A smart forest fire suppression system based on a six-axis rotary-wing UAV

By combining a six-axis rotary-wing UAV with airflow sensing and ballistic compensation modules, fire type identification, and infrared inspection technology, the precise delivery of fire extinguishing bombs and the quantitative processing of smoldering points in forest fires have been achieved. This solves the problems of accuracy and thoroughness in existing fire extinguishing schemes and improves the efficiency and safety of forest fire fighting.

CN123101918APending Publication Date: 2026-11-17SHANGHAI DEXINYI AVIATION TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611483781.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-22
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing drone-based firefighting solutions lack the ability to systematically perceive and adaptively handle complex fire environments, resulting in deviations between the calculated trajectory of fire extinguishing projectiles and their actual impact points, insufficient identification of fire types, and incomplete handling of smoldering points. This makes it difficult to achieve precise firefighting and timely removal of hidden dangers in forest fires.

Method used

A six-axis rotary-wing UAV is used in conjunction with an airflow sensing and ballistic compensation module. Through multi-source wind field sensing and iterative solution of the mass ballistic equation, the precise delivery of fire extinguishing bombs is achieved. Dual-light cameras and semantic segmentation networks are used to identify the fire type and dynamically adjust the fire extinguishing strategy. After the open flame is extinguished, infrared inspection is carried out, and the thermal influence depth of the smoldering point is estimated using a temperature decay model for precise re-extinguishing.

Benefits of technology

It improves the accuracy of fire extinguishing grenade throwing and the fire suppression effect, enhances the thoroughness of smoldering and the efficiency of fire extinguishing agent utilization, reduces the risk of reignition, and enhances the continuity and environmental adaptability of UAV autonomous fire extinguishing operations.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123101918A_ABST
    Figure CN123101918A_ABST
Patent Text Reader

Abstract

This invention discloses an intelligent forest fire fighting system based on a six-axis rotary-wing UAV, specifically relating to the field of forest fire fighting technology. The system includes an airflow sensing and ballistic compensation throwing module, a fire type identification and stratified fire suppression module, and an infrared inspection and precise extinguishing module for smoldering embers. The airflow sensing module acquires multi-source wind fields and performs vector superposition using an ultrasonic anemometer, a downwash flow field reduction model, and a fire plume model. It then iteratively solves the delivery compensation angle using ballistic equations to improve the accuracy of the fire extinguishing projectile delivery. The fire type identification module uses a dual-light camera and a semantic segmentation network to distinguish between crown fires and surface fires. Crown fires are treated with flame retardant atomized spray, while surface fires are dynamically calculated based on the fire spread rate to form a flame-retardant isolation zone. The smoldering ember inspection module identifies smoldering points using threshold segmentation and Kalman filtering. This invention achieves a closed-loop fire fighting operation from fire point targeting and stratified fire suppression to smoldering elimination.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of forest fire fighting technology, and more specifically, to an intelligent forest fire fighting system based on a six-axis rotary-wing unmanned aerial vehicle (UAV). Background Technology

[0002] Forest fires are frequent natural disasters worldwide, characterized by their suddenness, rapid spread, and difficulty in extinguishing, posing a serious threat to the ecological environment and the safety of people's lives and property. Traditional forest fire fighting methods mainly rely on ground firefighting forces, including firefighters carrying firefighting equipment into the fire scene for manual firefighting and creating firebreaks, as well as aerial water droplets or fire retardant spraying by manned firefighting aircraft or helicopters. However, ground firefighting forces have limited mobility when facing complex terrains such as steep mountains and canyons, and firefighters are directly exposed to high temperatures, dense smoke, and flying embers, posing extremely high safety risks. Manned aircraft firefighting is limited by high flight costs, demanding takeoff and landing conditions, and insufficient capabilities for operation at night and in dense smoke environments. It is also difficult to continuously and accurately locate and track fire points, and the accuracy of delivery is difficult to guarantee under complex and variable wind conditions at the fire scene. In addition, after a forest fire is extinguished, a large number of smoldering points often remain in the underground humus layer, which are very easy to reignite under the action of wind. Traditional inspection methods rely on manual ground inspection or observation from lookout towers, which are inefficient and have limited coverage, making it difficult to identify and completely eliminate hidden dangers in the early stages of a fire or after the open flames have been extinguished.

[0003] In recent years, with the rapid development of drone technology, multi-rotor drones, with their advantages of vertical takeoff and landing, stable hovering, high maneuverability, and relatively low cost, have gradually been applied to early reconnaissance and auxiliary firefighting in forest fires. Existing technologies include drones equipped with thermal imaging cameras for fire detection or carrying fire extinguishing bombs for deployment. However, most existing drone firefighting solutions remain at the level of single-function integration, lacking the ability to systematically perceive and adaptively respond to the complex environment of a fire scene. Specifically, existing solutions generally ignore or simplify the impact of the downwash flow field of the rotorcraft itself on the trajectory of the fire extinguishing projectile, and also fail to incorporate the thermal lifting effect of the fire plume into the wind field model, resulting in significant deviations between the trajectory calculation and the actual impact point. In terms of fire type identification, existing solutions often rely solely on simple threshold segmentation using single-spectral images, which cannot effectively distinguish between crown fires and surface fires, nor can they dynamically adjust the fire extinguishing strategy according to the fire spread, resulting in problems such as the deployment location deviating from the fire line and the flame-retardant zone being discontinuous. In the residual fire inspection stage after the open flame is extinguished, existing solutions lack the ability to accurately identify and quantitatively assess smoldering points, usually only performing heat source detection alarms, and cannot automatically calculate the required fire extinguishing dosage and implement precise point-to-point re-extinguishing, making it easy for smoldering points to be missed or not thoroughly dealt with.

[0004] To address the aforementioned issues, the industry urgently needs an intelligent forest fire suppression system that integrates multi-source wind field perception and ballistic compensation, possesses automatic fire type identification and tiered fire suppression decision-making capabilities, and performs infrared inspection and precise extinguishing of smoldering points. Existing publicly available literature lacks a complete closed-loop fire suppression scheme that couples a downwash flow field reduction model with a fire plume model for fire extinguishing projectile trajectory correction, combines semantic segmentation networks for fire type identification and coordinated fire suppression mode switching, and uses a temperature decay model to estimate the thermal impact depth of smoldering points to achieve quantitative calculation of fire extinguishing ball usage. Therefore, developing an intelligent forest fire suppression system based on a six-axis rotary UAV that can overcome the shortcomings of existing technologies has significant practical implications and application value. Summary of the Invention

[0005] To overcome the above-mentioned deficiencies of the prior art, embodiments of the present invention provide an intelligent forest fire fighting system based on a six-axis rotary unmanned aerial vehicle.

[0006] To achieve the above objectives, the present invention provides the following technical solution: An intelligent forest fire fighting system based on a six-axis rotary-wing UAV includes the following modules: The airflow sensing and ballistic compensation throwing module is used to sense the multi-source wind fields of natural wind, rotor downwash and fire plume in real time through airborne sensors and generate the total equivalent wind field by vector superposition. Combined with the mass ballistic equation, the landing point deviation is iteratively calculated. Based on the landing point deviation, the servo throwing rail is driven to adjust the throwing angle so that the fire extinguishing bomb is thrown towards the fire point along the compensation direction. The fire type identification and layered fire suppression module is used to acquire images through dual-light cameras and use semantic segmentation networks to identify crown fires and ground fires. When it is identified as a crown fire, water-based flame retardant is sprayed in atomized form. When it is identified as a ground fire, the deployment interval and lead time are dynamically calculated based on the fire spread rate. Dry powder fire extinguishing bombs are continuously deployed to form a flame-retardant isolation zone. The smoldering fire infrared inspection and precise extinguishing module is used to switch to low-altitude inspection after the open flame is extinguished. It uses threshold segmentation and Kalman filtering to track and identify smoldering points, estimates the thermal impact depth based on the temperature decay model, calculates the required number of extinguishing balls, and throws extinguishing balls point by point to eliminate smoldering.

[0007] Specifically, in the airflow sensing and ballistic compensation throwing module: An airborne ultrasonic anemometer acquires three-dimensional wind vectors, and an edge computing unit reads IMU data to calculate airspeed and ground speed. Based on the motor speed, a pre-calibrated downwash flow field reduced-order model is invoked to output the three components of rotor induced velocity at different elevations below the deployment height in real time; The natural wind vector and the downwash induced velocity vector are superimposed to construct the equivalent crosswind profile of the fire extinguishing bomb from the drop point to the ground; the equivalent crosswind profile is superimposed with the fire plume vector to generate the total equivalent wind field.

[0008] Specifically, the downwash flow field reduced-order model is constructed based on intrinsic orthogonal decomposition and radial basis function interpolation. It takes motor speed and incoming wind speed as input and outputs the three components of rotor induced velocity at different elevations from zero to twenty meters below the deployment height in real time. The calibration process is as follows: multiple incoming wind speed levels and motor speed levels are set in the wind tunnel. A hot-wire anemometer array is arranged within the elevation range directly below the delivery point. The measured values ​​are used as the reference true values. The intrinsic orthogonal decomposition mode is selected to make the cumulative energy contribution rate reach more than 95%. The radial basis function interpolation shape parameters are optimized and selected through leave-one-out cross-validation. The root mean square error of the model output is controlled within 0.15 meters per second.

[0009] Specifically, in the airflow sensing and ballistic compensation throwing module: The edge computing unit extracts the average radiation temperature and flame projection area of ​​the fire point area from the thermal infrared image of the dual-light camera, and reads the current ambient atmospheric temperature to estimate the average rising velocity of the fire plume center. The total equivalent wind field vector is generated by superimposing the fire plume velocity vector and the equivalent crosswind profile vector. The ballistic equation of the mass considering the wind field is used and solved iteratively by the fourth-order Runge-Kutta method to predict the horizontal deviation of the uncontrolled landing point relative to the target fire point. The pitch compensation angle and azimuth compensation angle required for the servo-deployed slide rail are calculated based on the horizontal deviation. The compensation angle is converted into PWM commands to drive the servo motor to adjust the slide rail to the compensation attitude before releasing the fire extinguishing projectile.

[0010] Specifically, in the fire type identification and stratified fire suppression module: The dual-light camera simultaneously acquires visible light and thermal infrared images, achieves frame-level alignment through hardware timestamps, and performs homography transformation using pre-calibrated camera parameters to register the thermal infrared image to the visible light image coordinate system. The edge computing unit runs a semantic segmentation network to extract the flame region from the visible light image, and generates a temperature field based on the radiometric calibration parameters of the thermal infrared image, and obtains the height distribution of the maximum temperature point in each frame based on the projective geometry relationship.

[0011] Specifically, combining the distances from the drone to the top of the tree canopy and to the ground obtained by the laser rangefinder, if the temperature peak areas are concentrated within a height range of ±1.5 meters from the top of the tree canopy and the thermal infrared image shows a cluster of high-temperature rising shapes, then it is determined to be a tree canopy fire. The aforementioned cluster-like high-temperature floating morphology refers to the following: the aspect ratio of the circumscribed rectangle of the connected domain of the high-temperature region in the thermal infrared image is between 0.7 and 1.5, the vertical upward displacement of the centroid of this region in three consecutive frames of images is more than twice the horizontal displacement, and the elevation of the highest temperature point in this region migrates upward at a speed of not less than 0.5 meters per second.

[0012] Specifically, when a crown fire is identified, the edge computing unit controls the electric push rod to extend the spray bar and start the high-pressure water pump to spray the water-based flame retardant in the airborne water tank into the base of the crown fire in a cone-shaped atomization mode. At the same time, the water pump flow rate is adjusted by the PID controller until the flame area is reduced to less than five percent of the total image area. When the fire is identified as a surface fire, the drone maintains a safe operating height of 15 meters relative to the ground surface. The edge computing unit acquires two frames of thermal infrared images, one from the current moment and the other from 0.5 seconds ago, extracts the minimum bounding rectangle of the surface fire area, and calculates the instantaneous spread rate of the fire line.

[0013] Specifically, the edge computing unit reads the effective suffocation radius of the fire extinguishing bomb and calculates the dynamic deployment interval based on the fire spread rate and the critical spread speed of the fire extinguishing bomb; If the fire spread rate reaches or exceeds the critical spread rate, switch to the highest density deployment mode and set the deployment interval to the minimum safe distance allowed by the system. Otherwise, the lead time for deployment is calculated, and the servo deployment rail continuously deploys dry powder fire extinguishing bombs at dynamic intervals and in front of the leading edge of the fire line. After the application or spraying is completed and the system has been allowed to stand, the dual-light camera will perform a second data collection and evaluation. If active fire points are still detected, the process will be repeated; otherwise, the process will exit and the fire point will be marked as cleared.

[0014] Specifically, in the residual fire infrared inspection and precise extinguishing module: The UAV scans at low altitude along a bow-shaped flight path. The edge computing unit performs Gaussian filtering on the thermal infrared image and then runs an adaptive threshold segmentation algorithm to extract candidate hot spot blocks. Kalman filtering is used to track and match the candidate hot spots in multiple consecutive frames. After successful association and smooth motion in multiple consecutive frames, the hot spots are confirmed as real smoldering points and their three-dimensional spatial coordinates are locked. An exponential decay model was used to fit and obtain the initial temperature of the smoldering point, the ambient background temperature, and the temperature decay coefficient. Based on this, the thermal influence depth was estimated. The required inerting volume was calculated by combining the actual surface area of ​​the smoldering point and the safety margin. Then, the required number of fire extinguishing balls was calculated based on the effective mass of the fire extinguishing ball and the fire extinguishing design concentration. The control pan-tilt unit pointed the dispenser sequentially to the coordinates of each aiming point and sent a current pulse to the heating wire fuse mechanism to make the fire extinguishing ball fall freely along the aiming direction and hit the smoldering point. The thermal imager continuously monitored the temperature of the point. If the temperature dropped below the threshold and did not rise, the mark was cleared. Otherwise, additional fire extinguishing balls were added until all were cleared before returning to base.

[0015] The technical effects and advantages of this invention are as follows: This invention incorporates a downwash flow field reduction model and the thermal lift effect of the fire plume into the trajectory calculation of the fire extinguishing projectile. By superimposing the effects of rotor-induced velocity and fire source thermal convection on natural wind, the synthesized wind field more closely resembles the actual airflow environment near the fire point. Furthermore, by iteratively solving the delivery compensation angle using the centroid trajectory equation, it effectively reduces the impact point deviation of the fire extinguishing projectile under complex fire weather conditions, improving delivery accuracy and the reliability of hitting the target fire point. Simultaneously, it utilizes dual-light image registration and a semantic segmentation network to extract the flame region, and combines the spatial distribution characteristics of the temperature field to distinguish between crown fires and surface fires. For different fire types, it initiates atomized spraying or continuous delivery of fire extinguishing projectiles, and can dynamically adjust the impact interval according to the fire spread rate and reserve lead time at the fire front edge, helping to form a continuous and reliable fire-retardant isolation zone, improving the efficiency of extinguishing agent utilization and the fire suppression effect.

[0016] This invention, after extinguishing the open flame, transitions to a low-altitude, bow-shaped inspection. Through local threshold segmentation and Kalman filtering with multi-frame correlation, it confirms the true smoldering point, reducing false alarm interference from sunlight reflection and residual heat from rocks. Then, it uses exponential fitting of the smoldering point's cooling curve to obtain the temperature decay coefficient, thereby inferring the thermal impact depth and quantitatively calculating the inerting volume and fire extinguishing ball dosage. This shifts smoldering point management from experience-based judgment to data-driven quantitative assessment, improving the thoroughness of re-extinguishing and the economy of pesticide application. The entire firefighting process forms a closed loop from front-end fire point targeting, mid-stage stratified fire suppression, to rear-end smoldering elimination. The coordinated operation of each module helps reduce the risk of forest fire reignition, decreases the frequency of firefighters entering dangerous areas, and enhances the continuity and environmental adaptability of autonomous drone firefighting operations. Attached Figure Description

[0017] Figure 1 This is a system block diagram of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] like Figure 1 As shown, a smart forest fire fighting system module based on a six-axis rotary-wing UAV is as follows: The airflow sensing and ballistic compensation throwing module uses airborne sensors to detect and vector-superimpose multi-source wind fields such as natural wind, rotor downwash, and fire plume in real time. It then iteratively calculates the impact point deviation using the particle trajectory equation, driving a servo rail to automatically correct the throwing angle. This enables high-precision throwing of fire extinguishing bombs under complex wind conditions. The specific working steps are as follows: The airborne ultrasonic anemometer acquires the three-dimensional wind vector at the location of the UAV at a frequency of 20Hz. At the same time, the edge computing unit reads the three-axis attitude, acceleration and barometric altimeter data output by the IMU and calculates the airspeed and ground speed of the UAV. Based on the current six-axis motor speed fed back by the ESC and the known blade geometry parameters, a downwash flow field reduced-order model, pre-calibrated through CFD simulation and wind tunnel testing and stored in the onboard memory, is invoked. This model is constructed based on intrinsic orthogonal decomposition and radial basis function interpolation, taking motor speed and incoming wind speed as inputs, and outputting in real time the three components of rotor-induced velocity at different elevations from 0 to 20 meters below the deployment height. Specifically, the calibration process involves setting multiple incoming wind speed levels in the wind tunnel, ranging from zero to 15 meters per second, and multiple incoming wind speed levels in the hovering state of the UAV, ranging from 3,000 to 6,000 revolutions per minute. For each motor speed level, a hot-wire anemometer array is placed every half meter within an elevation range of 0 to 20 meters directly below the deployment point, and its measured values ​​are used as the baseline true values. During model construction, the intrinsic orthogonal decomposition modes are sorted from high to low according to their energy contribution rate, and the top few principal modes are selected so that the cumulative energy contribution rate reaches more than 95%. For radial basis function interpolation, a Gaussian kernel function is used, and its shape parameters are optimized and selected through leave-one-out cross-validation. The model output must be verified by wind tunnel measured data, and its root mean square error with the measured value must be controlled within 0.15 meters per second before it can be installed and used.

[0020] The three-dimensional natural wind vector measured by the ultrasonic anemometer is vector-superimposed with the downwash induced velocity at the corresponding height level to construct the equivalent crosswind profile of the fire extinguishing bomb from the deployment point to the ground; after the third step is completed, the edge computing unit extracts the average radiation temperature of the fire point area from the latest frame of the thermal infrared image from the dual-light camera. (Unit: K) and flame projection area (unit: (and read the current ambient air temperature) Estimate the average upward velocity of the fire plume center using the following formula. : ,in The dimensionless empirical constant is taken as 0.15. Scalar of gravitational acceleration , Based on area Calculated equivalent fire source diameter ; The result Construct the vertically upward velocity vector of the fire plume And with the equivalent crosswind profile vector constructed in the third step Perform vector superposition to generate the total equivalent wind field vector. ; The target position is obtained by acquiring the slant distance of the UAV relative to the treetop apex or ground fire point as measured by a laser rangefinder. Given the mass, drag coefficient, and reference area of ​​the fire extinguishing projectile, a point mass ballistic equation considering the wind field is adopted. (in For the projectile velocity, The total equivalent wind field vector, It is the acceleration due to gravity. air density, The drag coefficient of the fire extinguishing projectile. (where m is the reference area of ​​the fire extinguishing projectile and m is the mass of the fire extinguishing projectile) is solved using the fourth-order Runge-Kutta method with a high-speed iterative solution at a step size of 0.01 seconds to predict the horizontal deviation of the uncontrolled landing point relative to the target fire point. The edge computing unit calculates the direction correction required for the initial velocity vector of the deployment based on the horizontal deviation, and solves the pitch compensation angle and azimuth compensation angle required for the multi-degree-of-freedom servo deployment rail. The rail consists of a universal bracket driven by two orthogonal servo motors and the rail body. The edge computing unit converts the compensation angle into a PWM command to drive the servo motor to adjust the rail to the compensation attitude. In the next millisecond after the attitude sensor confirms that the position is in place and the UAV is hovering stably, a release command is issued to accurately deploy the dry powder fire extinguishing projectile towards the fire point along the compensation direction.

[0021] The fire type identification and stratified fire suppression module uses dual-light cameras and a semantic segmentation network to automatically identify crown fires and surface fires. For crown fires, it employs water-based flame retardant atomized spraying. For surface fires, it dynamically calculates the deployment interval and lead time based on the fire spread rate, continuously deploying dry powder extinguishing bombs to form an adaptive flame-retardant firebreak, achieving precise fire suppression based on fire type. The specific working steps are as follows: The dual-light camera simultaneously acquires visible light images and 14-bit thermal infrared images. Frame-level alignment is achieved by hardware timestamps, and homography transformation is performed using pre-calibrated rotation and translation matrices of the two camera intrinsic and extrinsic parameters to register the thermal infrared images to the visible light image coordinate system for pixel-level alignment. The edge computing unit runs a DeepLabV3+ lightweight semantic segmentation network with MobileNetV3 as the backbone to extract flame regions from visible light images. This network has been trained offline using a dataset of forest fire images from the perspective of drones. At the same time, based on the radiometric calibration parameters, atmospheric transmittance and emissivity of the thermal infrared image, the gray value of each pixel is converted into a Celsius temperature value to generate a temperature field. Based on the ground reference distance provided by the laser rangefinder in the camera, the pixels of the flame region are back-projected into three-dimensional space using projective geometry to obtain the height distribution of the maximum temperature point in each frame. Combining the distances from the UAV to the top of the tree canopy and to the ground obtained by scanning the gaps between the top of the tree canopy and the ground with a laser rangefinder, if the temperature peak areas are concentrated within a height range of ±1.5 meters from the top of the tree canopy and the thermal infrared image shows a cluster of high-temperature rising patterns, it is determined to be a crown fire; if the high-temperature areas are continuously distributed along the ground surface and the temperature at the height of the tree canopy on the same vertical line is at least 30°C lower than the temperature at the ground surface, it is determined to be a surface fire; specifically, the cluster of high-temperature rising patterns refers to: the aspect ratio of the bounding rectangle of the connected region of the high-temperature area (grayscale value more than 2 standard deviations higher than the global mean) in the thermal infrared image is between 0.7 and 1.5, and the vertical upward displacement of the centroid of this area in three consecutive frames of images is more than twice the horizontal displacement, while the elevation of the highest temperature point in the area migrates upward at a speed of not less than 0.5 m / s; only when all the above conditions are met is it determined to be a "cluster of high-temperature rising patterns".

[0022] When a crown fire is identified, the edge computing unit controls the electric actuator to extend the foldable telescopic spraying boom and activates the high-pressure water pump. The water-based flame retardant, mixed in a 3:97 volume ratio from the onboard water tank, is sprayed in a cone-shaped atomization pattern towards the base of the crown fire. Simultaneously, the flame area output by the segmentation network is read in real time, and the spraying flow rate is dynamically adjusted by regulating the PWM duty cycle of the water pump via a PID controller until the flame area drops below a safe threshold of 5% of the total image area. When a surface fire is identified, the drone automatically uses laser ranging to maintain a safe operating altitude of 15 meters relative to the ground surface. Subsequently, the edge computing unit acquires two frames of registered thermal infrared images from the current moment and 0.5 seconds prior, extracts the minimum bounding rectangle of the surface fire area, and calculates the displacement of the center of the rectangle's leading edge in the geographic coordinate system in both frames. Utilizing frame intervals Calculate the instantaneous spread rate of the fire line ; Simultaneously read the factory-specified effective asphyxiation radius of the airborne fire extinguishing bomb. ; Calculate the dynamic deployment interval based on the spread rate and the effectiveness of the fire extinguishing bomb. ;in This refers to the instantaneous spread rate of the fire line. The critical spread velocity required for this type of fire extinguishing bomb to form a continuous flame-retardant zone under wind conditions is obtained from a preset table; if If the system detects an error, it switches to the highest density delivery mode, setting the delivery interval to the system's minimum safe distance (e.g., 0.5m); otherwise, it calculates the delivery lead time. ,in The estimated landing time of the fire extinguishing projectile is provided in real time by the airflow sensing and ballistic compensation module; finally, based on the compensation parameters calculated in real time by the airflow sensing and ballistic compensation throwing module, the servo-driven throwing rail is deployed at dynamic intervals. and in front of the leading edge of the fire line The system continuously delivers ultrafine dry powder fire extinguishing bombs to form an adaptive flame-retardant isolation zone. After the delivery or spraying is completed and the system is left to stand for 3 seconds, the dual-light camera performs a second acquisition and evaluation. If the segmentation network still detects active fire points, the system returns to the first step and executes the loop. Otherwise, the system exits the module and records that the fire point has been cleared.

[0023] The infrared detection and precise extinguishing module for smoldering fires involves the drone performing a low-altitude, bow-shaped inspection after the open flame is extinguished. By improving threshold segmentation and Kalman filtering, it identifies the actual smoldering points and estimates the heat impact depth based on a temperature decay model. Then, it calculates the required number of extinguishing balls and precisely drops perfluorohexanone extinguishing balls at each point to completely eliminate smoldering hazards. The specific steps are as follows: After the layered fire extinguishing module confirms that all open flames have been extinguished, the drone automatically switches to the residual fire inspection mode. The gain of the high-sensitivity thermal imager is switched to the high-temperature residual fire detection level that is sensitive to 50-300℃. The drone performs low-altitude scanning at a speed of 2 meters per second and a flight altitude of 10 meters along a pre-programmed bow-shaped route. The route spacing is set to 5 meters according to the field of view of the thermal imager to ensure the overlap rate. The edge computing unit first performs a 5×5 Gaussian filter on each frame of thermal infrared image, and then runs an adaptive thresholding algorithm based on an improved Otsu's algorithm. This algorithm first calculates the global Otsu threshold of the image, then calculates the local threshold in 16 blocks, and finally uses the weighted result of the global threshold and the local mean as the segmentation threshold to extract segments with gray values ​​higher than the environmental background and connected component areas between 5. Up to 300 Candidate hotspot blocks; Kalman filtering is used to track target hotspots in five consecutive frames of images. The state vector contains image coordinates. The depth value calculated by laser ranging combined with the gimbal angle is used to achieve inter-frame matching through nearest neighbor data association. After meeting the condition of five consecutive successful associations and smooth motion, it is confirmed as a real smoldering point, eliminating false alarm sources such as sunspots and rock heat reflection, and locking its three-dimensional spatial coordinates. After confirming the smoldering point and locking its coordinates, the edge computing unit extracts the highest temperature value sequence from five consecutive frames of Kalman filter tracking records of that smoldering point. to And the corresponding frame acquisition time, using an exponential decay model. The initial temperature of the smoldering point is obtained by performing nonlinear least squares fitting. Ambient background temperature and temperature decay coefficient time The first frame is set to 0; where The smoldering point surface temperature varies over time. is the base of the natural logarithm; Using attenuation coefficient With the thermal diffusivity of the material Empirical Relationship Estimating the equivalent thermal influence depth of smoldering point , Typical values ​​of forest humus layer This empirical formula is derived from the analytical solution of the one-dimensional heat conduction equation for a semi-infinite object. Its physical basis is that the smoldering point can be approximated as the free cooling process following instantaneous heating of a semi-infinite medium surface. Within this theoretical framework, the temperature decay coefficient and the material's thermal diffusivity have a definite inverse relationship. This formula has been verified by general literature in the field of forest fire dynamics and exhibits good engineering approximation accuracy within the range of 10% to 40% humus moisture content. When the humus layer type (such as coniferous forest, broadleaf forest, or mixed forest) in the operating area deviates significantly from the typical value, the thermal diffusivity should be recalibrated within the range of 0.6 x 10⁻⁷ to 3.0 x 10⁻⁷ square meters per second based on the results of on-site sampling and thermophysical property testing. The system reserves an interface for the ground station to issue correction coefficients.

[0024] Then, the pixel area is converted into the actual smoldering point surface area by combining the area of ​​the connected domain of the thermal infrared image with the laser ranging value. ; Calculate the required inerting volume , For surface coverage safety margin Read the parameters of the airborne fire extinguishing ball's charge; effective mass of a single perfluorohexanone ball. Perfluorohexanone fire extinguishing design concentration Calculate the required number of fire extinguishing balls. ; It is a rounding function; like If it is greater than 1, then use Generate a uniform coverage area based on the geometric center. The coordinates of the aiming points are sorted from closest to furthest from the drone; if If the value is 1, then the aiming point is the original smoldering point coordinate; Based on the generated aiming point sequence, the edge computing unit controls the two-axis stabilization gimbal to point the aiming axis of the miniature fire extinguishing ball dispenser to the coordinates of each aiming point in sequence. The bottom of the dispenser is equipped with multiple magazines containing fire extinguishing balls with a diameter of 5cm. Each fire extinguishing ball is constrained by a nylon binding line and is internally encapsulated with perfluorohexanone liquid and nitrogen gas. After the edge computing unit is in aiming position, it sends a 1A current pulse to the heating wire melting mechanism of the corresponding magazine. The heating wire melts the binding line instantly, and the fire extinguishing ball falls freely along the aiming direction and hits the smoldering point and breaks. The perfluorohexanone vaporizes to form a suffocating environment to eliminate the smoldering. After deployment, the thermal imager continuously monitors the original coordinates of the smoldering point for 10 seconds and extracts its highest temperature value. If the temperature drops below 50°C from the initial value within 10 seconds and shows no upward trend, the point is marked as cleared. If the temperature is still above 50°C or rises after 10 seconds, an additional fire extinguishing ball is automatically deployed at the location and the monitoring cycle is restarted until all confirmed smoldering points in the fire area are cleared. After the edge computing unit records the information of the extinguished points, the flight control command instructs the drone to return to base.

[0025] The above formulas are all dimensionless calculations. Dimensionless calculations can be performed using various methods such as standardization, which will not be elaborated here. The formulas are derived from software simulations based on a large amount of collected data, and the preset parameters in the formulas can be set by those skilled in the art according to the actual situation.

[0026] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, ATA hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. The semiconductor medium can be a solid-state ATA hard disk.

[0027] 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.

[0028] 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.

[0029] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0030] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0031] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0032] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable ATA hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0033] The above description is merely a specific embodiment 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 intelligent forest fire suppression system based on a six-axis rotary-wing unmanned aerial vehicle (UAV), characterized in that, Includes the following modules: The airflow sensing and ballistic compensation throwing module is used to sense the multi-source wind fields of natural wind, rotor downwash and fire plume in real time through airborne sensors and generate the total equivalent wind field by vector superposition. Combined with the mass ballistic equation, the landing point deviation is iteratively calculated. Based on the landing point deviation, the servo throwing rail is driven to adjust the throwing angle so that the fire extinguishing bomb is thrown towards the fire point along the compensation direction. The fire type identification and layered fire suppression module is used to acquire images through dual-light cameras and use semantic segmentation networks to identify crown fires and ground fires. When it is identified as a crown fire, water-based flame retardant is sprayed in atomized form. When it is identified as a ground fire, the deployment interval and lead time are dynamically calculated based on the fire spread rate. Dry powder fire extinguishing bombs are continuously deployed to form a flame-retardant isolation zone. The smoldering fire infrared inspection and precise extinguishing module is used to switch to low-altitude inspection after the open flame is extinguished. It uses threshold segmentation and Kalman filtering to track and identify smoldering points, estimates the thermal impact depth based on the temperature decay model, calculates the required number of extinguishing balls, and throws extinguishing balls point by point to eliminate smoldering.

2. The intelligent forest fire fighting system based on a six-axis rotary-wing UAV according to claim 1, characterized in that, In the airflow sensing and trajectory compensation throwing module: An airborne ultrasonic anemometer acquires three-dimensional wind vectors, and an edge computing unit reads IMU data to calculate airspeed and ground speed. Based on the motor speed, a pre-calibrated downwash flow field reduced-order model is invoked to output the three components of rotor induced velocity at different elevations below the deployment height in real time; The natural wind vector and the downwash induced velocity vector are superimposed to construct the equivalent crosswind profile of the fire extinguishing bomb from the drop point to the ground; the equivalent crosswind profile is superimposed with the fire plume vector to generate the total equivalent wind field.

3. The intelligent forest fire fighting system based on a six-axis rotary-wing UAV according to claim 2, characterized in that, The downwash flow field reduced-order model is constructed based on intrinsic orthogonal decomposition and radial basis function interpolation. It takes motor speed and incoming wind speed as inputs and outputs the three components of rotor induced velocity at different elevations from zero to twenty meters below the deployment height in real time. The calibration process is as follows: multiple incoming wind speed levels and motor speed levels are set in the wind tunnel. A hot-wire anemometer array is arranged within the elevation range directly below the delivery point. The measured values ​​are used as the reference true values. The intrinsic orthogonal decomposition mode is selected to make the cumulative energy contribution rate reach more than 95%. The radial basis function interpolation shape parameters are optimized and selected through leave-one-out cross-validation. The root mean square error of the model output is controlled within 0.15 meters per second.

4. The intelligent forest fire fighting system based on a six-axis rotary-wing UAV according to claim 2, characterized in that, In the airflow sensing and trajectory compensation throwing module: The edge computing unit extracts the average radiation temperature and flame projection area of ​​the fire point area from the thermal infrared image of the dual-light camera, and reads the current ambient atmospheric temperature to estimate the average rising velocity of the fire plume center. The total equivalent wind field vector is generated by superimposing the fire plume velocity vector and the equivalent crosswind profile vector. The ballistic equation of the mass considering the wind field is used and solved iteratively by the fourth-order Runge-Kutta method to predict the horizontal deviation of the uncontrolled landing point relative to the target fire point. The pitch compensation angle and azimuth compensation angle required for the servo-deployed slide rail are calculated based on the horizontal deviation. The compensation angle is converted into PWM commands to drive the servo motor to adjust the slide rail to the compensation attitude before releasing the fire extinguishing projectile.

5. The intelligent forest fire fighting system based on a six-axis rotary-wing UAV according to claim 1, characterized in that, In the fire type identification and tiered fire suppression module: The dual-light camera simultaneously acquires visible light and thermal infrared images, achieves frame-level alignment through hardware timestamps, and performs homography transformation using pre-calibrated camera parameters to register the thermal infrared image to the visible light image coordinate system. The edge computing unit runs a semantic segmentation network to extract the flame region from the visible light image, and generates a temperature field based on the radiometric calibration parameters of the thermal infrared image, and obtains the height distribution of the maximum temperature point in each frame based on the projective geometry relationship.

6. The intelligent forest fire fighting system based on a six-axis rotary-wing UAV according to claim 5, characterized in that, Combining the distances from the drone to the top of the tree canopy and to the ground obtained by the laser rangefinder, if the temperature peak areas are concentrated within a height range of ±1.5 meters from the top of the tree canopy and the thermal infrared image shows a cluster of high-temperature floating shapes, then it is determined to be a tree canopy fire. The aforementioned cluster-like high-temperature floating morphology refers to the following: the aspect ratio of the circumscribed rectangle of the connected domain of the high-temperature region in the thermal infrared image is between 0.7 and 1.5, the vertical upward displacement of the centroid of this region in three consecutive frames of images is more than twice the horizontal displacement, and the elevation of the highest temperature point in this region migrates upward at a speed of not less than 0.5 meters per second.

7. A smart forest fire suppression system based on a six-axis rotary-wing UAV according to claim 5, characterized in that, When the fire is identified as a crown fire, the edge computing unit controls the electric push rod to extend the spray bar and starts the high-pressure water pump to spray the water-based flame retardant in the airborne water tank into the base of the crown fire in a cone atomization mode. At the same time, the water pump flow rate is adjusted by the PID controller until the flame area is reduced to less than five percent of the total image area. When the fire is identified as a surface fire, the drone maintains a safe operating height of 15 meters relative to the ground surface. The edge computing unit acquires two frames of thermal infrared images, one from the current moment and the other from 0.5 seconds ago, extracts the minimum bounding rectangle of the surface fire area, and calculates the instantaneous spread rate of the fire line.

8. The intelligent forest fire fighting system based on a six-axis rotary-wing UAV according to claim 7, characterized in that, The edge computing unit reads the effective suffocation radius of the fire extinguishing bomb and calculates the dynamic deployment interval based on the fire spread rate and the critical spread speed of the fire extinguishing bomb. If the fire spread rate reaches or exceeds the critical spread rate, switch to the highest density deployment mode and set the deployment interval to the minimum safe distance allowed by the system. Otherwise, the lead time for deployment is calculated, and the servo deployment rail continuously deploys dry powder fire extinguishing bombs at dynamic intervals and in front of the leading edge of the fire line. After the application or spraying is completed and the system has been allowed to stand, the dual-light camera will perform a second data collection and evaluation. If active fire points are still detected, the process will be repeated; otherwise, the process will exit and the fire point will be marked as cleared.

9. A smart forest fire fighting system based on a six-axis rotary-wing UAV according to claim 1, characterized in that, In the residual fire infrared inspection and precise extinguishing module: The UAV scans at low altitude along a bow-shaped flight path. The edge computing unit performs Gaussian filtering on the thermal infrared image and then runs an adaptive threshold segmentation algorithm to extract candidate hot spot blocks. Kalman filtering is used to track and match the candidate hot spots in multiple consecutive frames. After successful association and smooth motion in multiple consecutive frames, the hot spots are confirmed as real smoldering points and their three-dimensional spatial coordinates are locked. An exponential decay model was used to fit and obtain the initial temperature of the smoldering point, the ambient background temperature, and the temperature decay coefficient. Based on this, the thermal influence depth was estimated. The required inerting volume was calculated by combining the actual surface area of ​​the smoldering point and the safety margin. Then, the required number of fire extinguishing balls was calculated based on the effective mass of the fire extinguishing ball and the fire extinguishing design concentration. The control pan-tilt unit pointed the dispenser sequentially to the coordinates of each aiming point and sent a current pulse to the heating wire fuse mechanism to make the fire extinguishing ball fall freely along the aiming direction and hit the smoldering point. The thermal imager continuously monitored the temperature of the point. If the temperature dropped below the threshold and did not rise, the mark was cleared. Otherwise, additional fire extinguishing balls were added until all were cleared before returning to base.