Fire-fighting unmanned aerial vehicle posture control and throwing method and system for complex environment

Abstract

The application provides a fire-fighting unmanned aerial vehicle posture control and throwing method and system for complex environments, and relates to the technical field of fire-fighting unmanned aerial vehicle control. The method comprises the following steps: collecting the image, temperature, flight attitude and height of a fire site by a color CCD task load, a thermal imaging system, an attitude sensor and a laser ranging device of a fire-fighting unmanned aerial vehicle, and carrying out noise reduction; processing the image based on an improved target detection algorithm and outputting the fire source category and coordinates; representing the flight attitude by using quaternions, combining an improved active disturbance rejection controller to build a posture control model, calculating the deviation and generating motor control quantity; calculating the optimal throwing parameters and generating instructions according to the fire source information, flight and environmental data and the dynamic characteristics of fire extinguishing bombs; executing the throwing and monitoring the trajectory of the fire extinguishing bomb, and adjusting the posture control model if the trajectory deviation exceeds the preset range, so that the stable flight of the unmanned aerial vehicle and the accurate throwing of the fire extinguishing bomb in the complex fire site environment can be realized, and the safety and effectiveness of the fire-fighting operation are ensured.

Description

Technical Field

[0001] This invention belongs to the field of firefighting drone control technology, and relates to a method and system for attitude control and throwing of firefighting drones in complex environments. Background Technology

[0002] Against the backdrop of global warming and frequent extreme weather events, fire accidents are becoming increasingly complex and severe. Whether it's urban high-rise fires, forest fires, or chemical fires, they all pose a significant threat to people's lives and property. In this context, firefighting drones, with their advantages of flexibility, efficiency, and rapid arrival at fire scenes, have become an important tool in the modern firefighting field, and their application in complex fire environments is increasingly in demand.

[0003] In recent years, with the continuous advancement of technology, significant progress has been made in the research of firefighting drones. In terms of attitude control, most drones use Euler angles to represent attitude and employ traditional proportional-integral-derivative (PID) controllers for control. For fire source detection, traditional target detection algorithms are primarily relied upon. The deployment of fire extinguishing bombs is based on simple mechanical models.

[0004] However, existing technologies have revealed numerous problems in practical applications. Attitude control, using Euler angles, is prone to gimbal lock issues, and traditional PID controllers, with their fixed parameters, have weak anti-interference capabilities in complex environments such as strong winds and hot air currents, resulting in poor flight stability. Traditional algorithms for fire source detection are ill-suited to complex fire environments, exhibiting low accuracy in identifying small fire sources and poor fire source localization. The throwing of fire extinguishing projectiles lacks a dynamic environmental compensation mechanism, failing to fully consider the impact of wind speed gradients and thermal flow fields on the trajectory, resulting in insufficient precision throwing capability, significant landing point deviations, and severely impacting fire extinguishing efficiency and safety. These shortcomings make it difficult for existing firefighting drone technology to meet the demands of stable flight and precise fire extinguishing in complex fire environments. Therefore, developing attitude control and throwing methods and systems for firefighting drones in complex environments is urgently needed. Summary of the Invention

[0005] To address the problems existing in the prior art, this invention provides a method and system for attitude control and throwing of firefighting drones in complex environments. This solves the problems of existing firefighting drone technology, which suffers from weak anti-interference and poor flight stability in complex environments due to the use of Euler angles and fixed-parameter PID controllers in attitude control; low accuracy and poor positioning of small targets due to reliance on traditional algorithms for fire source detection; and insufficient precision throwing capability and large landing point deviation due to the lack of dynamic compensation mechanism in fire extinguishing bomb throwing. These issues make it difficult to meet the requirements for stable flight and precise fire extinguishing in complex fire scenes.

[0006] This invention is achieved through the following technical solution: Attitude control and drop methods for firefighting drones in complex environments include: Data Acquisition and Noise Reduction: Using the color charge-coupled device payload, thermal imaging system, attitude sensor and laser rangefinder carried by the fire-fighting drone, fire scene images, temperature distribution, flight attitude parameters and altitude data are collected respectively, and an adaptive filtering algorithm is used for noise reduction. Fire source detection: Based on an improved target detection algorithm, fire scene images are processed. Features of fire sources of different sizes are extracted through multi-scale feature fusion. Dynamic loss function is combined to optimize and improve the recognition accuracy of small targets. The fire source category and spatial coordinates are output. Attitude control: Quaternions are used to represent flight attitude parameters. A model is built by combining an improved active disturbance rejection controller. Attitude deviation is calculated in real time, motor control quantities are dynamically generated and converted into pulse width modulation duty cycle signals for output. Drop decision: Based on fire source information, flight altitude, temperature distribution, attitude, wind speed and fire extinguishing bomb dynamics, calculate the optimal drop parameters and generate execution instructions; Execution and Adjustment: The system executes the throwing operation according to instructions, monitors the trajectory using visual sensors, and adjusts the attitude control model when deviations exceed limits.

[0007] Preferably, the improved target detection algorithm is the ASF-WIoU-YOLOv8 algorithm, and the processing steps of the ASF-WIoU-YOLOv8 algorithm are as follows: The feature extraction module performs convolution and pooling operations on the input fire scene image data to extract the fire scene feature information corresponding to the fire scene image data, forming fire scene feature maps at different scales. An attention scale sequence fusion mechanism is introduced into the feature enhancement module to perform weighted fusion of the fire scene feature maps at different scales to generate a fused fire scene feature map. The fused fire scene feature map is processed by a channel attention module and a spatial attention module to highlight the feature information of the fire source area from the channel dimension and the spatial dimension, respectively, and generate an enhanced fire scene feature map. The target detection module performs target classification and bounding box regression on the enhanced fire scene feature map, and outputs the fire source category and spatial coordinates. The ASF-WIoU-YOLOv8 algorithm uses the Wise-IoU loss function, which adaptively adjusts the weight coefficients of each module in the ASF-WIoU-YOLOv8 algorithm.

[0008] Preferably, the step of using quaternions to represent flight attitude parameters, combining an improved active disturbance rejection controller to construct a model, calculating attitude deviations in real time, dynamically generating current motor control quantities, and converting them into pulse width modulation duty cycle signals for output specifically includes: The flight attitude parameters include angular velocity data, linear acceleration data, and magnetic field strength data. Quaternions are used to parameterize the flight attitude parameters and the target flight attitude parameters, resulting in quaternion attitude representations and target quaternion attitude representations, respectively. Based on the quaternion attitude representation, an improved active disturbance rejection controller is constructed. An extended state observer is used to estimate the internal dynamic disturbances and external environmental disturbances of the system in real time. A nonlinear error feedback mechanism is introduced to form a robust control model for the flight attitude parameters. The quaternion attitude representation is compared with the target quaternion attitude representation, the flight attitude deviation is calculated in real time and input to the improved active disturbance rejection controller to generate the current motor control quantity; The current motor control quantity is limited and converted into the rotation speed command of the four rotors of the fire-fighting drone. The rotation speed command is converted into the pulse width modulation duty cycle signal through pulse width modulation technology and output to the motor drive module of the fire-fighting drone.

[0009] Preferably, the angular velocity data is represented as ,in, These represent the rotational angular velocities of the firefighting drone around the x, y, and z axes, respectively; the linear acceleration data are expressed as... ,in, These represent the acceleration components of the firefighting drone along the x, y, and z axes, respectively; the magnetic field strength data are expressed as... ,in, These represent the magnetic field strength components of the firefighting drone on the x, y, and z axes, respectively.

[0010] Preferably, when comparing the quaternion attitude representation with the target quaternion attitude representation, the quaternion attitude deviation corresponding to the flight attitude deviation is calculated in real time through quaternion multiplication and normalization operations.

[0011] Preferably, the step of calculating the optimal throwing parameters and generating execution instructions based on fire source information, flight altitude, temperature distribution, attitude, wind speed, and the dynamic characteristics of the fire extinguishing projectile is as follows: A three-dimensional fire scene model is constructed based on the type of fire source and spatial coordinates; The flight altitude data is mapped onto the three-dimensional fire field model to form a spatial positioning reference; the fire field temperature distribution data is fused into the three-dimensional fire field model to generate a thermal radiation intensity field; based on the three-dimensional fire field model and combined with the current flight attitude parameters, a relative coordinate system of the UAV is established to form a comprehensive fire field situation map. Based on the comprehensive fire situation map and the current environmental wind speed, an atmospheric flow field model is constructed, and the influence of wind speed gradients at different altitudes and thermal updrafts on the flight trajectory of fire extinguishing bombs is analyzed, establishing a thermal flow-airflow coupling model. Based on the aforementioned heat flow-air flow coupling model, the impact of environmental disturbances on the flight trajectory of the fire extinguishing projectile is quantified through fluid dynamics simulation, and a dynamic compensation parameter set is generated. Based on the dynamic compensation parameter set and combined with the dynamic characteristics of the fire extinguishing projectile, a ballistic trajectory prediction optimization model is constructed. The ballistic trajectory prediction optimization model takes the accuracy of the fire extinguishing projectile's landing point, the coverage range of the fire extinguishing agent, and the penetration capability of the fire extinguishing projectile as objectives, and takes the maneuverability of the fire-fighting drone and the dynamic changes in the fire scene as constraints. The model uses an intelligent optimization algorithm to iteratively solve the parameter combination of the optimal throwing height, the optimal throwing angle, and the optimal release timing. The calculated combination of the optimal throwing height, optimal throwing angle, and optimal release timing is converted into the current throwing execution command, which includes flight altitude adjustment parameters, flight attitude calibration parameters, and fire extinguishing grenade release timing parameters.

[0012] A firefighting drone attitude control and throwing system for complex environments includes: The data acquisition and noise reduction module utilizes the color charge-coupled device (CCD) payload, thermal imaging system, attitude sensor, and laser rangefinder mounted on the fire-fighting drone to acquire fire scene images, temperature distribution, flight attitude parameters, and altitude data, and uses an adaptive filtering algorithm for noise reduction. The fire source detection module processes fire scene images based on an improved target detection algorithm. It extracts features of fire sources of different sizes through multi-scale feature fusion, optimizes and improves the recognition accuracy of small targets by combining dynamic loss function, and outputs the fire source category and spatial coordinates. The attitude control module uses quaternions to represent flight attitude parameters, combines an improved active disturbance rejection controller to build a model, calculates attitude deviations in real time, dynamically generates motor control quantities, and converts them into pulse width modulation duty cycle signals for output. The throwing decision module calculates the optimal throwing parameters and generates execution instructions based on fire source information, flight altitude, temperature distribution, attitude, wind speed and fire extinguishing bomb dynamic characteristics. The execution and adjustment module executes the throwing according to instructions, monitors the trajectory with a visual sensor, and adjusts the attitude control model when the deviation exceeds the limit.

[0013] A computer device / apparatus / system includes a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above-described method.

[0014] A computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described method.

[0015] A computer program product includes a computer program / instructions that, when executed by a processor, implement the steps of the above-described method.

[0016] Compared with the prior art, the present invention has the following beneficial technical effects: This invention discloses a method for attitude control and deployment of firefighting drones in complex environments. The method utilizes a color CCD payload, thermal imaging system, attitude sensor, and laser rangefinder mounted on the firefighting drone to collect fire scene image data, fire temperature distribution data, flight attitude parameters, and flight altitude data of the target fire area. An adaptive filtering algorithm is used to denoise the fire scene image data, fire temperature distribution data, flight attitude parameters, and flight altitude data. An improved target detection algorithm is used to process the fire scene image data, extracting feature information of fire sources of different sizes through a multi-scale feature fusion mechanism. A dynamic loss function optimization strategy is combined to improve the recognition accuracy of small target fire sources, outputting the fire source category and spatial coordinates. Quaternions are used to represent flight attitude parameters, and an improved active disturbance rejection controller is used to construct an attitude control model. The method calculates the flight attitude parameters and target flight altitude data in real time. The flight attitude deviation of the flight attitude parameters is dynamically used to generate the current motor control quantity and convert it into a pulse width modulation duty cycle signal, which is then output to the motor drive module of the fire-fighting drone. Based on the fire source type, spatial coordinates, and flight altitude data, combined with fire temperature distribution data, current flight attitude parameters, current ambient wind speed, and fire extinguishing projectile dynamic characteristics, the optimal throwing height, optimal throwing angle, and optimal release timing are calculated, generating the current throwing execution command. Based on the current throwing execution command, the fire-fighting drone's fire extinguishing projectile delivery system executes the current throwing action. At the same time, the fire-fighting drone's visual sensors monitor the flight trajectory of the fire extinguishing projectile. If the flight trajectory deviation of the fire extinguishing projectile exceeds the preset trajectory deviation range, the attitude control model is adjusted. This achieves stable flight of the fire-fighting drone, accurate fire source identification, and efficient delivery of fire extinguishing projectiles in complex fire environments, ensuring the safety and effectiveness of fire-fighting operations.

[0017] Furthermore, this invention employs an improved ASF-WIoU-YOLOv8 algorithm, which uses an attention-scale sequence fusion mechanism to weightedly fuse fire scene feature maps at different scales, enhancing the feature representation capability of multi-scale fire sources. Combined with the Wise-IoU loss function to dynamically adjust weight coefficients, it optimizes target classification and bounding box regression accuracy in complex backgrounds, solving the problem of poor detection performance for small target fire sources in traditional algorithms. This ensures accurate output of fire source category and spatial coordinates, providing a reliable basis for subsequent throwing decisions. This invention uses quaternions to represent flight attitude parameters, effectively avoiding the gimbal lock problem in Euler angle attitude calculation. Combined with an improved active disturbance rejection controller, it uses an extended state observer to estimate internal and external disturbances in real time, introducing a nonlinear error feedback mechanism to dynamically generate motor control quantities. This significantly improves the firefighting drone's resistance to environmental disturbances such as strong winds and hot air currents, ensuring that the firefighting drone can maintain stable flight attitude and adaptive attitude control under complex conditions such as high and low temperatures and electromagnetic interference, laying the foundation for accurate throwing. This invention constructs a heat flow-airflow coupling model based on fire source location, flight altitude, temperature distribution, ambient wind speed, and the dynamic characteristics of the fire extinguishing projectile. It quantifies the impact of environmental disturbances on the trajectory through fluid dynamics simulation and generates dynamic compensation parameters. Combined with an intelligent optimization algorithm, it iteratively solves for the optimal throwing height, angle, and timing. Furthermore, it uses a visual sensor to monitor the flight trajectory deviation of the fire extinguishing projectile in real time, triggering the attitude control model to make secondary adjustments, forming a closed-loop correction mechanism. This effectively reduces the landing point error of the fire extinguishing projectile, ensures the accuracy of the projectile's throwing, and significantly improves fire extinguishing efficiency and safety. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 The flowchart shows the attitude control and throwing method of the firefighting drone for complex environments according to the present invention. Figure 2 This is a system block diagram of the attitude control and throwing system for firefighting drones in complex environments according to the present invention. Figure 3 This is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present invention.

[0020] Among them, 30 is electronic equipment; 31 is processor; 32 is memory; and 33 is bus. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0022] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0023] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0024] In the description of the embodiments of the present invention, it should be noted that if terms such as "upper," "lower," "horizontal," or "inner" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship commonly used when the product of the invention is in use, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention. Furthermore, terms such as "first" and "second" are only used to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0025] Furthermore, the use of the term "horizontal" does not imply that the component must be absolutely horizontal, but rather that it can be slightly tilted. For example, "horizontal" simply means that its direction is more horizontal than "vertical," and does not mean that the structure must be completely horizontal, but can be slightly tilted.

[0026] In the description of the embodiments of the present invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," and "link" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in the present invention according to the specific circumstances.

[0027] The present invention will now be described in further detail with reference to the accompanying drawings: This invention provides an attitude control and throwing method for firefighting drones in complex environments, addressing the problems of firefighting drones' inability to achieve stable flight and adaptive attitude control in complex environments, as well as insufficient accuracy in fire extinguishing bomb throwing, resulting in low firefighting efficiency and safety. The method includes: Data Acquisition and Noise Reduction: Using the color charge-coupled device (CCD) payload, thermal imaging system, attitude sensor and laser rangefinder carried by the fire-fighting drone, fire scene images, temperature distribution, flight attitude parameters and altitude data are acquired respectively, and an adaptive filtering algorithm is used for noise reduction. Fire source detection: Based on an improved target detection algorithm, fire scene images are processed. Features of fire sources of different sizes are extracted through multi-scale feature fusion. Dynamic loss function is combined to optimize and improve the recognition accuracy of small targets. The fire source category and spatial coordinates are output. Attitude control: Flight attitude parameters are represented by quaternions, and a model is built by combining an improved active disturbance rejection controller. Attitude deviations are calculated in real time, motor control quantities are dynamically generated and converted into pulse width modulation duty cycle signals for output. Drop decision: Based on fire source information, flight altitude, temperature distribution, attitude, wind speed and fire extinguishing bomb dynamics, calculate the optimal drop parameters and generate execution instructions; Execution and Adjustment: The system executes the throwing operation according to instructions, monitors the trajectory using visual sensors, and adjusts the attitude control model when deviations exceed limits.

[0028] More specifically, such as Figure 1 The diagram shown is a flowchart of a firefighting drone attitude control and throwing method for complex environments provided by an embodiment of the present invention. Figure 1 The execution entity of the method shown can be a software and / or hardware device. The execution entity of this application can include, but is not limited to, at least one of the following: user equipment, network equipment, etc. User equipment can include, but is not limited to, computers, smartphones, personal digital assistants (PDAs), and the aforementioned electronic devices. Network equipment can include, but is not limited to, a single network server, a server group consisting of multiple network servers, or a cloud based on cloud computing consisting of a large number of computers or network servers. Cloud computing is a type of distributed computing, consisting of a super virtual computer composed of a group of loosely coupled computers. This embodiment does not limit this. Steps S1 to S5 are detailed as follows: S1, the fire scene image data, fire scene temperature distribution data, flight attitude parameters and flight altitude data are collected by the color CCD mission payload, thermal imaging system, attitude sensor and laser rangefinder carried by the fire-fighting drone, and the fire scene image data, the fire scene temperature distribution data, the flight attitude parameters and the flight altitude data are processed by an adaptive filtering algorithm to reduce noise. The color CCD payload refers to the color image acquisition device mounted on the firefighting drone. It uses charge-coupled device (CCD) technology to acquire visible light image data of the fire area to achieve fire source identification and location. The thermal imaging system is a device that generates temperature distribution images by detecting the infrared radiation intensity of objects. It is used to collect temperature gradient data of the fire area, identify high-temperature fire source areas, and output temperature distribution information. The attitude sensor is an integrated sensor composed of a gyroscope, accelerometer, and magnetometer. The gyroscope, accelerometer, and magnetometer are used to collect angular velocity data, linear acceleration data, and environmental magnetic field strength data of the drone around three-dimensional axes, respectively. These data are then combined to generate flight attitude parameters reflecting the drone's spatial attitude. The laser rangefinder is a distance measurement device based on the laser ranging principle, used to collect the vertical distance between the firefighting drone and the ground or fire source in real time, providing a measurement basis for flight altitude data.

[0029] Fire scene image data refers to visible light image information of the fire scene area acquired by a color CCD payload, including visual features such as fire source morphology, distribution range, and surrounding environment. It is the fundamental data for fire source identification and location. Fire scene temperature distribution data refers to the temperature gradient and spatial distribution information of the fire scene area detected by a thermal imaging system. It reflects the heat intensity, spread trend, and high-temperature zone boundaries of the fire source, providing a basis for fire assessment. Flight attitude parameters refer to the three-dimensional spatial attitude information of the UAV acquired by attitude sensors, used to characterize the real-time attitude state of the firefighting UAV. Flight altitude data refers to the vertical distance information of the firefighting UAV relative to the fire scene area or the ground, measured by a laser rangefinder. It is key data for calculating the parameters for fire extinguishing bomb deployment. Adaptive filtering algorithms are signal processing algorithms that can dynamically adjust filtering parameters. They are used to suppress noise in fire scene images, temperature distribution, flight attitude, and altitude data, reducing data distortion caused by environmental interference and improving data reliability.

[0030] The firefighting drone, equipped with a color CCD payload, can capture visible light images of the fire area, forming fire image data to visually present the shape and distribution of the fire source. Through a thermal imaging system, it can sense infrared radiation and generate fire temperature distribution data, accurately reflecting the temperature gradient and high-temperature zone boundaries, providing thermal dimension information for fire assessment. An attitude sensor integrating a gyroscope, accelerometer, and magnetometer collects the drone's angular velocity, linear acceleration, and environmental magnetic field strength, summarizing these data to generate flight attitude parameters, characterizing the three-dimensional spatial attitude of the firefighting drone. A laser rangefinder measures the vertical distance of the firefighting drone relative to the ground or fire source, generating flight attitude parameters that provide a benchmark for subsequent spatial positioning and deployment calculations.

[0031] Due to electromagnetic interference, vibration, and other disturbances in complex fire scenes, the raw data is easily contaminated by noise. Therefore, an adaptive filtering algorithm is needed to process the fire scene image data, fire scene temperature distribution data, flight attitude parameters, and flight altitude data. This algorithm can dynamically adjust the filtering parameters, effectively suppress noise interference, reduce data distortion, and provide a reliable data foundation for subsequent analysis and calculation.

[0032] S2, the fire scene image data is processed based on the improved target detection algorithm, the feature information of fire sources of different sizes is extracted through the multi-scale feature fusion mechanism, and the recognition accuracy of small target fire sources is improved by combining the dynamic loss function optimization strategy, and the fire source category and spatial coordinates are output. It is understandable that the improved target detection algorithm refers to the ASF-WIoU-YOLOv8 algorithm, which introduces the Attention Scale Sequence Fusion (ASF) mechanism and the Wise-IoU loss function on the basis of YOLOv8 to optimize the detection performance of multi-scale fire sources in complex fire scenes. The multi-scale feature fusion mechanism refers to the technique used to weightedly fuse feature maps of different levels, which can enhance the feature representation ability of fire sources of different sizes, especially small target fire sources, and improve the accuracy of cross-scale target detection. The dynamic loss function optimization strategy can adopt the Wise-IoU loss function, which optimizes the loss calculation of fire source detection by adaptively adjusting the weight coefficients of each module in the ASF-WIoU-YOLOv8 algorithm, thereby improving the robustness of fire source detection. Small target fire sources refer to fire sources that occupy few pixels and are small in size in fire scene images, such as initial fire sources and concealed fire sources. Fire source category refers to the type of fire source identified from fire scene image data based on an improved target detection algorithm, such as open flame, smoldering, electrical fire, etc. Fire source spatial coordinates refer to the location information of the fire source in a three-dimensional coordinate system, including latitude and longitude, relative altitude, etc., which is the core reference for accurate throwing.

[0033] Based on an improved target detection algorithm, this algorithm enables deep analysis of preprocessed fire scene image data. It is optimized to address common characteristics of fire environments, such as large scenes, multiple scales, small targets, and complex backgrounds. In the feature processing stage, a multi-scale feature fusion mechanism is used to weightedly fuse feature maps at different levels, enhancing the feature representation capabilities of fire sources of different sizes. This ensures that feature information can be fully extracted from both large-area open flames and concealed initial fires. Simultaneously, a dynamic loss function optimization strategy is introduced. By adaptively adjusting the weight coefficients, the training effect of samples of different quality is balanced, effectively improving the recognition accuracy of small target fire sources and reducing false positives and false negatives. Finally, the algorithm outputs the fire source category and spatial coordinates, providing core target parameters for subsequent precise throwing.

[0034] S3, using quaternions to represent the flight attitude parameters, and combining an improved active disturbance rejection controller to construct an attitude control model, dynamically generating the current motor control quantity by calculating the flight attitude deviation between the flight attitude parameters and the target flight attitude parameters in real time, and converting it into a pulse width modulation duty cycle signal to output to the motor drive module of the fire-fighting drone. It should be noted that quaternions are a mathematical representation consisting of one scalar and three vectors, used to describe the three-dimensional attitude of firefighting drones. This avoids the gimbal lock problem in Euler angle attitude calculation, ensuring the continuity and stability of the attitude representation. An improved active disturbance rejection controller (ADRC) can be based on an ADRC architecture optimized controller. By extending the state observer, it estimates internal and external disturbances in real time and generates control quantities using nonlinear error feedback, achieving robust attitude control. Target flight attitude parameters refer to the preset desired attitude parameters of the firefighting drone when performing firefighting tasks, such as the horizontal attitude when hovering and the tilt angle when tracking a fire source. These serve as the benchmark for the attitude control of the firefighting drone. Flight attitude deviation refers to the difference between the current flight attitude parameters and the target attitude parameters of the firefighting drone, and is the core basis for attitude adjustment.

[0035] The current motor control quantity refers to the physical quantity used to adjust motor operation based on flight attitude deviation, such as torque and speed commands, used to drive UAV attitude correction. The pulse width modulation (PWM) duty cycle signal is an electrical signal that controls motor speed by adjusting the pulse width duty cycle. The motor drive module is a device that receives the PWM duty cycle signal and drives the motor, used to adjust the motor speed and direction of the firefighting UAV according to the PWM duty cycle signal, achieving real-time attitude adjustment of the firefighting UAV.

[0036] Compared to the traditional Euler angle representation, using quaternions to mathematically represent flight attitude parameters effectively avoids gimbal lock problems, ensuring the continuity and stability of flight attitude description. Then, an attitude control model can be constructed by combining an improved active disturbance rejection controller. This controller estimates the internal dynamics and external environmental disturbances of the system in real time through an extended state observer and introduces a nonlinear error feedback mechanism to enhance resistance to complex disturbances. By calculating the flight attitude deviation between the current flight attitude parameters and the preset target flight attitude parameters in real time—that is, the difference between the actual attitude and the desired attitude—the current motor control quantity is dynamically generated. This control quantity is the force or torque command that adjusts the firefighting drone from the current attitude to the target attitude. Subsequently, the continuous motor control quantity can be amplitude-limited and converted into speed commands adapted to the four rotors of the firefighting drone. Pulse width modulation (PWM) technology is then used to convert the speed commands into PWM duty cycle signals. Finally, this signal can be output to the motor drive module to directly control the speed and direction of each rotor, achieving real-time and precise adjustment of the firefighting drone's attitude and ensuring its stable flight in complex environments.

[0037] S4. Based on the fire source type and spatial coordinates and the flight altitude data, combined with the fire temperature distribution data, current flight attitude parameters, current ambient wind speed and fire extinguishing bomb dynamics, calculate the optimal throwing height, optimal throwing angle and optimal release timing, and generate the current throwing execution command. Among them, the current flight attitude parameters refer to the actual flight attitude parameters of the firefighting drone at the current moment. The current ambient wind speed refers to the real-time airflow speed and direction information in the fire area, including the wind speed gradient at different altitudes, which is a key environmental factor affecting the flight trajectory of the fire extinguishing projectile. The dynamic characteristics of the fire extinguishing projectile refer to the motion law of the projectile under the influence of gravity, air resistance, wind speed, etc. during flight, including flight speed decay characteristics and airburst response characteristics, which are the core basis for calculating the throwing parameters.

[0038] The optimal launch height refers to the height at which the fire extinguishing projectile accurately covers the core area of ​​the fire source, calculated by comprehensively considering the spatial coordinates of the fire source, flight altitude data, ambient wind speed, and the dynamic characteristics of the fire extinguishing projectile. To ensure the projectile's trajectory adapts to environmental interference and minimizes impact deviation, an optimal launch angle is set, which is the optimal angle between the projectile and the horizontal plane at the time of release. The optimal release timing refers to the time point at which the projectile is released, determined by combining the UAV's current flight attitude parameters, the projectile's dynamic characteristics, and ambient wind speed, to match the real-time location of the fire source with the predicted trajectory. The current launch execution command is a set of instructions containing parameters such as the optimal launch height, launch angle, and release timing, used to control the fire extinguishing projectile delivery system to complete aiming and release actions.

[0039] Based on the acquired fire source type, spatial coordinates, and flight altitude data, a multi-dimensional optimization model can be constructed by comprehensively considering fire temperature distribution data, current flight attitude parameters, current ambient wind speed, and the dynamic characteristics of the fire extinguishing projectile. The impact of environmental disturbances on the projectile's trajectory is quantified through fluid dynamics simulation, generating dynamic compensation parameters to solve for the optimal throwing height, optimal launch angle, and optimal release timing. Finally, these parameters can be integrated into the current throwing execution command, which includes flight altitude adjustment parameters, flight attitude calibration parameters, and fire extinguishing projectile release timing parameters, providing precise guidance for the throwing action.

[0040] S5. Based on the current throwing execution command, the current throwing action is executed through the fire extinguishing bomb delivery system of the fire-fighting drone. At the same time, the visual sensor of the fire-fighting drone is used to monitor the flight trajectory of the fire extinguishing bomb. If the flight trajectory deviation of the fire extinguishing bomb exceeds the preset trajectory deviation range, the attitude control model is adjusted.

[0041] It is understandable that the current throwing action refers to the specific operation performed by the fire extinguishing grenade delivery system according to the current throwing execution command, including continuous actions such as angle adjustment, target locking, and releasing the fire extinguishing grenade. The fire extinguishing grenade delivery system refers to a device composed of a suspended aiming and releasing device, a laser rangefinder, etc., used to receive throwing commands and execute the aiming and releasing actions of the fire extinguishing grenade. The visual sensor is a device used to acquire real-time images of the fire extinguishing grenade's flight and monitor its trajectory deviation. The fire extinguishing grenade's flight trajectory refers to the spatial movement path of the fire extinguishing grenade from release to arrival at the fire extinguishing area, monitored in real-time by the visual sensor to determine whether the trajectory deviation is within a preset range, providing a basis for secondary adjustments to the attitude control model. The preset trajectory deviation range refers to a pre-set threshold for allowable trajectory deviation of the fire extinguishing grenade. If the actual deviation exceeds this range, the attitude control model is triggered to perform secondary adjustments to ensure the accuracy of the fire extinguishing grenade throwing.

[0042] Based on the current delivery command, the fire-fighting drone's fire extinguishing bomb delivery system executes the current delivery action, including adjusting the delivery angle, locking onto the target, and releasing the fire extinguishing bomb in a continuous sequence. Simultaneously, the drone's onboard visual sensors monitor the fire extinguishing bomb's flight trajectory in real time. Through image recognition and trajectory fitting, the deviation between the actual and predicted trajectories is obtained. If this deviation exceeds a preset trajectory deviation range, the attitude control model is triggered for secondary adjustments. By recalculating the attitude deviation, new motor control quantities are generated to correct the drone's attitude, indirectly adjusting the delivery trajectory of subsequent fire extinguishing bombs or dynamically correcting the delivery parameters of unreleased fire extinguishing bombs. This ensures that the fire extinguishing bombs ultimately and accurately cover the fire source area, improving fire-fighting efficiency and safety.

[0043] The improved target detection algorithm is the ASF-WIoU-YOLOv8 algorithm, and the processing steps of the ASF-WIoU-YOLOv8 algorithm are as follows: The feature extraction module performs convolution and pooling operations on the input fire scene image data to extract the fire scene feature information corresponding to the fire scene image data, forming fire scene feature maps at different scales. An attention scale sequence fusion mechanism is introduced into the feature enhancement module to perform weighted fusion of the fire scene feature maps at different scales to generate a fused fire scene feature map. The fused fire scene feature map is processed by a channel attention module and a spatial attention module to highlight the feature information of the fire source area from the channel dimension and the spatial dimension, respectively, and generate an enhanced fire scene feature map. The target detection module performs target classification and bounding box regression on the enhanced fire scene feature map, and outputs the fire source category and spatial coordinates.

[0044] First, the input fire scene image data is processed by a feature extraction module. This module performs layer-by-layer feature extraction on the image through continuous convolution operations, using convolution kernels to capture basic features such as edges and textures in the image, and combining this with pooling operations to compress the feature map size, retaining key information while reducing computational load. After multiple rounds of convolution and pooling, fire scene feature information corresponding to different levels is finally extracted from the original fire scene image, forming a series of fire scene feature maps of different scales. Among them, small-scale feature maps retain more detailed information and are suitable for identifying small target fire sources; large-scale feature maps contain richer global semantic information, which is beneficial for capturing the overall features of large-area fire sources.

[0045] Subsequently, an attention-scale sequence fusion mechanism was introduced into the feature enhancement module. This mechanism employs a weighted fusion approach, dynamically assigning weights to feature maps at different scales: smaller-scale feature maps containing details of small target fire sources are given higher weights, strengthening their contribution to the fusion process; simultaneously, the global information of larger-scale feature maps is preserved, ensuring the complete retention of multi-scale fire source features. This fusion strategy effectively addresses the problem of insufficient representation of multi-scale targets by single-scale feature maps, generating fused fire scene feature maps that more comprehensively cover fire sources of different sizes and distributions.

[0046] Next, the fused fire scene feature map is enhanced using a channel attention module and a spatial attention module. The channel attention module analyzes the importance of each feature channel, assigning higher weights to channels representing key fire source features such as flame color and outline, while suppressing interference from irrelevant background channels, thus highlighting fire source features from a channel perspective. The spatial attention module focuses on the spatial distribution of the image, calculating the correlation between pixels to strengthen the spatial response of the fire source area and weaken the feature signals of the background area, further focusing on the fire source from a spatial perspective. After processing by these two modules, an enhanced fire scene feature map is generated, significantly highlighting the feature information of the fire source area in both channel and spatial dimensions.

[0047] Finally, the enhanced fire scene feature map is processed by the target detection module. This module performs target classification and bounding box regression based on the enhanced feature information: the classification task determines the type of fire source, such as open flame or smoldering, by analyzing feature differences; the bounding box regression accurately predicts the location range of the fire source in the image based on feature distribution, forming spatial coordinate information. Simultaneously, the classification and regression processes are dynamically optimized using the Wise-IoU loss function, and the robustness of detection against complex backgrounds, occluded or blurred fire sources is improved by adaptively adjusting the weight coefficients. Ultimately, the module accurately outputs the fire source type and its corresponding spatial coordinates, providing reliable target localization information for subsequent firefighting operations.

[0048] The ASF-WIoU-YOLOv8 algorithm employs the Wise-IoU loss function, which adaptively adjusts the weight coefficients of each module in the ASF-WIoU-YOLOv8 algorithm to improve the algorithm's adaptability to the fire scene image data and optimize the recognition accuracy of the fire source type and spatial coordinates.

[0049] The Wise-IoU loss function used in the ASF-WIoU-YOLOv8 algorithm is an optimized upgrade of the traditional Cross-Union Ratio (CUI) loss function. Through a dynamic non-monotonic focusing mechanism, the Wise-IoU loss function adaptively adjusts the weight coefficients of each module in the ASF-WIoU-YOLOv8 algorithm, enabling the algorithm to better adapt to the complex characteristics of fire scene image data.

[0050] To address common issues in fire scene images, such as strong background interference, large differences in fire source scale, and occlusion or blurring of targets, the Wise-IoU loss function can dynamically allocate weights based on the quality of the training data: for high-quality samples, such as clear and complete fire source images, the weights are appropriately reduced to avoid overfitting; for low-quality samples, such as small targets, occluded or smoke-affected fire source images, the weights are increased to enhance the ASF-WIoU-YOLOv8 algorithm's learning of these types of samples.

[0051] This adaptive adjustment mechanism enhances the ASF-WIoU-YOLOv8 algorithm's adaptability to complex fire environments, enabling it to more accurately capture key fire source information during feature extraction, effectively preserve multi-scale fire source features during feature fusion, and optimize classification and bounding box regression accuracy during target detection. Ultimately, through dynamic optimization of the weights of each module in the ASF-WIoU-YOLOv8 algorithm, the accuracy of fire source classification and spatial coordinate positioning is significantly improved, providing more reliable target information for subsequent firefighting operations.

[0052] The process employs quaternions to represent the flight attitude parameters, constructs an attitude control model using an improved active disturbance rejection controller, and dynamically generates the current motor control quantity by calculating the flight attitude deviation between the flight attitude parameters and the target flight attitude parameters in real time. This quantity is then converted into a pulse width modulation duty cycle signal and output to the motor drive module of the firefighting drone. Specifically, this includes: The attitude sensor includes a gyroscope, an accelerometer, and a magnetometer. The gyroscope, the accelerometer, and the magnetometer respectively collect angular velocity data, linear acceleration data, and magnetic field strength data of the firefighting drone. The angular velocity data, the linear acceleration data, and the magnetic field strength data are combined to generate three-dimensional spatial attitude data, i.e., the flight attitude parameters. Quaternions are used to parameterize the flight attitude parameters and the target flight attitude parameters, resulting in quaternion attitude representations and target quaternion attitude representations, respectively. Based on the quaternion attitude representation, an improved active disturbance rejection controller is constructed. An extended state observer is used to estimate the internal dynamic disturbances and external environmental disturbances of the system in real time. A nonlinear error feedback mechanism is introduced to form a robust control model for the flight attitude parameters. The quaternion attitude representation is compared with the target quaternion attitude representation. The flight attitude deviation is calculated in real time through quaternion multiplication and normalization and input to the improved active disturbance rejection controller to generate the current motor control quantity. The continuous current motor control quantity output by the improved active disturbance rejection controller is limited and converted into the rotation speed command of the four rotors of the fire-fighting drone. The rotation speed command is converted into the pulse width modulation duty cycle signal through pulse width modulation technology and output to the motor drive module of the fire-fighting drone.

[0053] The attitude sensor consists of a gyroscope, an accelerometer, and a magnetometer. The gyroscope collects angular velocity data of the UAV around its three-dimensional axes, the accelerometer acquires linear acceleration data along each axis, and the magnetometer captures magnetic field strength data from the environment. These three types of data are aggregated and integrated to form three-dimensional spatial attitude data, or flight attitude parameters, that fully characterize the UAV's spatial attitude, providing the initial basis for subsequent attitude representation and control.

[0054] First, the attitude parameters can be parameterized using quaternions. Quaternions are used to represent both the flight attitude parameters and the target flight attitude parameters, resulting in corresponding quaternion attitude representations and target quaternion attitude representations. This representation method effectively avoids the gimbal lock problem that may occur in traditional Euler angle representations, ensuring the continuity and stability of the attitude description and laying the foundation for accurate attitude comparison and control.

[0055] Next, an improved active disturbance rejection control model can be built based on quaternion attitude representation. This model uses an extended state observer to sense and estimate the dynamic characteristic disturbances inside the system in real time, such as structural vibration and component errors, as well as external environmental disturbances, such as strong winds and hot air currents. At the same time, a nonlinear error feedback mechanism is introduced to enhance the ability to suppress complex disturbances, forming a robust attitude control model that ensures that control accuracy can still be maintained in complex fire environments.

[0056] Subsequently, the quaternion attitude representation can be compared with the target quaternion attitude representation, and the flight attitude deviation can be solved in real time through quaternion multiplication and normalization. Then, this deviation can be input into the improved disturbance rejection controller, which dynamically generates the current motor control quantity based on the magnitude and trend of the deviation. This control quantity directly reflects the force or torque required to adjust the attitude of the firefighting drone.

[0057] Finally, the continuous motor control output of the improved active disturbance rejection controller can be limited to prevent it from exceeding the motor's operating range, and then converted into speed commands adapted to the four rotors of the firefighting drone. Using pulse width modulation (PWM) technology, the speed commands are converted into PWM duty cycle signals and transmitted to the motor drive module, thereby driving each rotor to adjust its speed and direction according to the commands, ultimately achieving real-time and precise control of the firefighting drone's flight attitude.

[0058] The quaternion attitude representation is obtained by parameterizing the flight attitude parameters using quaternions. Specifically, it includes: For the angular velocity data, the angular velocity data is represented as ,in, These represent the rotational angular velocities of the firefighting drone around the x, y, and z axes, respectively. The angular velocity data is parameterized using quaternions to obtain the angular velocity quaternion attitude representation. ,in, The scalar part representing the attitude representation of the angular velocity quaternion. The vector part representing the attitude representation of the angular velocity quaternion; The angular velocity quaternion attitude representation The time derivative is ,in, The scalar part representing the time reciprocal of the attitude representation of the angular velocity quaternion. The vector part representing the time derivative of the attitude representation of the angular velocity quaternion; The angular velocity quaternion attitude representation and time derivative The angular velocity data w satisfies the following conversion relationship:

[0059] After unfolding, we get:

[0060]

[0061]

[0062]

[0063] By discretizing the integral, the initial state's angular velocity quaternion attitude is represented using the transformation relation. Update:

[0064] In the formula, The quaternion attitude representation of the angular velocity at time k+1; The quaternion attitude representation of the angular velocity at time k; The time derivative of the quaternion attitude representation of the angular velocity at time k; This represents the update time interval; k represents the time step, i.e., the update time step. The updated angular velocity quaternion attitude representation Perform normalization until the condition is met. ; For the linear acceleration data, the linear acceleration data is represented as ,in, Let x, y, and z represent the acceleration components of the firefighting drone along the x, y, and z axes, respectively. Calculate the flight pitch angle of the firefighting drone. and flight roll angle :

[0065]

[0066] The linear acceleration quaternion attitude representation is obtained by parameterizing the linear acceleration data using quaternions. ; The magnetic field strength data is represented as follows: ,in, These represent the magnetic field strength components of the firefighting drone along the x, y, and z axes, respectively. From the quaternion attitude representation Extracting the quaternion attitude vector part ; Projecting the magnetic field strength data m onto a horizontal plane yields the horizontal component of the magnetic field strength data m. :

[0067] After unfolding, we get:

[0068]

[0069]

[0070] Calculate the flight heading angle of the firefighting drone. :

[0071] The magnetic field strength data is parameterized using quaternions to obtain the magnetic field strength quaternion attitude representation. ; The complementary filtering algorithm is used to fuse the angular velocity quaternion attitude representation. The linear acceleration quaternion attitude representation The magnetic field strength quaternion attitude representation Generate the quaternion attitude representation q:

[0072] The steps for parameterizing the target flight attitude parameters using quaternions to obtain the target quaternion attitude representation are the same as above.

[0073] It should be noted that the angular velocity data is first represented as a set of rotational angular velocities around three-dimensional coordinate axes, where each element corresponds to the rotational angular velocity of the firefighting drone around these three axes. Then, it is parameterized using quaternions to obtain the angular velocity quaternion attitude representation. This quaternion contains a scalar part and a vector part, and its time derivative has a certain transformation relationship with the angular velocity data. After expansion, the specific correlation formulas for each component can be obtained. Through discretized integration, this transformation relationship is used to iteratively update the initial state angular velocity quaternion attitude representation. During the update process, the angular velocity quaternion attitude representation at the next time step is equal to the angular velocity quaternion attitude representation at the current time step plus the product of the time derivative of the current angular velocity quaternion attitude representation and the update time interval. The time parameter involved is used to identify different update time steps, and after the update, the obtained angular velocity quaternion attitude representation needs to be normalized until the quaternion normalization constraint condition is met.

[0074] For linear acceleration data, it is represented as a set of acceleration components along three-dimensional coordinate axes, where each element corresponds to the acceleration component of the firefighting drone on these three axes. Based on these components, the flight pitch angle and roll angle are calculated, and then quaternions are used to parameterize these angle information to generate a linear acceleration quaternion attitude representation.

[0075] For magnetic field strength data, it is represented as a set of magnetic field strength components along three-dimensional coordinate axes, where each element corresponds to the magnetic field strength component of the firefighting drone on these three axes. The vector part of the quaternion attitude is extracted from the existing quaternion attitude representation, and the magnetic field strength data is projected onto the horizontal plane through projection operations to obtain the horizontal components. After expansion, the specific expression of each horizontal component can be clearly defined. The flight heading angle is calculated based on the horizontal components, and then parameterized in quaternion form to obtain the magnetic field strength quaternion attitude representation.

[0076] Finally, a complementary filtering algorithm can be used to fuse the aforementioned angular velocity quaternion attitude representation, linear acceleration quaternion attitude representation, and magnetic field strength quaternion attitude representation to generate the final quaternion attitude representation. The complementary filtering algorithm can combine the characteristics of different sensor data. Angular velocity data has high accuracy over short periods, suitable for reflecting rapid attitude changes; linear acceleration data and magnetic field strength data have good stability over long periods, providing an absolute reference for attitude. Through reasonable weight allocation and filtering, the limitations of single data types are overcome, ultimately generating a more accurate and stable quaternion attitude representation that comprehensively reflects the true attitude of the firefighting drone in three-dimensional space.

[0077] Furthermore, the process of using quaternions to parameterize the target flight attitude parameters to obtain the target quaternion attitude representation is consistent with the above-mentioned flight attitude parameter processing steps.

[0078] The quaternion pose representation q is compared with the target quaternion pose representation. The comparison is performed, and the quaternion attitude deviation corresponding to the flight attitude deviation is calculated in real time through quaternion multiplication and normalization operations:

[0079] Extract the attitude deviation vector corresponding to the quaternion attitude deviation:

[0080] The attitude deviation vector is then input into the improved active disturbance rejection controller to generate the current motor control quantity. :

[0081] In the formula, Indicates the proportional gain coefficient; Represents the differential gain coefficient; Represents the angular velocity error vector; This represents the total disturbance estimated by the extended state observer, which is the sum of internal dynamic disturbances and external environmental disturbances.

[0082] It should be noted that the target quaternion attitude representation can be processed inversely, and then multiplied with the quaternion attitude representation. After normalization, a quaternion attitude deviation that reflects the flight attitude deviation is obtained. This quaternion attitude deviation fully contains the rotational information required to adjust from the current attitude to the target attitude, including the rotation axis and rotation angle.

[0083] Subsequently, the corresponding attitude deviation vector can be extracted from the quaternion attitude deviation. This process is achieved by processing the vector part of the quaternion attitude deviation and adjusting it in conjunction with the sign characteristics of its scalar part to ensure that the attitude deviation vector can accurately reflect the direction and magnitude of the minimum rotation path, providing an intuitive error metric for subsequent control.

[0084] Furthermore, the extracted attitude deviation vector can be input into the improved self-disturbance rejection controller, which integrates multiple information sources to generate the current motor control input. Specifically, the proportional gain coefficient multiplied by the attitude deviation vector forms the proportional term, which directly generates control action based on the current deviation magnitude, quickly responding to attitude deviations; the differential gain coefficient multiplied by the angular velocity error vector forms the differential term, providing damping based on the rate of change of the deviation, suppressing oscillations during attitude adjustment; and the total disturbance estimated by the extended state observer is used to compensate for internal dynamic disturbances, such as component characteristic fluctuations, and external environmental disturbances, such as airflow disturbances, thereby enhancing the robustness of UAV attitude control.

[0085] The combined effect of these three factors generates the current motor control quantity, which can precisely drive the firefighting drone to adjust from its current attitude to the target attitude, ensuring the accuracy and stability of attitude control in complex fire environments and providing a reliable attitude basis for subsequent fire extinguishing bomb throwing.

[0086] The current motor control quantity Represented as , This represents the roll moment of the firefighting drone around the x-axis, used to adjust the roll attitude of the firefighting drone. This represents the pitch moment of the firefighting drone around the y-axis, used to adjust the pitch attitude of the firefighting drone; This represents the yaw moment of the firefighting drone about the z-axis, used to adjust the drone's heading. The continuous current motor control quantity output by the improved active disturbance rejection controller. Amplitude limiting is applied and converted into rotational speed commands for the four rotors of the firefighting drone. :

[0087] In the formula, This command represents the rotational speed of the i-th rotor of the firefighting drone. This indicates the rotational speed command for the first rotor of the firefighting drone. This command indicates the rotational speed of the second rotor of the firefighting drone. This command indicates the rotational speed of the third rotor of the firefighting drone. This indicates the rotational speed command for the fourth rotor of the firefighting drone; b indicates the lift coefficient; l indicates the arm length, i.e., the distance from the rotor's rotation center to the geometric center of the firefighting drone's body; d indicates the drag coefficient; F indicates the total lift required by the firefighting drone. The speed command is transmitted using pulse width modulation technology. Converted into the pulse width modulation duty cycle signal :

[0088] In the formula, This indicates the minimum pulse width modulation duty cycle signal when the motor of the firefighting drone is operating normally. This indicates the maximum pulse width modulation duty cycle signal when the motor of the firefighting drone is operating normally; This indicates the minimum value of the speed command; Indicates the maximum value of the speed command; The pulse width modulation duty cycle signal is output to the motor drive module of the firefighting drone.

[0089] The current motor control quantities are represented in the form of a three-dimensional vector, including the roll torque about the x-axis, the pitch torque about the y-axis, and the yaw torque about the z-axis, which are used to adjust the roll attitude, pitch attitude, and heading of the firefighting drone, respectively.

[0090] To improve the continuous motor control output of the active disturbance rejection controller, the first step is to limit the amplitude to prevent it from exceeding the physical operating range of the motor. Then, it is converted into speed commands for the four rotors. This conversion process is based on rotor dynamics. A matrix relationship is constructed using the lift coefficient, arm length, drag coefficient, and required total lift to obtain the square value of each rotor speed command, thereby determining the speed command for each rotor. The lift coefficient reflects the correlation between rotor speed and lift, the arm length affects the torque generation efficiency, the drag coefficient relates to the drag characteristics during rotor rotation, and the total lift is used to balance the UAV's own weight and dynamic loads.

[0091] Speed ​​commands need to be converted into pulse width modulation (PWM) duty cycle signals using pulse width modulation (PWM) technology. This conversion process is based on the linear correspondence between speed commands and duty cycles. It uses the minimum and maximum duty cycles during normal motor operation as boundaries, and maps them to the minimum and maximum values ​​of the speed commands to ensure that the output duty cycle signal is within a safe operating range and accurately reflects the speed requirements.

[0092] Finally, the generated pulse width modulation duty cycle signal can be output to the motor drive module to drive each rotor to adjust its speed according to the command, thereby achieving precise control of roll, pitch, and yaw torque, thus completing the real-time adjustment of the attitude of the firefighting drone and ensuring flight stability in complex fire scene environments.

[0093] The process involves calculating the optimal throwing height, optimal launching angle, and optimal release timing based on the fire source type and spatial coordinates, the flight altitude data, the fire field temperature distribution data, the current flight attitude parameters, the current ambient wind speed, and the dynamic characteristics of the fire extinguishing projectile, thereby generating the current throwing execution command. Specifically, this includes: A three-dimensional fire scene model is constructed based on the fire source type and spatial coordinates. The flight altitude data is mapped to the three-dimensional fire field model to form a spatial positioning reference; the fire field temperature distribution data is fused into the three-dimensional fire field model to generate a thermal radiation intensity field; based on the three-dimensional fire field model and combined with the current flight attitude parameters, a relative coordinate system of the UAV is established to form a comprehensive fire field situation map. Based on the comprehensive fire situation map and the current environmental wind speed, an atmospheric flow field model is constructed, and the influence of wind speed gradients at different altitudes and thermal updrafts on the flight trajectory of fire extinguishing bombs is analyzed, establishing a thermal flow-airflow coupling model. Based on the aforementioned heat flow-air flow coupling model, the impact of environmental disturbances on the flight trajectory of the fire extinguishing projectile is quantified through fluid dynamics simulation, and a dynamic compensation parameter set is generated. Based on the dynamic compensation parameter set and combined with the dynamic characteristics of the fire extinguishing projectile, a ballistic trajectory prediction optimization model is constructed. The ballistic trajectory prediction optimization model takes the accuracy of the fire extinguishing projectile's landing point, the coverage range of the fire extinguishing agent, and the penetration capability of the fire extinguishing projectile as objectives, and takes the maneuverability of the fire-fighting drone and the dynamic changes in the fire scene as constraints. The model uses an intelligent optimization algorithm to iteratively solve the parameter combination of the optimal throwing height, the optimal throwing angle, and the optimal release timing. The calculated combination of the optimal throwing height, optimal throwing angle, and optimal release timing is converted into the current throwing execution command, which includes flight altitude adjustment parameters, flight attitude calibration parameters, and fire extinguishing grenade release timing parameters.

[0094] In practical applications, a three-dimensional fire scene model reflecting the spatial structure of the fire can be constructed based on the type of fire source and its spatial coordinates. First, flight altitude data is mapped onto this model to establish a spatial positioning benchmark. Then, fire temperature distribution data is fused to generate a thermal radiation intensity field. This field not only clearly shows the differences in heat intensity across different areas of the fire but also reflects the potential spread trend of the fire through temperature gradient changes, such as the direction and speed of diffusion from high-temperature areas to low-temperature areas. Finally, a relative coordinate system for the UAV is established based on the current flight attitude parameters. This coordinate system uses the UAV itself as the reference origin and converts information such as the fire source location and thermal radiation intensity field into spatial data relative to the UAV. Ultimately, this data is integrated to form a comprehensive fire situation map, thus fully reflecting the correlation between the fire source location, the environmental thermal state, and the UAV's spatial attitude.

[0095] Based on a comprehensive fire situation map and current environmental wind speed, an atmospheric flow field model reflecting the airflow state around the fire is constructed. This model focuses on the wind speed gradient characteristics at different altitudes, that is, the differences in wind speed magnitude and direction with altitude, such as lower wind speeds near the ground and higher wind speeds at higher altitudes. This difference directly leads to different air resistance experienced by the fire extinguishing projectiles at different stages of flight. Simultaneously, the influence of thermal updrafts caused by the high temperature at the fire site on the flight trajectory of the fire extinguishing projectiles is analyzed. Thermal updrafts generate upward thrust, which may cause the flight path of the fire extinguishing projectiles to deviate upwards. By coupling the interaction between the thermal radiation intensity field and the atmospheric flow field model—for example, heat flow changes the density and direction of local airflow, while airflow affects the propagation range and intensity of thermal radiation—a heat flow-airflow coupling model is finally established to accurately characterize the comprehensive interference of the complex environment on the trajectory.

[0096] Based on a heat-fluid-airflow coupling model, fluid dynamics simulation technology is used to simulate the flight process of a fire extinguishing projectile under different environmental conditions. The simulation process calculates in detail the magnitude and direction of various forces, such as airflow impact force and thermal buoyancy, experienced by the projectile, thereby quantifying the impact of environmental disturbances on the projectile's flight trajectory. Based on these quantification results, a dynamic compensation parameter set is generated, which includes trajectory correction coefficients and velocity compensation values ​​for different environmental disturbances. This provides data support for subsequent trajectory correction, ensuring that the fire extinguishing projectile can approach the expected flight path as closely as possible in complex environments.

[0097] Based on a dynamic compensation parameter set and combined with the dynamic characteristics of fire extinguishing projectiles, such as ballistic laws and airburst characteristics, a ballistic trajectory prediction optimization model is constructed. This model focuses on the accuracy of the fire extinguishing projectile's impact point, the coverage area of ​​the extinguishing agent, and the penetration capability of the projectile. Simultaneously, it incorporates the maneuverability of the UAV and the dynamic changes in the fire scene as constraints. UAV maneuverability includes maximum payload and attitude adjustment rate, while fire scene dynamic changes include the fire spread rate. The model is iteratively solved using a simulated annealing algorithm, continuously trying different combinations of throwing height, launching angle, and release timing, evaluating their performance in terms of target achievement and constraint satisfaction, and gradually optimizing and determining the optimal combination of throwing height, launching angle, and release timing.

[0098] The combined parameters obtained from the solution are transformed into the current delivery command, which includes flight altitude adjustment parameters, flight attitude calibration parameters, and fire extinguishing grenade release timing parameters, providing a precise operational basis for the fire extinguishing grenade delivery system.

[0099] Among these parameters, the flight altitude adjustment parameter ensures the firefighting drone reaches the optimal throwing height. This parameter, based on the fire source's spatial coordinates and flight altitude data, adjusts the vertical distance to ensure the fire extinguishing projectile is released at the optimal spatial position. The flight attitude calibration parameter adjusts the drone's pitch, roll, and yaw attitude to match the optimal throwing angle and counteract the impact of current attitude deviations on the trajectory. The fire extinguishing projectile release timing parameter controls the timing of the projectile release, linking it to the optimal release moment to ensure the projectile is released at the predetermined time to accurately cover the fire source. These three parameters work together to form the core of the throwing execution command, ensuring the fire extinguishing projectile accurately targets the fire source.

[0100] Example 2 In addition, this invention also discloses a firefighting drone attitude control and throwing system for complex environments, the system comprising: The data acquisition and noise reduction module utilizes the color charge-coupled device (CCD) payload, thermal imaging system, attitude sensor, and laser rangefinder mounted on the fire-fighting drone to acquire fire scene images, temperature distribution, flight attitude parameters, and altitude data, and uses an adaptive filtering algorithm for noise reduction. The fire source detection module processes fire scene images based on an improved target detection algorithm. It extracts features of fire sources of different sizes through multi-scale feature fusion, optimizes and improves the recognition accuracy of small targets by combining dynamic loss function, and outputs the fire source category and spatial coordinates. The attitude control module uses quaternions to represent flight attitude parameters, combines an improved active disturbance rejection controller to build a model, calculates attitude deviations in real time, dynamically generates motor control quantities, and converts them into pulse width modulation duty cycle signals for output. The throwing decision module calculates the optimal throwing parameters and generates execution instructions based on fire source information, flight altitude, temperature distribution, attitude, wind speed and fire extinguishing bomb dynamic characteristics. The execution and adjustment module executes the throwing according to instructions, monitors the trajectory with a visual sensor, and adjusts the attitude control model when the deviation exceeds the limit.

[0101] Specifically, such as Figure 2 The diagram shown is a system block diagram of a firefighting drone attitude control and throwing system for complex environments provided in an embodiment of the present invention. The system includes: The multi-source data acquisition module is used to acquire fire scene image data, fire scene temperature distribution data, flight attitude parameters and flight altitude data through the color CCD payload, thermal imaging system, attitude sensor and laser rangefinder carried by the fire-fighting drone, and to perform noise reduction processing on the fire scene image data, the fire scene temperature distribution data, the flight attitude parameters and the flight altitude data using an adaptive filtering algorithm. The target fire source identification module is used to process the fire scene image data based on the improved target detection algorithm, extract feature information of fire sources of different sizes through a multi-scale feature fusion mechanism, improve the identification accuracy of small target fire sources by combining dynamic loss function optimization strategy, and output the fire source category and spatial coordinates. The control signal output module is used to represent the flight attitude parameters using quaternions, construct an attitude control model in combination with an improved active disturbance rejection controller, dynamically generate the current motor control quantity by calculating the flight attitude deviation between the flight attitude parameters and the target flight attitude parameters in real time, and convert it into a pulse width modulation duty cycle signal to output to the motor drive module of the fire-fighting drone. The throwing instruction generation module is used to calculate the optimal throwing height, optimal throwing angle and optimal release timing based on the fire source type and spatial coordinates and the flight altitude data, combined with the fire temperature distribution data, current flight attitude parameters, current ambient wind speed and fire extinguishing bomb dynamic characteristics, and generate the current throwing execution instruction. The fire extinguishing bomb throwing module is used to execute the current throwing action through the fire extinguishing bomb throwing system of the fire-fighting drone based on the current throwing execution command. At the same time, it uses the visual sensor of the fire-fighting drone to monitor the flight trajectory of the fire extinguishing bomb. If the flight trajectory deviation of the fire extinguishing bomb exceeds the preset trajectory deviation range, the attitude control model is adjusted.

[0102] Figure 2 The apparatus of the illustrated embodiment can be used to perform corresponding actions. Figure 1 The steps in the method embodiments shown are implemented in a similar manner and have similar technical effects, and will not be repeated here.

[0103] Example 3 Meanwhile, the present invention also discloses an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the processor runs the computer program stored in the memory, the processor executes the steps of the attitude control and throwing method for firefighting drones in complex environments as described in any of the above.

[0104] like Figure 3 The diagram shown is a hardware structure schematic of an electronic device according to an embodiment of the present invention. The electronic device 30 includes: a processor 31, a memory 32, and a computer program; wherein... The memory 32 is used to store the computer program, and the memory may also be flash memory. The computer program is, for example, an application program or functional module that implements the above method.

[0105] The processor 31 is configured to execute the computer program stored in the memory to implement the various steps performed by the device in the above method. For details, please refer to the relevant descriptions in the preceding method embodiments.

[0106] Alternatively, the memory 32 can be either standalone or integrated with the processor 31.

[0107] When the memory 32 is a device independent of the processor 31, the device may further include: Bus 33 is used to connect the memory 32 and the processor 31.

[0108] A readable storage medium storing a computer program, which, when executed by a processor, is used to implement the steps of the attitude control and throwing method for firefighting drones in complex environments as described in any of the above claims.

[0109] The readable storage medium can be a computer storage medium or a communication medium. A communication medium includes any medium that facilitates the transfer of computer programs from one location to another. A computer storage medium can be any available medium accessible to a general-purpose or special-purpose computer. For example, a readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application-Specific Integrated Circuit (ASIC). Alternatively, the ASIC can be located in a user equipment. Of course, the processor and the readable storage medium can also exist as discrete components in a communication device. The readable storage medium can be a read-only memory (ROM), random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage device, etc.

[0110] The present invention also provides a program product including executable instructions stored in a readable storage medium. At least one processor of the device can read the executable instructions from the readable storage medium, and the at least one processor executes the executable instructions to cause the device to implement the methods provided in the various embodiments described above.

[0111] In the embodiments of the above-described device, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly manifested as execution by a hardware processor, or execution by a combination of hardware and software modules within the processor.

[0112] Through the above embodiments, this invention provides a method and system for attitude control and deployment of firefighting drones in complex environments. The method utilizes a color CCD payload, thermal imaging system, attitude sensor, and laser rangefinder mounted on the firefighting drone to collect fire scene image data, fire temperature distribution data, flight attitude parameters, and flight altitude data of the target fire area. An adaptive filtering algorithm is used to denoise the fire scene image data, fire temperature distribution data, flight attitude parameters, and flight altitude data. An improved target detection algorithm is used to process the fire scene image data, extracting feature information of fire sources of different sizes through a multi-scale feature fusion mechanism. A dynamic loss function optimization strategy is combined to improve the recognition accuracy of small target fire sources, outputting the fire source category and spatial coordinates. Quaternions are used to represent flight attitude parameters, and an attitude control model is constructed using an improved active disturbance rejection controller. The flight attitude parameters are calculated in real time. The system dynamically generates the current motor control quantity based on the flight attitude deviation between the target and the target flight attitude parameters, and converts it into a pulse width modulation duty cycle signal for output to the motor drive module of the fire-fighting drone. Based on the fire source type, spatial coordinates, and flight altitude data, combined with fire temperature distribution data, current flight attitude parameters, current ambient wind speed, and the dynamic characteristics of the fire extinguishing projectile, it calculates the optimal throwing height, optimal throwing angle, and optimal release timing, generating the current throwing execution command. Based on this command, the fire-fighting drone's fire extinguishing projectile delivery system executes the current throwing action. Simultaneously, the fire-fighting drone's visual sensors monitor the fire extinguishing projectile's flight trajectory. If the trajectory deviation exceeds the preset deviation range, the attitude control model is adjusted. This achieves stable flight of the fire-fighting drone in complex fire environments, accurate fire source identification, and efficient fire extinguishing projectile delivery, ensuring the safety and effectiveness of fire-fighting operations.

[0113] This invention employs an improved ASF-WIoU-YOLOv8 algorithm, which enhances the feature representation capability of multi-scale fire sources by weighted fusion of fire scene feature maps at different scales through an attention-scale sequence fusion mechanism. Combined with the Wise-IoU loss function to dynamically adjust weight coefficients, it optimizes the accuracy of target classification and bounding box regression in complex backgrounds, solving the problem of poor detection performance of small target fire sources in traditional algorithms. This ensures accurate output of fire source category and spatial coordinates, providing a reliable basis for subsequent throwing decisions. This invention uses quaternions to represent flight attitude parameters, effectively avoiding the gimbal lock problem in Euler angle attitude calculation. Combined with an improved active disturbance rejection controller, it estimates internal and external disturbances in real time through an extended state observer and introduces a nonlinear error feedback mechanism to dynamically generate motor control quantities. This significantly improves the firefighting drone's resistance to environmental disturbances such as strong winds and hot air currents, ensuring that the firefighting drone can maintain stable flight attitude and adaptive attitude control under complex conditions such as high and low temperatures and electromagnetic interference, laying the foundation for accurate throwing. This invention constructs a heat flow-airflow coupling model based on fire source location, flight altitude, temperature distribution, ambient wind speed, and the dynamic characteristics of the fire extinguishing projectile. It quantifies the impact of environmental disturbances on the trajectory through fluid dynamics simulation and generates dynamic compensation parameters. Combined with an intelligent optimization algorithm, it iteratively solves for the optimal throwing height, angle, and timing. Furthermore, it uses a visual sensor to monitor the flight trajectory deviation of the fire extinguishing projectile in real time, triggering the attitude control model to make secondary adjustments, forming a closed-loop correction mechanism. This effectively reduces the landing point error of the fire extinguishing projectile, ensures the accuracy of the projectile's throwing, and significantly improves fire extinguishing efficiency and safety.

[0114] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for attitude control and deployment of firefighting drones in complex environments, characterized in that, include: Data Acquisition and Noise Reduction: Using the color charge-coupled device payload, thermal imaging system, attitude sensor and laser rangefinder carried by the fire-fighting drone, fire scene images, temperature distribution, flight attitude parameters and altitude data are collected respectively, and an adaptive filtering algorithm is used for noise reduction. Fire source detection: Based on an improved target detection algorithm, fire scene images are processed. Features of fire sources of different sizes are extracted through multi-scale feature fusion. Dynamic loss function is combined to optimize and improve the recognition accuracy of small targets. The fire source category and spatial coordinates are output. Attitude control: Quaternions are used to represent flight attitude parameters. A model is built by combining an improved active disturbance rejection controller. Attitude deviation is calculated in real time, motor control quantities are dynamically generated and converted into pulse width modulation duty cycle signals for output. Drop decision: Based on fire source information, flight altitude, temperature distribution, attitude, wind speed and fire extinguishing bomb dynamics, calculate the optimal drop parameters and generate execution instructions; Execution and Adjustment: The system executes the throwing operation according to instructions, monitors the trajectory using visual sensors, and adjusts the attitude control model when deviations exceed limits.

2. The attitude control and throwing method for firefighting drones in complex environments according to claim 1, characterized in that, The improved target detection algorithm is the ASF-WIoU-YOLOv8 algorithm, and the processing steps of the ASF-WIoU-YOLOv8 algorithm are as follows: The feature extraction module performs convolution and pooling operations on the input fire scene image data to extract the fire scene feature information corresponding to the fire scene image data, forming fire scene feature maps at different scales. An attention scale sequence fusion mechanism is introduced into the feature enhancement module to perform weighted fusion of the fire scene feature maps at different scales to generate a fused fire scene feature map. The fused fire scene feature map is processed by a channel attention module and a spatial attention module to highlight the feature information of the fire source area from the channel dimension and the spatial dimension, respectively, and generate an enhanced fire scene feature map. The target detection module performs target classification and bounding box regression on the enhanced fire scene feature map, and outputs the fire source category and spatial coordinates. The ASF-WIoU-YOLOv8 algorithm uses the Wise-IoU loss function, which adaptively adjusts the weight coefficients of each module in the ASF-WIoU-YOLOv8 algorithm.

3. The attitude control and throwing method for firefighting drones in complex environments according to claim 1, characterized in that, The method employs quaternions to represent flight attitude parameters, combines an improved active disturbance rejection controller to construct a model, calculates attitude deviations in real time, dynamically generates current motor control quantities, and converts them into pulse width modulation duty cycle signals for output. Specifically, this includes: The flight attitude parameters include angular velocity data, linear acceleration data, and magnetic field strength data. Quaternions are used to parameterize the flight attitude parameters and the target flight attitude parameters, resulting in quaternion attitude representations and target quaternion attitude representations, respectively. Based on the quaternion attitude representation, an improved active disturbance rejection controller is constructed. An extended state observer is used to estimate the internal dynamic disturbances and external environmental disturbances of the system in real time. A nonlinear error feedback mechanism is introduced to form a robust control model for the flight attitude parameters. The quaternion attitude representation is compared with the target quaternion attitude representation, the flight attitude deviation is calculated in real time and input to the improved active disturbance rejection controller to generate the current motor control quantity; The current motor control quantity is limited and converted into the rotation speed command of the four rotors of the fire-fighting drone. The rotation speed command is converted into the pulse width modulation duty cycle signal through pulse width modulation technology and output to the motor drive module of the fire-fighting drone.

4. The attitude control and throwing method for firefighting drones in complex environments according to claim 3, characterized in that, The angular velocity data is represented as ,in, These represent the rotational angular velocities of the firefighting drone around the x, y, and z axes, respectively; the linear acceleration data are expressed as... ,in, These represent the acceleration components of the firefighting drone along the x, y, and z axes, respectively; the magnetic field strength data are expressed as... ,in, These represent the magnetic field strength components of the firefighting drone on the x, y, and z axes, respectively.

5. The attitude control and throwing method for firefighting drones in complex environments according to claim 1, characterized in that, When comparing the quaternion attitude representation with the target quaternion attitude representation, the quaternion attitude deviation corresponding to the flight attitude deviation is calculated in real time through quaternion multiplication and normalization operations.

6. The attitude control and throwing method for firefighting drones in complex environments according to claim 1, characterized in that, The optimal throwing parameters are calculated based on fire source information, flight altitude, temperature distribution, attitude, wind speed, and the dynamic characteristics of the fire extinguishing projectile, and an execution command is generated, specifically as follows: A three-dimensional fire scene model is constructed based on the type of fire source and spatial coordinates; The flight altitude data is mapped onto the three-dimensional fire field model to form a spatial positioning reference; the fire field temperature distribution data is fused into the three-dimensional fire field model to generate a thermal radiation intensity field; based on the three-dimensional fire field model and combined with the current flight attitude parameters, a relative coordinate system of the UAV is established to form a comprehensive fire field situation map. Based on the comprehensive fire situation map and the current environmental wind speed, an atmospheric flow field model is constructed, and the influence of wind speed gradients at different altitudes and thermal updrafts on the flight trajectory of fire extinguishing bombs is analyzed, establishing a thermal flow-airflow coupling model. Based on the aforementioned heat flow-air flow coupling model, the impact of environmental disturbances on the flight trajectory of the fire extinguishing projectile is quantified through fluid dynamics simulation, and a dynamic compensation parameter set is generated. Based on the dynamic compensation parameter set and combined with the dynamic characteristics of the fire extinguishing projectile, a ballistic trajectory prediction optimization model is constructed. The ballistic trajectory prediction optimization model takes the accuracy of the fire extinguishing projectile's landing point, the coverage range of the fire extinguishing agent, and the penetration capability of the fire extinguishing projectile as objectives, and takes the maneuverability of the fire-fighting drone and the dynamic changes in the fire scene as constraints. The model uses an intelligent optimization algorithm to iteratively solve the parameter combination of the optimal throwing height, the optimal throwing angle, and the optimal release timing. The calculated combination of the optimal throwing height, optimal throwing angle, and optimal release timing is converted into the current throwing execution command, which includes flight altitude adjustment parameters, flight attitude calibration parameters, and fire extinguishing grenade release timing parameters.

7. A firefighting drone attitude control and throwing system for complex environments, characterized in that, include: The data acquisition and noise reduction module utilizes the color charge-coupled device payload, thermal imaging system, attitude sensor and laser rangefinder carried by the fire-fighting drone to acquire fire scene images, temperature distribution, flight attitude parameters and altitude data, and uses an adaptive filtering algorithm to reduce noise. The fire source detection module processes fire scene images based on an improved target detection algorithm. It extracts features of fire sources of different sizes through multi-scale feature fusion, optimizes and improves the recognition accuracy of small targets by combining dynamic loss function, and outputs the fire source category and spatial coordinates. The attitude control module uses quaternions to represent flight attitude parameters, combines an improved active disturbance rejection controller to build a model, calculates attitude deviations in real time, dynamically generates motor control quantities, and converts them into pulse width modulation duty cycle signals for output. The throwing decision module calculates the optimal throwing parameters and generates execution instructions based on fire source information, flight altitude, temperature distribution, attitude, wind speed and fire extinguishing bomb dynamic characteristics. The execution and adjustment module executes the throwing according to instructions, monitors the trajectory with a visual sensor, and adjusts the attitude control model when the deviation exceeds the limit.

8. A computer device / equipment / system, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method as described in any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method as described in any one of claims 1 to 6.

10. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method as described in any one of claims 1 to 6.

Images