated by the fire source continuously lose energy and are affected by fuel moisture during their flight. The probability of successful ignition after each distance is the product of independent events, so using cumulative multiplication is consistent with physical reality.

[0045] Meanwhile, fuel humidity has an exponentially decreasing relationship with the ignition probability, because when humidity increases, water evaporation requires more heat, making it difficult for flying sparks to ignite. Compared with the fixed-radius circular diffusion model, this method can distinguish the asymmetric flying spark risk under different wind directions and distances, and retains directional information through sector segment division, making the prediction map closer to the actual fire behavior.

[0046] In the fire early warning module, the estimated trigger time is calculated as follows: Calculate the straight-line distance between each potential fire landing zone and its corresponding propagation source, and record it as the distance value. The straight-line distance refers to the Euclidean distance from the center point of the propagation source (e.g., the coordinates of the fire source location) to the geometric center of the potential fire landing zone (e.g., the midpoint of the sector segment), and the unit is meters.

[0047] Extract the wind speed component value in the direction of propagation from real-time meteorological data, and denot it as the wind speed component value. The direction of propagation refers to the direction from the propagation source to the center of the potential fire landing area. The wind speed component value is equal to the projection of the total wind speed vector in that direction, and the unit is meters per second. The calculation formula is the total wind speed magnitude multiplied by the cosine value of the angle between the wind direction and the direction of propagation.

[0048] Determine whether the absolute value of the wind speed component is greater than the preset minimum wind speed value. The preset minimum wind speed value is set based on the minimum effective measurement accuracy of the meteorological sensor and the minimum wind speed at which the fire can drift. According to multiple field fire experiment data, the typical value is 0.5 meters per second.

[0049] If the absolute value of the wind speed component is less than or equal to the preset minimum wind speed, it indicates extremely low wind speed or headwind, making it difficult for the fire to spread effectively. In this case, the estimated trigger time is set to a preset maximum time value. The preset maximum time value is set based on the maximum standby time of the drone and the upper limit of the fire suppression resource rescheduling cycle, typically set to 300 seconds (five minutes), indicating that the arrival time of the fire under this wind speed condition exceeds the system's waiting tolerance limit. If the absolute value of the wind speed component is greater than the preset minimum wind speed value, the subsequent steps continue.

[0050] Dividing the distance by the wind speed component yields the baseline flight time. The baseline flight time, in seconds, represents the time required for sparks to travel at a constant linear velocity at the wind speed. Multiplying this baseline flight time by the fire propagation velocity correction factor gives the predicted triggering time. This correction factor is a dimensionless factor used to correct for the difference between the actual fire propagation velocity and the wind speed, as sparks are affected by air resistance, rising thermals, and other factors, resulting in a horizontal propagation velocity that is typically lower than the wind speed.

[0051] The fire propagation speed correction factor is obtained by looking up the thermal radiation intensity value of the propagation source and the average fuel moisture content along the propagation path. The experimental data used for looking up the table comes from standard fire dynamics tests (such as the flying fire propagation speed database published by the International Union of Forest Research Organizations). The thermal radiation intensity in the table is divided into three levels: low, medium, and high (typical values: below 300 degrees Celsius per square meter is low, 300 to 600 is medium, and above 600 is high). The average fuel moisture content is divided into three levels: dry, moderate, and wet (typical values: below 30% is dry, 30% to 60% is moderate, and above 60% is wet). The corresponding correction factor ranges from 0.3 to 0.9.

[0052] For example: the heat radiation intensity is 500 degrees Celsius per square meter (medium level), the average fuel moisture is 40% (moderate level), and the correction factor is 0.6 from the table. If the distance is 150 meters and the wind speed component is 5 meters per second, then the base flight time is 30 seconds, and the estimated trigger time is 30 seconds multiplied by 0.6 equals 18 seconds.

[0053] The time it takes for a fire to drift from its source to its landing zone directly impacts the lead time for pre-deployment. Too short an estimated trigger time will result in drones not having enough time to reach the landing zone, while too long a time wastes resources. The directional component of wind speed reflects the asymmetry of wind direction on the fire's path, avoiding errors introduced by the isotropic assumption. Introducing a preset minimum wind speed avoids division-by-zero errors when the wind speed is zero, and also reflects that under windless conditions, fire mainly relies on thermal buoyancy for propagation, making the time unpredictable. The fire transition speed correction coefficient utilizes existing experimental data through a lookup table, which is more accurate than a single fixed coefficient without increasing online computational complexity.

[0054] The mission airspace reconfiguration module receives the predicted fire chain propagation map and the location of each UAV nest, and generates a mission queue adjustment command. This command is used to reallocate the UAV deployment order and flight paths of each nest, ensuring that firefighting resources are prioritized for deployment to the highest-risk fire impact areas. The specific method for generating the mission queue adjustment command is as follows: The potential fire landing zones are sorted from largest to smallest according to their propagation probabilities in the fire chain propagation prediction diagram. The propagation probability is a value output by the preceding module, ranging from 0 to 1. For example, if the propagation probabilities of the three potential fire landing zones A, B, and C are 0.12, 0.08, and 0.05 respectively, then the sorting result is A, B, and C.

[0055] In the task airspace reassignment module, the specific method for generating task queue adjustment instructions is as follows: sorting the potential fire landing areas in the fire chain propagation prediction map from largest to smallest. This sorting ensures that high-risk landing areas are given priority in resource allocation, and the landing area with the highest propagation probability is allocated the nearest available UAV nest. The prerequisite for this allocation method is that the UAVs in all available UAV nests have the same flight speed, at which point the distance is shortest, i.e. the flight time is shortest.

[0056] If the drones have different flight speeds, the allocation will be based on the available drone nest with the shortest estimated arrival time. The estimated arrival time is equal to the straight-line distance between the nest location and the landing area location divided by the flight speed of the drone in that nest. When the same nest is selected by multiple landing areas at the same time, the drones in that nest will be dispatched in order of their propagation probability from highest to lowest.

[0057] If there are multiple drones in the same nest with the same speed, they are dispatched in order of probability. If the speeds are different, the priority of flight time and probability needs to be comprehensively evaluated. In this case, a weighted scoring method can be used, where the score is equal to the propagation probability divided by the estimated arrival time, and the drone with the higher score is dispatched first.

[0058] The nearest available drone nest is selected by comparing the straight-line distance between the nest location and the potential fire impact zone. When the speeds are the same, the straight-line distance is directly compared; when the speeds are different, the estimated arrival time of each nest is calculated first, and then the time lengths are compared.

[0059] The reason for adopting this allocation method is that the fire extinguishing window is usually only a few minutes, and the time it takes for the drone to arrive at the scene is the key to the success or failure of fire extinguishing. In a simplified scenario where the speed is the same, the closest distance directly corresponds to the shortest time, with the least amount of calculation, which is suitable for real-time scheduling. In actual scenarios where the speed is different, allocating according to the estimated arrival time can more accurately reflect the accessibility of resources and avoid delays in actual arrival due to ignoring speed differences.

[0060] In the fire chain suppression module, the specific method for controlling the drone to perform pre-deployment is as follows: When the fire chain propagation prediction map indicates that the propagation probability of a potential fire landing area exceeds a preset risk threshold, a drone carrying flame retardant is dispatched to fly over the landing area in advance before the landing area is actually ignited. The drone sprays flame retardant with the center of the landing area as the center and with a preset radius to form a flame-retardant isolation zone.

[0061] "Ahead of schedule" here means that the flame retardant spraying is completed before the estimated trigger time of the potential fire landing zone. The estimated trigger time is the time value calculated by the pre-fire observation module, which represents the estimated arrival time of the stray sparks from the propagation source to the landing zone. The pre-deployed drones must arrive and complete the spraying before this time in order to effectively block the ignition of the fire source.

[0062] The preset risk threshold is set based on the fire department's classification standard for flying fire risk levels. According to statistics from multiple forest fire operations, preventive measures are recommended when the propagation probability is greater than or equal to 0.05 (i.e., 5%). Therefore, the typical value of the preset risk threshold is 0.05. For example, if the calculated cumulative propagation probability of a certain landing area is 0.0588, exceeding 0.05, then pre-deployment is triggered.

[0063] The preset radius is set based on the effective coverage area of ​​the flame retardant carried by a single drone. Depending on the nozzle flow rate and flight speed of the drone spraying equipment, the typical value is 20 to 30 meters, for example, 25 meters, to ensure that a continuous isolation zone is formed around the center of the landing area.

[0064] The specific timing requirements for scheduling drones to arrive above the landing area in advance are as follows: The flight time (distance divided by flight speed) required for a drone to take off from its nest and reach the landing area, plus the spraying operation time (e.g., five seconds), must be less than the expected trigger time. If the sum of the flight time and the operation time exceeds the expected trigger time, the system should select a closer nest or pre-arrange multiple drones to operate simultaneously. For example, if the expected trigger time for a potential fire landing area is eighteen seconds, and it takes twelve seconds for a drone to fly from the nearest nest to the landing area, plus five seconds for spraying, the total time is seventeen seconds, which is less than eighteen seconds. This meets the "advance" condition, and pre-deployment can be performed normally. If the total time is twenty seconds, which is greater than eighteen seconds, the system determines that pre-deployment cannot be completed. In this case, the system will instead dispatch a fire-fighting drone to suppress the fire after it is ignited in the landing area, thus downgrading the deployment from prevention to firefighting.

[0065] The reason for adopting a pre-deployment approach is that the ignition of sparks is sudden and short-lived; once a spark lands and comes into contact with dry fuel, a new ignition point can form within seconds. If drones are deployed only after the ignition point appears, the optimal suppression window will be missed, and multiple fire extinguishing bomb drops will be needed to control the fire. By spraying flame retardant before the expected trigger time provided by the pre-application module, the moisture content of the fuel in the landing area can be increased to a level that makes it difficult to ignite, thereby cutting off the chain propagation path of the sparks and preventing a fire from breaking out.

[0066] The status changes of the thermal anomaly area reported by the fire chain suppression module include the suppressed area marker, the unsuppressed area marker, and the newly appeared area marker; the suppressed area marker indicates that the thermal radiation intensity value of the thermal anomaly area has decreased by more than the preset decrease threshold within two consecutive sampling cycles.

[0067] The preset reduction threshold is set based on the typical attenuation rate of the heat radiation intensity of the fire source after the fire extinguishing bomb is thrown. According to the experimental data of the drone fire extinguishing bomb being thrown in multiple fire drills, a decrease in heat radiation intensity of more than 50% within three seconds is considered as effective suppression. Therefore, the typical value of the preset reduction threshold is 50%.

[0068] The sampling interval between two consecutive sampling cycles is related to the sensing sampling frequency of the smoke and heat fusion sensing module. For example, if the sensing sampling frequency is 1 Hz and the sampling interval is 1 second, then two consecutive sampling cycles mean that the decrease exceeds 50% within two seconds. For instance, if the thermal radiation intensity of a certain thermal anomaly area was 120 degrees Celsius in the previous sampling cycle and is 50 degrees Celsius in the current sampling cycle, the decrease is approximately 58.3% (70 divided by 120), which exceeds 50%, and is therefore marked as a suppressed area.

[0069] An unsuppressed area is marked as an area where the decrease in thermal radiation intensity does not exceed a preset threshold. For example, if the thermal radiation intensity in the same area decreases from 120 degrees Celsius to 90 degrees Celsius, the decrease is 30 divided by 120, which equals 25%. If the decrease is less than 50%, it is marked as an unsuppressed area.

[0070] Upon receiving an unsuppressed area marker, the fire chain suppression module sends a reinforcement request to the mission airspace reassignment module. The mission airspace reassignment module then reassigns additional drones from the same or adjacent drone nests, carrying fire extinguishing bombs, to perform another fire extinguishing drop. Alternatively, if the drones have sufficient battery power and ammunition, they can simply repeat the drop. If the fire is not suppressed after two consecutive attempts, the warning level is raised, and the information is sent to the command center, recommending the dispatch of ground firefighting forces.

[0071] The newly appearing area identifier indicates that a thermal anomaly area that was not present in the previous sampling period has appeared in the current sampling period. This identifier is used to determine whether a new fire source has been ignited by a stray fire or has changed from a concealed state to an exposed state. When a newly appearing area identifier is received, the cluster self-stabilizing feedback module uses this identifier as a trigger condition to increase the sensing sampling frequency. At the same time, the task spatial re-division module treats this new area as a new propagation source, re-executes the stray fire chain propagation prediction and task allocation of the fire situation observation module, and inserts it into the task queue, sorted by propagation probability.

[0072] For example: If there is no thermal anomaly area at a certain coordinate location in the previous sampling period, and an area with a thermal radiation intensity of 130 degrees Celsius appears at that location in the current sampling period, it will be marked as a newly appeared area. The system will immediately add the coordinate to the propagation source list, calculate the possible fire impact zone, and assign it to a drone.

[0073] The reason for adopting the above feedback mechanism is that: suppressed indicators are used to release fire-fighting resources to avoid repeated dispatching in the same area; unsuppressed indicators trigger secondary fire-fighting or escalation responses to prevent the fire from reigniting; newly emerging indicators are used to dynamically adjust the prediction model and task queue to achieve real-time tracking of the fire's evolution and form a complete perception-decision-execution-feedback closed loop.

[0074] In the cluster self-stabilizing feedback module, the specific method for generating adjustment parameters and feeding them back to the smoke and heat fusion sensing module is as follows: The system monitors the communication bandwidth utilization rate and thermal anomaly status changes of each drone nest in real time, setting a first threshold and a second threshold, where the first threshold value is greater than the second threshold value. Communication bandwidth utilization rate refers to the ratio of currently used communication bandwidth to total available bandwidth, expressed as a percentage, such as the measured value of a 4G or 5G link.

[0075] The first threshold is set based on the critical point at which the communication link is about to become congested. Based on statistics of the minimum bandwidth required for drone swarm control, a typical value is 70%, indicating that packet loss or delay may occur when the occupancy rate exceeds 70%. The second threshold is set based on a relatively idle communication link, typically set at 30%, indicating that when the occupancy rate is below 30%, there is excess bandwidth that can be used to increase the sampling frequency. When a new area identifier appears, the sensing sampling frequency in the adjustment parameters is increased to twice the current value.

[0076] The appearance of a new area marker indicates the emergence of a new fire source, requiring faster detection to track the fire's spread. Therefore, the sampling frequency is immediately doubled, for example, from the default 1 Hz to 2 Hz. The decision to double the frequency is based on experimental results showing that doubling improves temporal resolution without excessively increasing bandwidth; higher multipliers, such as quadrupling, would lead to excessive data volume. When the communication bandwidth occupancy exceeds the first threshold for three consecutive sampling cycles, the sensing sampling frequency in the adjustment parameters is reduced to half its current value. Three consecutive sampling cycles are used to prevent misjudgments caused by momentary jitter. Each sampling cycle is related to the sensing sampling frequency; for example, at a sampling frequency of 1 Hz, three cycles are three seconds.

[0077] A typical scenario for halving the sampling frequency: If the current sampling frequency is 2 Hz, and the bandwidth utilization rate reaches 75% exceeding the first threshold for three consecutive seconds, then the frequency will be reduced to 1 Hz. The reduced sampling frequency must not be lower than a preset minimum frequency (e.g., 0.2 Hz) to ensure basic monitoring capabilities.

[0078] When the communication bandwidth utilization rate is below the second threshold for three consecutive sampling periods and no new area markers appear, the sensing sampling frequency in the adjustment parameters will be restored to a preset reference frequency. The preset reference frequency is set based on the normal sampling requirements of the system under stable conditions. According to the time resolution requirements of fire inspection tasks, a typical value is 0.5 Hz (sampling once every two seconds). For example, if the current sampling frequency is reduced to 0.25 Hz due to high bandwidth, and then the bandwidth utilization rate is below 30% for three consecutive seconds and no new fire source appears, it will be restored to 0.5 Hz.

[0079] In other cases, the current sensing sampling frequency remains unchanged, specifically including: when the bandwidth utilization rate is between the first and second thresholds, or when the bandwidth is below the second threshold but a new area identifier appears, the doubling rule is executed first, or when the bandwidth exceeds the first threshold but less than three cycles have been completed, etc.

[0080] The reason for adopting this adjustment mechanism is that: if the sensing sampling frequency is too high, it will occupy a lot of communication bandwidth and increase the computing load; if it is too low, it may miss the rapid spread of fire. Through dynamic adjustment, the system actively increases the sampling rate to capture details when there is a high risk, i.e. when a new fire source appears, and actively reduces the frequency when the bandwidth is tight to ensure that control commands are transmitted first. Setting three consecutive periodic conditions avoids frequent fluctuations caused by signal noise, making the system behavior more stable.

[0081] The above-mentioned models or function formulas are all dimensionless and numerical calculations. The models or function formulas are obtained by software simulation based on a large amount of collected data to obtain the most recent real situation. The preset parameters in the models or function formulas are set by those skilled in the art according to the actual situation.

[0082] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0083] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0084] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0085] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations 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. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An automatic dispatch and rescue system for drone swarms based on dynamic matching of fire alarms, characterized in that, include: The smoke and heat fusion sensing module receives multi-source data from infrared sensors, millimeter-wave radar, and spectral sensors, identifies all thermal anomaly areas, and outputs type labels and smoke and heat dynamic characteristics of each thermal anomaly area. The thermal anomaly areas include shielded thermal areas and directly visible exposed thermal areas. The type labels are used to distinguish between shielded and exposed thermal areas. The smoke and heat dynamic characteristics include the location coordinates, thermal radiation intensity values, and smoke obstruction degree of each thermal anomaly area. The smoke obstruction degree is used to dynamically adjust the first and third preset thresholds during the identification process. The fire prediction module receives the location coordinates and thermal radiation intensity values ​​of the thermal anomaly area, as well as the wind speed vector field and fuel humidity from real-time meteorological data. It obtains a fire chain propagation prediction map through nonlinear fire transition calculation. The fire chain propagation prediction map includes the location of potential fire landing areas, the expected trigger time, and the propagation probability of each fire landing area. The mission airspace reassignment module receives the fire chain propagation prediction map and the location of each UAV nest, and generates a mission queue adjustment command. The mission queue adjustment command is used to redistribute the UAV launch order and flight path of each UAV nest. The fire chain suppression module receives task queue adjustment instructions, controls the drone to perform forward deployment and fire suppression dropping, and updates the status of the thermal anomaly area after each dropping, providing feedback on the changes in the status of the thermal anomaly area. The cluster self-stabilizing feedback module receives the status changes of the thermal anomaly area and the communication bandwidth occupancy rate of each UAV nest, generates adjustment parameters, and feeds them back to the smoke and heat fusion sensing module. The adjustment parameters are used to adjust the sensing sampling frequency of the smoke and heat fusion sensing module.

2. The automatic dispatch and rescue system for unmanned aerial vehicle (UAV) swarms based on dynamic matching of fire alarms according to claim 1, characterized in that, The specific method for identifying all thermal anomaly regions and outputting type labels is as follows: The thermal radiation intensity distribution map obtained by the infrared sensor and the penetration echo map obtained by the millimeter-wave radar are aligned at the pixel level. For continuous areas where the thermal radiation intensity exceeds the first preset threshold after alignment, a thermal anomaly candidate area is determined. Within the thermal anomaly candidate area, if the millimeter-wave radar echo intensity is lower than the second preset threshold and the spatial gradient change rate of the infrared thermal radiation intensity exceeds the third preset threshold, the thermal anomaly area is marked as a shielded thermal area; otherwise, it is marked as an exposed thermal area.

3. The automatic dispatch and rescue system for unmanned aerial vehicle (UAV) swarms based on dynamic matching of fire alarms according to claim 2, characterized in that, The method for dynamically adjusting the smoke obstruction level in the output smoke-thermal dynamic characteristics is as follows: If the thermal anomaly area is labeled as an exposed thermal area, the smoke obscuration level is set to zero; if the thermal anomaly area is labeled as a shielded thermal area, the smoke obscuration level is calculated using the following steps: In the area where the marked shading heat zone is located, the average brightness value of the area in the visible light band is collected and recorded as the first brightness; The average brightness value in the visible light band of a reference area with no thermal anomalies and no smoke around the area is collected and recorded as the second brightness. Determine if the second brightness is greater than zero. If the second brightness is greater than zero, divide the first brightness by the second brightness to obtain the brightness attenuation ratio. If the second brightness is equal to zero, directly set the brightness attenuation ratio to zero. Subtract the brightness attenuation ratio from the constant to obtain the value of the smoke obstruction level; The threshold correction amount is obtained by multiplying the smoke obscuration value by a preset adjustment coefficient. The first preset threshold is subtracted from the threshold correction amount to obtain the updated first preset threshold. The third preset threshold is added to the threshold correction amount to obtain the updated third preset threshold, which is used for thermal anomaly area identification in subsequent cycles.

4. The automatic dispatch and rescue system for unmanned aerial vehicle (UAV) swarms based on dynamic matching of fire alarms according to claim 3, characterized in that, The specific method for obtaining the predicted fire chain propagation map through nonlinear fire transition calculation is as follows: Each identified thermal anomaly area is considered a propagation source, and the wind speed vector field and fuel humidity distribution map in real-time meteorological data are used as the environmental field. The following steps are used to calculate the propagation probability of each potential fireball impact zone: Centered on each propagation source, multiple fan-shaped segments are divided downwind according to wind speed, with each fan-shaped segment corresponding to a distance step. For each sector segment, extract the average fuel moisture value within that sector segment and substitute it into the propagation probability function. The propagation probability function is an exponential decay function with fuel moisture as the independent variable, and output the sector propagation probability of that sector segment. The segment propagation probability of the first sector segment is denoted as the cumulative propagation probability of the first sector segment. The segment propagation probability of the Nth sector segment is multiplied by the cumulative propagation probability of the Nth minus one sector segment to obtain the cumulative propagation probability of the Nth sector segment, where N is an integer greater than one. The geographical locations covered by sector segments whose cumulative propagation probability exceeds the propagation probability threshold are marked as potential fire landing areas, and the cumulative propagation probability of the sector segment is output as the propagation probability of the fire landing area.

5. The automatic dispatch and rescue system for unmanned aerial vehicle (UAV) swarms based on dynamic matching of fire alarms according to claim 4, characterized in that, The estimated trigger time is calculated as follows: Calculate the straight-line distance between each potential fireball impact zone and its corresponding propagation source, and record it as the distance value; Extract the wind speed component in the direction of propagation from real-time meteorological data and record it as the wind speed component value. Determine whether the absolute value of the wind speed component is greater than the preset minimum wind speed value. If the absolute value of the wind speed component is less than or equal to the preset minimum wind speed value, set the expected trigger time to a preset maximum time value. If the absolute value of the wind speed component is greater than the preset minimum wind speed value, continue to execute the subsequent steps. Divide the distance value by the wind speed component value to obtain the baseline flight time; The predicted trigger time is obtained by multiplying the baseline flight time by the fire transition velocity correction factor. The fire transition rate correction factor is obtained by looking up the thermal radiation intensity value of the propagation source and the average fuel humidity along the propagation path from a table.

6. The automatic dispatch and rescue system for unmanned aerial vehicle (UAV) swarms based on dynamic matching of fire alarms according to claim 5, characterized in that, The specific method for generating task queue adjustment instructions is as follows: According to the propagation probability of each potential fire landing area in the fire chain propagation prediction map, they are sorted from largest to smallest. The nearest drone nest is assigned to the fire landing area with the highest propagation probability. When the same nest is selected by multiple landing areas at the same time, the drones of the drone nest are arranged in order of descending propagation probability.

7. The automatic dispatch and rescue system for unmanned aerial vehicle (UAV) swarms based on dynamic matching of fire alarms according to claim 6, characterized in that, The specific methods for controlling drones to perform pre-deployment are as follows: When the fire chain propagation prediction map indicates that the propagation probability of a potential fire landing area exceeds a preset risk threshold, before the fire landing area is actually ignited, a drone carrying flame retardant is dispatched to fly over the landing area in advance and spray flame retardant with the center of the fire landing area as the center and a preset radius to form a flame-retardant isolation zone.

8. The automatic dispatch and rescue system for unmanned aerial vehicle (UAV) swarms based on dynamic matching of fire alarms according to claim 7, characterized in that, The feedback on the status changes of thermal anomaly areas includes suppressed area markers, unsuppressed area markers, and newly appearing area markers. Among them, the suppressed area marker indicates that the thermal radiation intensity value of the thermal anomaly area has decreased by more than a preset decrease threshold within two consecutive sampling periods; the unsuppressed area marker indicates that the thermal radiation intensity value of the thermal anomaly area has not decreased by more than the preset decrease threshold; and the newly appearing area marker indicates that a thermal anomaly area that did not exist in the previous sampling period has appeared in the current sampling period.

9. The automatic dispatch and rescue system for unmanned aerial vehicle (UAV) swarms based on dynamic matching of fire alarms according to claim 8, characterized in that, The specific method for generating adjustment parameters and feeding them back to the smoke and heat fusion sensing module is as follows: Real-time monitoring of the communication bandwidth occupancy rate of each UAV nest and the status changes of the thermal anomaly area, and setting a first threshold and a second threshold, wherein the value of the first threshold is greater than the value of the second threshold; When a new area identifier appears, the sensing sampling frequency in the adjustment parameters will be increased to twice the current value; When the communication bandwidth utilization rate exceeds the first threshold for three consecutive sampling periods, the sensing sampling frequency in the adjustment parameters will be reduced to half of the current value. When the communication bandwidth utilization rate is below the second threshold for three consecutive sampling periods and no new area identifier appears, the sensing sampling frequency in the adjustment parameters will be restored to a preset reference frequency. In other cases, the current sensing sampling frequency remains unchanged.

Images