A multi-light fusion target detection method based on high-altitude air-drop coaxial unmanned aerial vehicle

Abstract

The application relates to a multi-light fusion target detection method based on high-altitude air-drop coaxial unmanned aerial vehicle, belonging to the field of unmanned aerial vehicle and photoelectric detection, and comprising the following steps: the coaxial unmanned aerial vehicle is separated from an aircraft at a high altitude, adaptive attitude back correction is performed, the unmanned aerial vehicle is adaptively back corrected to a stable descending attitude; during the stable descending process of the unmanned aerial vehicle, the confidence weight of a multi-optical sensor is dynamically adjusted based on real-time absolute height of falling; based on the confidence weight, multi-modal feature data collected by each optical sensor is fused and judged, a target is identified and continuously confirmed in the continuous descending process, and the three-dimensional space position of the target relative to the coaxial unmanned aerial vehicle is solved in real time; the motion state of the target is predicted based on the three-dimensional space position, and an expected accompanying tracking position located above the target is dynamically generated based on the prediction result; flight control instructions are generated to guide the unmanned aerial vehicle to continuously approach and track the target; when the detection task is completed and the height of the unmanned aerial vehicle is lowered to a preset landing threshold, landing recovery is performed.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) and photoelectric detection technology, and in particular to a multi-light fusion target detection method based on a high-altitude airdropped coaxial UAV. Background Technology

[0002] With the evolution of wide-area search and rescue, target search, and situational awareness missions in modern complex environments, extremely high demands are placed on reconnaissance capabilities that are timely, deep-penetrating, and cross-regional. Currently, wide-area target search mainly relies on satellites, high-altitude long-endurance UAVs, or ground-launched small and medium-sized UAVs. However, satellites have limited transit windows and are easily constrained by orbital patterns; large high-altitude long-endurance UAVs are expensive and lack deployment flexibility; ground-launched small and medium-sized UAVs are limited by battery capacity and flight speed, resulting in very limited range and altitude, making it impossible to quickly reach distant rear areas for sudden reconnaissance. "High-altitude deployment and low-altitude reconnaissance," as a new application model combining the high-speed maneuverability of large carrier aircraft with the close-range detection advantages of small and medium-sized UAVs, is becoming an important development direction in this field.

[0003] However, existing drones are only capable of performing 8,000 High-altitude airdrop missions face severe physical environment and flight control challenges. 8000 The extremely low air density at high altitudes causes conventional rotary-wing drones to experience a sharp decrease in lift after detaching from their carrier aircraft, making them highly susceptible to uncontrollable tumbling. Forcibly starting the motors and rotors directly in this thin air environment not only fails to generate effective thrust but also easily leads to motor overload and damage, or even catastrophic failures such as coaxial rotor "flailing." Therefore, achieving a smooth transition from freefall to a stable and controllable flight attitude and safely completing power startup in such extremely thin air remains a challenging problem that has yet to be properly solved.

[0004] In addition, from 8000 During their long vertical descent from high altitude to the ground, drones need to penetrate multiple layers of the troposphere, thick clouds, and near-surface fog. Traditional reconnaissance drones typically carry only visible light payloads or a single infrared payload. Visible light payloads become completely ineffective when penetrating clouds and fog, and are limited by day and night lighting conditions; while single infrared payloads offer some night vision and penetration capabilities, their resolution is low, and they are prone to missed detections and false positives when facing targets covered with infrared camouflage nets or using highly biomimetic camouflage coatings. Existing dual-light detection methods that only include visible and infrared light are no longer sufficient to meet the high-reliability target identification requirements under complex weather conditions, all-weather conditions, and heavily camouflaged environments due to the limitations of spectral information. Summary of the Invention

[0005] Based on the above analysis, the present invention aims to provide a multi-light fusion target detection method based on a high-altitude airdrop coaxial UAV, in order to solve the technical problems of existing UAVs having difficulty starting up in extremely high-altitude airdrop environments and insufficient detection performance under complex weather conditions.

[0006] The objective of this invention is mainly achieved through the following technical solutions: This invention provides a multi-optical fusion target detection method based on a high-altitude airdropped coaxial UAV, comprising the following steps: The coaxial UAV detaches from the carrier aircraft at high altitude and performs adaptive attitude correction, enabling the coaxial UAV to adaptively correct itself to a stable descent attitude; During the stable descent of the coaxial UAV, the confidence weights of the multiple optical sensors carried by the coaxial UAV are dynamically adjusted based on the real-time absolute descent altitude. Based on the confidence weights, the multimodal feature data collected by each optical sensor are fused and judged to identify and continuously confirm the target during the continuous descent, and the three-dimensional spatial position of the target relative to the coaxial UAV is calculated in real time. Based on the real-time calculation of the target's three-dimensional spatial position relative to the coaxial UAV, the target's motion state is predicted, and a desired accompanying tracking position located above and behind the target is dynamically generated based on the prediction results. Flight control commands are generated based on the desired accompanying tracking position and the current position of the coaxial UAV to guide the coaxial UAV to continuously approach and track the target. When the detection mission is completed and the coaxial UAV's altitude drops to a preset landing threshold, it is landed and recovered.

[0007] Furthermore, adaptive attitude correction is performed by actively stimulating the gyro precession effect based on the coaxial dual rotor speed difference, including: Calculate the attitude error of a coaxial UAV; Based on the acquired roll rate, pitch rate, and yaw rate of the coaxial UAV, as well as the attitude error, the desired control torque to bring the coaxial UAV back to center is calculated. Based on the desired control torque, the UAV's physical rotational mass, roll rate, and pitch rate, the target value of the optimal rotational speed difference between the two rotors is obtained; Calculate the residual aerodynamic blade tilt angle based on the target value of the optimal speed difference between the two rotors; Based on the base sustaining speed, the target value of the optimal speed difference between the two rotors, and the residual aerodynamic blade tilt angle set by the coaxial UAV flight control system, the coaxial UAV return-to-center control command is generated.

[0008] Furthermore, the desired control torque includes the desired roll torque. Desired pitch moment and desired yaw moment ; The target value for the optimal speed difference between the two rotors is as follows:

[0009] in, The target value for the optimal speed difference between the two rotors. The roll rate is angular velocity. The pitch angular velocity, For the physical rotational mass of the drone, To prevent extremely small constants with a denominator of zero.

[0010] Furthermore, the generation of coaxial UAV homing control commands includes: A first speed command is issued to the upper rotor drive motor of the coaxial UAV, and a second speed command is issued to the lower rotor drive motor of the coaxial UAV; wherein, the first speed command is the base maintenance speed plus half of the target value of the optimal speed difference between the two rotors; the second speed command is the base maintenance speed minus half of the target value of the optimal speed difference between the two rotors. Based on the residual aerodynamic blade tilt angle, a periodic pitch control command is issued to the tilting disk of the coaxial UAV.

[0011] Furthermore, the multiple optical sensors include a short-wave infrared camera, a long-wave infrared camera, a hyperspectral camera, and a visible light camera; the optical axes of each optical sensor are parallel to each other and are arranged in an array symmetrically around the central vertical axis of the camera load in space; Based on the real-time absolute fall height, the confidence weights of multiple optical sensors are dynamically adjusted, including: when At the same time, the highest confidence level is assigned to both the short-wave infrared camera and the long-wave infrared camera; when At that time, the hyperspectral camera is assigned a dominant confidence level; when At that time, based on the logistic inverse function model, the confidence weight of the visible light camera jumps as the height decreases, until it takes over the task of confirming the fine texture of the target. in, This represents the absolute height of the fall in real time. These are the first, second, and third height thresholds, respectively. .

[0012] Furthermore, based on the aforementioned confidence weights, the multimodal feature data collected by each optical sensor are fused and judged to identify and continuously confirm the target during the continuous descent, while simultaneously calculating the target's three-dimensional spatial position relative to the coaxial UAV in real time, including: Image data acquired by short-wave infrared camera, long-wave infrared camera, hyperspectral camera and visible light camera are extracted respectively. Candidate targets in each image data are detected and identified, and the probability of target existence corresponding to each optical sensor is generated. Obtain the confidence weights of each optical sensor at the current altitude, multiply the target existence probability of each optical sensor by the corresponding confidence weight, sum them up, and calculate the fusion decision probability. When the fusion decision probability is greater than or equal to a preset decision threshold, the existence of the target is confirmed and continuously tracked and confirmed. From images of at least two optical sensors with known relative positions, the pixel coordinates of the same target matching point are extracted. Combined with the intrinsic and extrinsic parameter matrices of the optical sensors, the three-dimensional spatial position coordinates of the target relative to the coaxial UAV are calculated.

[0013] Furthermore, based on the real-time calculated three-dimensional spatial position, the target's motion state is predicted, and based on the prediction results, a desired accompanying tracking position located above and behind the target is dynamically generated, including: The instantaneous velocity vector of the target is obtained by performing differential calculations on the target's three-dimensional spatial position calculated at consecutive time points. The instantaneous velocity vector of the target is smoothed by an exponential moving average filter to generate a smoothed velocity vector of the target. Based on the smoothed velocity vector of the target, the motion state of the target is determined and the predicted motion direction of the target is determined; Based on the preset safe tracking distance and height offset vector, the desired accompanying tracking position located above and behind the target is dynamically generated on the reverse extension line of the target's predicted motion direction.

[0014] Furthermore, the desired accompanying tracking position is as follows:

[0015] in, To accompany the tracking position, The target's current position, The preset safe tracking distance, Let be the unit vector representing the predicted direction of motion of the target. This is the preset height offset vector.

[0016] Furthermore, based on the desired accompanying tracking position and the current position of the coaxial UAV, flight control commands are generated to guide the coaxial UAV to continuously approach and track the target, including: Real-time calculation of the three-dimensional error vector between the desired tracking position and the current position of the coaxial UAV; Separate the horizontal and height components of the three-dimensional error vector; A proportional control law is used to convert the horizontal component into a horizontal speed command and the height component into a vertical speed command. The horizontal and vertical speed commands are sent to the UAV's flight control system to guide the UAV to the desired tracking position, thereby achieving continuous close-range tracking of the target.

[0017] Furthermore, when the reconnaissance mission is completed and the coaxial UAV descends to a preset landing threshold, a landing and recovery process is initiated, including: When the detection mission is completed and the coaxial UAV descends to the preset landing threshold, the downward-looking visible light camera is activated to collect ground images. Combined with the ground distance information obtained by the airborne laser ranging module, optical flow analysis is performed on the ground texture. Based on optical flow analysis and laser ranging information, a local three-dimensional elevation map of the area directly below the coaxial UAV is constructed. The surface flatness of the candidate landing points in the local three-dimensional elevation map is calculated, and the flat area with the smallest surface variance and no protruding obstacles is selected as the target landing point. Control the coaxial UAV to fly towards the target landing point, make the coaxial UAV descend vertically, and perform autonomous landing and recovery.

[0018] Compared with the prior art, the present invention can achieve at least one of the following beneficial effects: 1. This invention overcomes the bottleneck of flight safety in extreme high-altitude environments, achieving reliable attitude recovery. Addressing the problem of traditional rotor aerodynamic torque failure and the high risk of uncontrolled rollovers and crashes due to the extremely thin air at altitudes of 8000 meters, this invention innovatively proposes an attitude recovery method that actively stimulates gyro precession torque using the rotational speed difference of coaxial dual rotors. Through an optimal gyro torque allocation algorithm based on least-squares projection, the two-dimensional attitude recovery requirement is adaptively mapped to one-dimensional rotational speed difference control, supplemented by residual aerodynamic compensation. This allows the UAV to quickly recover to a stable descent attitude during freefall by utilizing the physical rotational mass of the rotors, completely solving the pain point of uncontrollable flight after deployment at extremely high altitudes. 2. This invention constructs a cross-spatial multimodal adaptive fusion detection system, significantly improving target recognition capabilities in complex environments. Addressing the inherent limitations of single or dual-light sensors in cloud and fog penetration, anti-camouflage, and all-weather detection, this invention establishes a dynamic confidence fusion strategy based on altitude perception. In high-altitude clouds, the Beer-Lambert law transmission model of short-wave infrared light is used to penetrate clouds and fog. In the mid-altitude region, a hyperspectral camera and its spectral angle mapping algorithm are introduced to accurately identify strongly camouflaged targets, such as those with infrared camouflage, through material spectral fingerprinting. In low-altitude regions, a visible light camera takes over for fine texture confirmation. The three-stage sensor system seamlessly switches adaptively with altitude, achieving all-altitude, all-weather, and highly reliable target detection from tens of thousands of meters above the ground. 3. This invention achieves intelligent prediction and dynamic positioning tracking, ensuring safe approach and autonomous recovery. Addressing the command lag and tracking collision risks inherent in traditional proportional control, this invention employs exponential moving average filtering to smoothly predict target motion and innovatively generates the desired accompanying tracking position dynamically above and behind the target. It combines flight control with camera gimbal arctangent closed-loop calculation to achieve close, smooth, and safe approach tracking of high-speed maneuvering non-cooperative targets. After mission completion, a local 3D elevation map is constructed by fusing downward visible light and laser ranging. The flattest area is automatically selected, and a vertical, compliant descent is executed, forming a complete intelligent closed loop from high-altitude deployment, detection, tracking to recovery.

[0019] In this invention, the above-described technical solutions can be combined with each other to achieve more preferred combinations. Other features and advantages of this invention will be set forth in the following description, and some advantages may become apparent from the description or be learned by practicing the invention. The objects and other advantages of this invention can be realized and obtained from what is particularly pointed out in the description and drawings. Attached Figure Description

[0020] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts.

[0021] Figure 1 This is a flowchart of a multi-light fusion target detection method based on a high-altitude airdropped coaxial UAV in an embodiment of the present invention; Figure 2 This is a flowchart of the four-light fusion detection process in an embodiment of the present invention; Figure 3 This is a schematic diagram of the payload structure of the four-light camera in an embodiment of the present invention.

[0022] Figure label: 1-Camera gimbal; 2-Visible light camera; 3-Hyperspectral camera; 4-Short-wave infrared camera; 5-Long-wave infrared camera. Detailed Implementation

[0023] Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not intended to limit the scope of the present invention.

[0024] There is an urgent need in this technical field for a novel unmanned aerial vehicle (UAV) system and method that can combine high-altitude self-stabilized descent initiation with multimodal fusion depth detection. This invention relates to a coaxial UAV capable of high-altitude deployment, self-stabilized flight, and target detection using a four-light fusion optoelectronic pod, and its control method.

[0025] The hardware system utilized in this invention mainly consists of a foldable coaxial dual-rotor UAV platform and a four-light camera payload.

[0026] like Figure 3 As shown, the four-light camera payload integrates four sensor modules: a short-wave infrared camera 4, a long-wave infrared camera 5, a hyperspectral camera 3, and a high-resolution visible light camera 2. In terms of structure and arrangement, the optical axes of the four sensor modules are mutually parallel and aligned, and they are arranged in a compact, symmetrical array around the central vertical axis of the camera payload.

[0027] Four optical sensor modules are fixed together inside the pod shell of the streamlined coaxial UAV. The lower half of the pod shell is equipped with a high-strength protective cover that encloses the four optical sensor modules.

[0028] ① Shortwave infrared camera 4, which uses its band characteristics to penetrate clouds and fog to obtain the target outline; ② Long-wave infrared camera 5, captures thermal radiation signals to detect distant heat sources; ③ Hyperspectral camera 3, to acquire continuous spectral reflectance information to extract spectral fingerprints of materials for identification of camouflage materials such as infrared camouflage nets or camouflage paint; ④ A high-resolution visible light camera 2 is used to provide target detail texture and color information under good lighting conditions.

[0029] Four optical sensor modules work together to achieve synchronous acquisition of multi-band feature data.

[0030] The foldable coaxial dual-rotor UAV and the four-light camera payload adopt a modular plug-and-play connection method.

[0031] The four-light camera payload is mounted at the bottom of the coaxial drone fuselage, providing an unobstructed omnidirectional hemispherical downward-facing field of view for the optical sensor. A detachable universal drone payload interface is fixedly located at the center of the bottom of the coaxial drone fuselage, and a quick-release connector (camera gimbal 1) that mates with the coaxial drone payload interface is fixedly located at the center of the top of the four-light camera payload.

[0032] The universal coaxial drone payload interface and quick-release connector are respectively integrated with mechanical locking mechanism and male and female electrical communication contact bar. After the two are connected, they can realize one-step rigid mechanical fixation between the camera payload and the coaxial drone, as well as blind plug-in interconnection of power supply and high-speed data link, realizing the structural detachable and plug-and-play of the four-light camera payload.

[0033] A specific embodiment of the present invention discloses a multi-optical fusion target detection method based on a high-altitude airdropped coaxial UAV, such as... Figure 1 As shown, it includes the following steps: Step S1: The coaxial UAV detaches from the carrier aircraft at high altitude and performs adaptive attitude correction to make the coaxial UAV adaptively correct to a stable descent attitude; Step S2: During the stable descent of the coaxial UAV, the confidence weights of the multiple optical sensors carried by the coaxial UAV are dynamically adjusted based on the real-time absolute descent altitude; based on the confidence weights, the multimodal feature data collected by each optical sensor are fused and judged, the target is identified and continuously confirmed during the continuous descent, and the three-dimensional spatial position of the target relative to the coaxial UAV is calculated in real time. Step S3: Based on the real-time calculation of the target's three-dimensional spatial position relative to the coaxial UAV, predict the target's motion state and dynamically generate the desired accompanying tracking position above and behind the target based on the prediction results; generate flight control commands based on the desired accompanying tracking position and the current position of the coaxial UAV to guide the coaxial UAV to continuously approach and track the target; when the detection mission is completed and the coaxial UAV's altitude drops to a preset landing threshold, perform landing and recovery.

[0034] Step S1, specifically.

[0035] This step involves the high-altitude deployment of the coaxial UAV and adaptive attitude homing based on optimal decoupling from gyroscope precession.

[0036] For example, coaxial drones from 8000 After detaching from the carrier aircraft at high altitude, it enters a state of disordered tumbling and free fall due to the disturbance of the initial release velocity and the effect of gravity.

[0037] In 8000 At high altitudes, where air density is extremely low, the aerodynamic restoring torque generated by traditional methods of using swashplates to periodically change blade pitch is almost ineffective.

[0038] Adaptive attitude correction is performed by actively stimulating gyro precession effect based on the speed difference of coaxial dual rotors, including: Calculate the attitude error of a coaxial UAV; Based on the acquired roll rate, pitch rate, and yaw rate of the coaxial UAV, as well as the attitude error, the desired control torque to bring the coaxial UAV back to center is calculated. Based on the desired control torque, the UAV's physical rotational mass, roll rate, and pitch rate, the target value of the optimal rotational speed difference between the two rotors is obtained; Calculate the residual aerodynamic blade tilt angle based on the target value of the optimal speed difference between the two rotors; Based on the base sustaining speed, the target value of the optimal speed difference between the two rotors, and the residual aerodynamic blade tilt angle set by the coaxial UAV flight control system, the coaxial UAV return-to-center control command is generated.

[0039] This invention innovatively proposes an attitude correction method that utilizes the speed difference of coaxial dual rotors to actively excite the gyro precession effect. The specific derivation and implementation process are as follows: First, the attitude error and the desired torque are calculated.

[0040] The flight control system of a coaxial UAV uses onboard inertial navigation sensors to obtain the real-time angular velocity of the coaxial UAV in the body coordinate system: roll rate. Pitch angular velocity yaw rate .

[0041] Roll rate The real-time angular velocity of the UAV rotating around the x-axis of its body; Pitch angular velocity The real-time angular velocity of the UAV rotating around the y-axis of its body; Yaw angular velocity This is the real-time angular velocity of the UAV rotating around its body axis.

[0042] Simultaneously, the projection vector of the current gravity direction in the body coordinate system is obtained, which is a 3×1 column vector, as shown below: Formula (1) in, This is the projection vector of the current gravity direction in the body coordinate system; These are the projection components of the gravity direction onto the x, y, and z axes of the body, respectively; the three components satisfy the normalization constraints. .

[0043] When the drone is horizontal, the projection component characteristics are represented as follows: Formula (2) To align the coaxial UAV vertically downwards, the target gravity vector is defined as follows:

[0044] in, Let be the target gravity vector.

[0045] When the coaxial drone tilts Non-zero values ​​will appear, the magnitude and sign of which reflect the direction and degree of tilt.

[0046] The flight control system of a coaxial UAV first uses two vectors , The cross product is used to calculate the attitude error of the coaxial UAV. As shown below: Formula (3) Attitude error Disassembled into , , respectively used in formulas (5) and (4).

[0047] Using a proportional-derivative control law, the desired three-axis control torque required to return the coaxial UAV to center is calculated. , , As shown below: Formula (4) Formula (5) Formula (6) in, These are the desired roll moment, desired pitch moment, and desired yaw moment for righting the coaxial UAV, respectively. For attitude scaling gain, the attitude error ( The coefficient for converting the example sentence into a correction factor indicates that the larger the value, the more aggressive the correction of the tilt. For angular velocity damping gain, the angular velocity The proportionality coefficient, which is converted into damping torque, has a larger value. The larger the value, the stronger the suppression of rotational motion, and the more it helps to prevent oscillation.

[0048] Desired Rolling Moment The desired torque that causes the coaxial UAV to rotate around the x-axis of the fuselage, used to correct tilt in the left and right directions; Desired pitch moment The desired torque that causes the coaxial UAV to rotate around the y-axis of the fuselage, used to correct tilt in the forward and backward directions; Desired yaw moment : The desired torque that causes the coaxial UAV to rotate around the z-axis of the fuselage, used to suppress the spin oscillation of the nose.

[0049] Formula (4) Physical meaning: Roll channel, determined by roll attitude error Generate corrective torque, while using roll velocity Provides damping to suppress oscillations in the roll direction; Formula (5) physical meaning: pitch channel, derived from pitch attitude error Generate corrective torque, while using pitch angular velocity Provide damping; The physical meaning of formula (6): yaw channel, no attitude error term (because yaw around the gravity axis does not affect attitude recovery), only yaw angular velocity is used. It generates a damping torque to suppress spin.

[0050] With the desired torque clearly defined Next, a gyroscopic dynamics model of the coaxial dual-rotor is established. Let the rotational speed of the upper rotor of the coaxial dual-rotor be... The lower rotor speed is The moment of inertia of a single rotor is .

[0051] The speed difference between the upper and lower rotors is The net angular momentum vector generated by the rotation of the twin rotors. Concentrated along the Z-axis of the body; The calculation is as follows: Formula (7) in, This is the net angular momentum vector generated by the rotation of the twin rotors.

[0052] When the aircraft undergoes a violent roll, the angular velocity of the aircraft interacts with the angular momentum of the rotor, generating a powerful gyroscopic precession torque. .

[0053] According to Euler rigid body dynamics, gyroscopic torque The calculation is as follows: Formula (8) Expanding equation (8) by cross product, we obtain the projections of the gyroscopic torque onto the roll along the x-axis and the pitch along the y-axis of the aircraft, as shown below: Formula (9) Formula (10) in, This represents the roll component of the gyroscopic torque along the X-axis of the machine. This represents the pitch component of the gyroscopic torque along the Y-axis of the machine.

[0054] The gyro torque depends entirely on the physical rotational mass of the rotor. and speed difference It is not affected by the thin air at high altitudes.

[0055] Furthermore, the optimal torque distribution solution is calculated for extremely thin air at high altitudes. Traditional coaxial UAVs rely on the lateral tilt angle of the propeller blades. and longitudinal tilt angle This generates aerodynamic roll torque. and aerodynamic pitch moment .

[0056] Aerodynamic coefficient is Then we have: Formula (11) Formula (12) in, It is the aerodynamic roll moment; It is the aerodynamic pitching moment.

[0057] Due to 8000 The aerodynamic coefficient at high altitudes is extremely small. If the desired torque is provided entirely by aerodynamics, the calculated blade tilt angle will exceed the mechanical limit and cause loss of control.

[0058] Therefore, this invention maximizes the use of gyroscopic torque to replace aerodynamic force by dynamically adjusting the rotational speed difference. This is an overdetermined equation problem in which a one-dimensional rotational speed difference control quantity satisfies the two-dimensional torque requirements of roll and pitch.

[0059] To find the optimal solution, this invention employs the least squares projection algorithm.

[0060] Let the desired roll and pitch moment vectors be... The coefficient vector of the gyroscopic effect is It must meet the following requirements: Formula (13) in, The moment of inertia of a single rotor.

[0061] By adjusting the speed difference between the upper and lower rotors The resulting gyroscopic torque is approximately equal to the roll and pitch control torques desired by the coaxial UAV flight control system.

[0062] Formula (13) uses The reason is: ①The desired roll and pitch moment vectors are There are two dimensions, but the control quantity There is only one, and one control quantity cannot perfectly satisfy the torque requirements of two dimensions at the same time; ② Least square solution: The physical objective is to minimize the error between the gyro torque and the desired torque (i.e., minimize the mean square error). The least squares projection algorithm is used to find the optimal speed difference target value for the dual rotors. This allows the gyro torque to approximate the desired torque as closely as possible, with the remaining small residuals then compensated by the periodic pitch variation of the conventional aerodynamic swashplate.

[0063] The desired control torque includes the desired roll torque. Desired pitch moment and desired yaw moment ; The target value for the optimal speed difference between the two rotors is as follows: Formula (14) in, The target value for the optimal speed difference between the two rotors. The roll rate is angular velocity. The pitch angular velocity, For the physical rotational mass of the drone, To prevent extremely small constants with a denominator of zero.

[0064] To minimize the mean square error between the gyro torque and the desired torque, the desired control torque vector is projected onto the coefficient vector using a dot product, thus deriving the target value for the optimal speed difference between the two rotors. .

[0065] Formula (14) accurately realizes the adaptive mapping of the two-dimensional attitude recovery requirement to the one-dimensional difference in the rotational speed of the upper and lower rotors.

[0066] Finally, residual compensation and motor drive are performed by a traditional aerodynamic swashplate for periodic pitch changes.

[0067] After calculating the target value of the optimal speed difference of the dual rotor Then, the optimal speed difference is substituted into the aforementioned gyroscope dynamics formulas (7)-(8), and the optimal speed difference is used. Substituting the value into formula (7) The torque actually provided by the gyro effect is calculated.

[0068] Since the rotational speed difference cannot fully meet the torque requirements in all dimensions, the remaining residual torque is compensated by the periodic pitch variation of the tilting disk of the coaxial UAV.

[0069] The coaxial UAV consists of, from top to bottom, an upper rotor, a swashplate mechanism, a lower rotor, a fuselage, a payload, and a landing gear.

[0070] Calculate the required residual aerodynamic blade tilt angle and As shown below: Formula (15) Formula (16) This decoupling formula significantly reduces the mechanical burden on the swashplate.

[0071] The generated coaxial UAV homing control command includes: A first speed command is issued to the upper rotor drive motor of the coaxial UAV, and a second speed command is issued to the lower rotor drive motor of the coaxial UAV; wherein, the first speed command is the base maintenance speed plus half of the target value of the optimal speed difference between the two rotors; the second speed command is the base maintenance speed minus half of the target value of the optimal speed difference between the two rotors. Based on the residual aerodynamic blade tilt angle, a periodic pitch control command is issued to the tilting disk of the coaxial UAV.

[0072] Ultimately, the flight control system maintains the rotational speed based on the set baseline. The upper rotor is driven by two motors on the upper and lower layers. Add half of the optimal speed difference, the lower rotor is The final control command is the result of subtracting half of the optimal speed difference.

[0073] The flight control system of the coaxial UAV executes the above steps in real time. During the free fall phase, the UAV actively adjusts the speed difference between the upper and lower propellers and uses the optimally excited physical gyro precession torque to perfectly compensate for the lack of high-altitude dynamics, quickly and adaptively returning from a disordered tumbling state to a stable state with the nose pointing vertically downward.

[0074] Step S1 is used to stimulate the gyro precession torque by actively adjusting the speed difference of the coaxial dual rotors under the condition that the traditional aerodynamic torque fails due to the extremely thin air at high altitudes. This enables the UAV to quickly and adaptively return to a stable descent attitude with the nose pointing vertically downwards from a state of disordered tumbling free fall.

[0075] Step S2, specifically.

[0076] This step involves cross-spatial multimodal detection and dynamic confidence fusion, such as... Figure 2 As shown.

[0077] During the stable descent of the coaxial drone, the confidence weights of the multiple optical sensors carried by the coaxial drone are dynamically adjusted based on the real-time absolute descent altitude.

[0078] The multiple optical sensors include a short-wave infrared camera, a long-wave infrared camera, a hyperspectral camera, and a visible light camera; the optical axes of each optical sensor are parallel to each other and are arranged in an array symmetrically around the central vertical axis of the camera load in space; Based on the real-time absolute fall height, the confidence weights of multiple optical sensors are dynamically adjusted, including: when At the same time, the highest confidence level is assigned to both the short-wave infrared camera and the long-wave infrared camera; when At that time, the hyperspectral camera is assigned a dominant confidence level; when At that time, based on the logistic inverse function model, the confidence weight of the visible light camera increases as the height decreases until it takes over the task of confirming the fine texture of the target; the confidence of the visible light camera is determined as shown in formula (21).

[0079] in, This represents the absolute height of the fall in real time. These are the first, second, and third height thresholds, respectively. .

[0080] For example, The distances are 8000m, 5000m, and 2000m respectively.

[0081] For example, the confidence weight allocation for each height interval is shown in Table 1.

[0082]

[0083] In each height interval, the sum of the four confidence weights is 1.0.

[0084] After the UAV completes adaptive return to center and establishes a controllable descent trajectory, the flight control system immediately initiates the cross-airspace multimodal detection program.

[0085] At this stage, the coaxial drone's altitude will increase from 8000 meters. As it gradually descends to the ground, its line of sight must penetrate clouds of varying thicknesses and near-surface smog. At 8000... Up to 5000 During the high-altitude phase, the visible light camera 2 became almost completely ineffective due to the strong scattering effect of clouds and fog. Therefore, the system prioritized the use of the short-wave infrared camera 4 and the long-wave infrared camera 5 for wide-area scanning detection.

[0086] This invention establishes a method for calculating short-wave infrared fog penetration based on an atmospheric scattering physics model. The extinction coefficient of light waves in the atmosphere is also discussed. With the wavelength of light waves Closely related, the flight control system of a coaxial UAV relies on real-time atmospheric visibility data obtained from onboard environmental sensors. The extinction distribution model is constructed as follows: Formula (17) in, The reference wavelength for visible light. The atmospheric scattering constant is determined by the size of cloud and fog particles.

[0087] According to Beer-Lambert's law, the light reflected from the target penetrates at a distance of... After passing through the cloud and fog band, the transmittance reaching the camera lens... As shown below: Formula (18) Because the wavelength is 1.5 The shortwave infrared wavelength in the frequency band is significantly larger than the visible light wavelength and the radius of most cloud and fog particles, causing its extinction coefficient to decrease exponentially, resulting in a substantial increase in transmittance. Based on the above physical formula, the flight control system sets rigorous camera switching and parameter adaptation rules: When airborne sensors measure high-altitude visibility 8000 At that time, the system forcibly activates short-wave infrared camera 4 and long-wave infrared camera 5 to perform wide-area scanning to detect non-cooperative targets. In order to compensate for the attenuation of photon quantity caused by absorption by high-altitude clouds, the flight control system of the coaxial UAV dynamically adjusts the integration time and digital gain of short-wave infrared camera 4 according to the calculated current transmittance, so as to reconstruct the high-contrast outline of the ground obscured by thick clouds and fog.

[0088] As the coaxial drone continued its descent and crossed the main cloud belt, it entered the 5000-meter range. By 2000 In the mid-to-high altitude zone, the hyperspectral camera 3 began to be fully engaged, conducting hyperspectral high-resolution detection.

[0089] At this extremely high altitude and from a distance, ground targets often appear as tiny, sub-pixel-level objects with no distinct shape in the image, rendering traditional visual recognition algorithms based on image geometric features completely ineffective. To address this challenge, the flight control system utilizes a hyperspectral camera to extract the measured pixels... A one-dimensional high-dimensional spectral feature vector is constructed from the light intensity reflectance sequence under several continuous subdivided wavelength bands: Formula (19) The onboard edge computing platform of the coaxial UAV will extract the spectral vector in real time and compare it with the reference spectral vectors of known targets such as specific metal anti-rust paint or carbon fiber composite materials in the material database built into the flight control system. Compare them.

[0090] This invention employs a spectral angle mapping algorithm, treating two spectral features as... Vectors in Euclidean space are determined by calculating the generalized spatial angle between them. To assess the similarity of materials, as shown below: Formula (20) The flight control system is set to operate within this altitude range, covering a frequency band of 400. Up to 1000 The visible and near-infrared region, with a spectral resolution set to 5. Since the spectral angle is only related to the shape of the spectral curve representing the vector direction, it is naturally immune to linear scaling interference caused by changes in high-altitude light intensity. Therefore, the flight control system sets its strict material matching tolerance threshold to 0.1. It can tolerate errors in the refraction of natural light while accurately distinguishing the unique red-edge reflection effect of chlorophyll in infrared camouflage nets from that in real vegetation. Using this material's spectral fingerprint, coaxial drones can precisely separate extremely small targets at high altitudes where there is no spatial contour information.

[0091] During the descent of the coaxial UAV, the flight control system deployed a sensor dynamic confidence fusion strategy based on altitude perception within the edge computing platform. Let the absolute descent altitude calculated in real-time by the coaxial UAV using a barometer be... The flight control system is equipped with a confidence weight distribution function that is dynamically updated with altitude for each of the four cameras.

[0092] At height 5000 When the distance is ≤8000m, the system will forcibly assign the highest confidence level to the infrared cameras (short-wave infrared camera 4 and long-wave infrared camera 5). Entering 2000 Up to 5000 During the interval, the weights of the system controlling the hyperspectral camera 3 surged to dominance in a parabolic model; When height 2000 Furthermore, as it gradually approaches the Earth's surface, the spatial resolution of the optical lens of the visible light camera 2 increases exponentially.

[0093] At this point, the flight control system sets the confidence weights for visible light camera 2 based on the logistic inverse function model, as shown below: Formula (21) in, The confidence weights for visible light camera 2; This refers to the fusion gain slope coefficient; This is the absolute height of the fall; is the threshold height at which meteorological and visual abilities intersect; exp(.) is the natural exponential function.

[0094] The threshold height for the intersection of meteorological and visual capabilities is set at 2000. The fusion gain slope coefficient was set to 0.01. According to the formula, when the drone drops to 1500... The visible light weight rapidly jumps to over 0.99. Based on this result, the flight control system automatically instructs the visible light camera 2 to activate continuous optical zoom, take over the task of confirming fine textures, and sets the electronic shutter speed of the visible light camera to above 1 / 2000. To overcome the image blurring caused by descent.

[0095] Based on the aforementioned confidence weights, the multimodal feature data collected by each optical sensor are fused and judged to identify and continuously confirm the target during continuous descent. Simultaneously, the three-dimensional spatial position of the target relative to the coaxial UAV is calculated in real time, including: Image data acquired by short-wave infrared camera, long-wave infrared camera, hyperspectral camera and visible light camera are extracted respectively. Candidate targets in each image data are detected and identified, and the probability of target existence corresponding to each optical sensor is generated. Obtain the confidence weights of each optical sensor at the current altitude, multiply the target existence probability of each optical sensor by the corresponding confidence weight, sum them up, and calculate the fusion decision probability. When the fusion decision probability is greater than or equal to a preset decision threshold, the existence of the target is confirmed and continuously tracked and confirmed. From images of at least two optical sensors with known relative positions, the pixel coordinates of the same target matching point are extracted. Combined with the intrinsic and extrinsic parameter matrices of the optical sensors, the three-dimensional spatial position coordinates of the target relative to the coaxial UAV are calculated.

[0096] The edge computing platform of the coaxial UAV extracts the target existence probability generated by four sensors, combines it with dynamic weights and sums it to obtain the final fusion decision probability, and maintains normalization constraints throughout the process.

[0097] For example, the preset decision threshold is 0.85.

[0098] The threshold setting takes into account both the noise interference from multi-sensor data and the accuracy requirements for target recognition under complex high-altitude weather conditions. When the fusion decision probability is ≥0.85, the flight control system determines that a non-cooperative target exists and immediately starts the close tracking mode, synchronously outputting the target's three-dimensional spatial coordinates to the flight control and navigation module; If the fusion decision probability is lower than 0.85, the flight control system will automatically trigger the wide-area scanning and multimodal feature extraction process, and dynamically adjust the parameters of each sensor (such as the band range of the hyperspectral camera and the integration time of the infrared camera) in combination with the current altitude until the fusion probability meets the threshold condition or it is confirmed that there is no effective target.

[0099] The function of step S2 is to adaptively allocate the confidence weights of the four types of cameras (shortwave infrared, longwave infrared, hyperspectral, and visible light) based on the real-time altitude during the entire process of the UAV descending from 8,000 meters to the ground, identify and continuously confirm the target through weighted fusion decision, and solve the target's three-dimensional spatial position to achieve continuous and stable detection across airspace and against camouflage.

[0100] Step S3 includes steps S31-S32.

[0101] This step involves close-range tracking and autonomous landing recovery based on predictive algorithms.

[0102] Step S31: Based on the real-time calculation of the target's three-dimensional spatial position relative to the coaxial UAV, predict the target's motion state, and dynamically generate the desired accompanying tracking position located above and behind the target based on the prediction results.

[0103] Based on the real-time calculated 3D spatial position, the target's motion state is predicted, and the expected accompanying tracking position located above and behind the target is dynamically generated based on the prediction results, including: The instantaneous velocity vector of the target is obtained by performing differential calculations on the target's three-dimensional spatial position calculated at consecutive time points. The instantaneous velocity vector of the target is smoothed by an exponential moving average filter to generate a smoothed velocity vector of the target. Based on the smoothed velocity vector of the target, the motion state of the target is determined and the predicted motion direction of the target is determined; Based on the preset safe tracking distance and height offset vector, the desired accompanying tracking position located above and behind the target is dynamically generated on the reverse extension line of the target's predicted motion direction.

[0104] Once the visible light camera 2 takes over the fine texture confirmation and calculates the continuous three-dimensional relative coordinates of the target in real time, the flight control system immediately starts the flight control algorithm based on target motion prediction to achieve close-range tracking.

[0105] The flight control system records the current moment. The target's three-dimensional position vector in the UAV's body coordinate system and the position vector of the previous time step The instantaneous velocity vector of the target is obtained through differential calculation, as shown below: Formula (22) in, The instantaneous velocity vector of the target; The target position vector at the current moment; The target position vector at the previous moment; This represents the time interval between two location samplings.

[0106] To eliminate sudden velocity changes caused by high-altitude wind disturbances or visual pixel fluctuations, the flight control system employs an exponential moving average filtering algorithm to smooth the instantaneous velocity, deriving a smoothed velocity vector. As shown below: Formula (23) in, The smoothing speed of the previous control cycle; These are the filter coefficients.

[0107] For example, the flight control system will filter the coefficients Set precisely to 0.3.

[0108] The scientific basis for this parameter setting is that a weighting ratio of 0.3 can quickly respond to the actual physical acceleration actions of ground vehicles or sea vessels within an extremely short control cycle, while also effectively attenuating high-frequency visual noise interference. Based on the smoothed horizontal velocity vector, the flight control system intelligently determines the target's motion state through a set velocity threshold.

[0109] When the target's horizontal velocity exceeds a threshold, the flight control system determines that the target is in an active escape state and directly uses the unit vector in the current velocity direction. As a predictor of the direction of motion; When the speed is below the threshold, the flight control system determines that the target is in a hovering or low-speed drifting state and continues to use the previously valid high-speed motion direction vector.

[0110] After accurately obtaining the predicted direction of motion of the target, the control algorithm dynamically generates an optimal desired tracking position. To prevent dangerous dive-collisions when the coaxial UAV encounters strong airflow, the desired tracking position is not to directly coincide with the target's current position, but rather based on a preset safe tracking distance. With height offset vector ,in, This is the height offset, the component of the drone along the Z-axis, which represents the height at which the drone should be above the target.

[0111] Construct a safe accompanying occupant point on the reverse extension line of the target's predicted motion direction.

[0112] The expected accompanying tracking position is as follows:

[0113] in, To accompany the tracking position, The target's current position, The preset safe tracking distance, Let be the unit vector representing the predicted direction of motion of the target. This is the preset height offset vector.

[0114] Step S32: Generate flight control commands based on the desired accompanying tracking position and the current position of the coaxial UAV to guide the coaxial UAV to continuously approach and track the target; when the detection mission is completed and the altitude of the coaxial UAV drops to a preset landing threshold, perform landing and recovery.

[0115] Based on the desired tracking position and the current position of the coaxial UAV, flight control commands are generated to guide the coaxial UAV to continuously approach and track the target, including: Real-time calculation of the three-dimensional error vector between the desired tracking position and the current position of the coaxial UAV; Separate the horizontal and height components of the three-dimensional error vector; A proportional control law is used to convert the horizontal component into a horizontal speed command and the height component into a vertical speed command. The horizontal and vertical speed commands are sent to the UAV's flight control system to guide the UAV to the desired tracking position, thereby achieving continuous close-range tracking of the target.

[0116] When the reconnaissance mission is completed and the coaxial UAV descends to a preset landing threshold, a landing and recovery process will be initiated, including: When the detection mission is completed and the coaxial UAV descends to the preset landing threshold, the downward-looking visible light camera is activated to collect ground images. Combined with the ground distance information obtained by the airborne laser ranging module, optical flow analysis is performed on the ground texture. Based on optical flow analysis and laser ranging information, a local three-dimensional elevation map of the area directly below the coaxial UAV is constructed. The surface flatness of the candidate landing points in the local three-dimensional elevation map is calculated, and the flat area with the smallest surface variance and no protruding obstacles is selected as the target landing point. Control the coaxial UAV to fly towards the target landing point, make the coaxial UAV descend vertically, and perform autonomous landing and recovery.

[0117] The flight control system of the coaxial UAV calculates the desired tracking position and the current position of the coaxial UAV in real time. The three-dimensional error vector between .

[0118] The flight control system separates the horizontal component of the error vector. and height components It then uses a proportional control law to linearly convert the command into the horizontal speed command required for UAV flight control. Vertical lift speed command As shown below: Formula (25) Formula (26) in, and These are the proportional gain coefficients for the horizontal and vertical directions, respectively. During this close-range flight, to ensure that the four-optical pods can remain stably aligned with the target, the flight control system synchronously extracts the relative position vector from the coaxial UAV body pointing towards the target. .

[0119] Decompose the relative vector into components in a three-dimensional orthogonal coordinate system. , as well as Then, the flight control system uses the arctangent function to calculate the yaw angle command required by the pod gimbal. With pitch angle command : Formula (27) Formula (28) The aforementioned angle commands are directly sent to the three-axis mechanically stabilized camera gimbal 1 on the pod, driving it to rotate in real time to ensure that the high-speed maneuvering target is always locked in the center of the camera's high-resolution field of view. When the reconnaissance mission is complete and the UAV's altitude has decreased to approximately 50 meters above the ground... At that time, the flight control system shuts down the tracking algorithm and uses the downward-looking visible light camera 2 to perform optical flow analysis on the ground texture.

[0120] Combined with the airborne laser ranging module, the flight control system quickly constructs a local 3D elevation map of the area directly below. The calculation unit performs a comprehensive calculation of the surface flatness variance of each candidate landing point, automatically selecting the absolutely flat area with the smallest surface variance and no protruding obstacles.

[0121] After selecting the landing site, the flight control system controls the coaxial brushless motor to gradually and smoothly reduce the collective thrust output, and finally executes a vertical and gentle autonomous landing and recovery procedure.

[0122] Step S3 is to dynamically generate a safe accompanying tracking point above and behind the target based on the target's three-dimensional position prediction and exponential moving average filtering. It then converts the tracking point into flight commands through a proportional control law to achieve close-range tracking. After the mission is completed, it guides the UAV to autonomously select a flat area for a smooth vertical landing through optical flow analysis and laser ranging fusion.

[0123] In summary, the multi-light fusion target detection method based on a high-altitude airdropped coaxial UAV, according to an embodiment of the present invention, has the following beneficial effects: 1. This invention overcomes the bottleneck of flight safety in extreme high-altitude environments, achieving reliable attitude recovery. Addressing the problem of traditional rotor aerodynamic torque failure and the high risk of uncontrolled rollovers and crashes due to the extremely thin air at altitudes of 8000 meters, this invention innovatively proposes an attitude recovery method that actively stimulates gyro precession torque using the rotational speed difference of coaxial dual rotors. Through an optimal gyro torque allocation algorithm based on least-squares projection, the two-dimensional attitude recovery requirement is adaptively mapped to one-dimensional rotational speed difference control, supplemented by residual aerodynamic compensation. This allows the UAV to quickly recover to a stable descent attitude during freefall by utilizing the physical rotational mass of the rotors, completely solving the pain point of uncontrollable flight after deployment at extremely high altitudes. 2. This invention constructs a cross-spatial multimodal adaptive fusion detection system, significantly improving target recognition capabilities in complex environments. Addressing the inherent limitations of single or dual-light sensors in cloud and fog penetration, anti-camouflage, and all-weather detection, this invention establishes a dynamic confidence fusion strategy based on altitude perception. In high-altitude clouds, the Beer-Lambert law transmission model of short-wave infrared light is used to penetrate clouds and fog. In the mid-altitude region, a hyperspectral camera and its spectral angle mapping algorithm are introduced to accurately identify strongly camouflaged targets, such as those with infrared camouflage, through material spectral fingerprinting. In low-altitude regions, a visible light camera takes over for fine texture confirmation. The three-stage sensor system seamlessly switches adaptively with altitude, achieving all-altitude, all-weather, and highly reliable target detection from tens of thousands of meters above the ground. 3. This invention achieves intelligent prediction and dynamic positioning tracking, ensuring safe approach and autonomous recovery. Addressing the command lag and tracking collision risks inherent in traditional proportional control, this invention employs exponential moving average filtering to smoothly predict target motion and innovatively generates the desired accompanying tracking position dynamically above and behind the target. It combines flight control with camera gimbal arctangent closed-loop calculation to achieve close, smooth, and safe approach tracking of high-speed maneuvering non-cooperative targets. After mission completion, a local 3D elevation map is constructed by fusing downward visible light and laser ranging. The flattest area is automatically selected, and a vertical, compliant descent is executed, forming a complete intelligent closed loop from high-altitude deployment, detection, tracking to recovery.

[0124] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware, and the program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0125] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A multi-optical fusion target detection method based on a high-altitude airdropped coaxial UAV, characterized in that, include: The coaxial UAV detaches from the carrier aircraft at high altitude and performs adaptive attitude correction, enabling the coaxial UAV to adaptively correct itself to a stable descent attitude; During the stable descent of the coaxial drone, the confidence weights of the multiple optical sensors carried by the coaxial drone are dynamically adjusted based on the real-time absolute descent altitude. Based on the confidence weight, the multimodal feature data collected by each optical sensor are fused and judged to identify and continuously confirm the target during the continuous descent, and the three-dimensional spatial position of the target relative to the coaxial UAV is calculated in real time. Based on the real-time calculation of the target's three-dimensional spatial position relative to the coaxial UAV, the target's motion state is predicted, and the expected accompanying tracking position located above and behind the target is dynamically generated based on the prediction results. Flight control commands are generated based on the expected tracking position and the current position of the coaxial UAV to guide the coaxial UAV to continuously approach and track the target; when the detection mission is completed and the coaxial UAV's altitude drops to a preset landing threshold, it will land and be recovered.

2. The multi-optical fusion target detection method based on a high-altitude airdropped coaxial UAV according to claim 1, characterized in that, Adaptive attitude correction is performed by actively stimulating gyro precession effect based on the speed difference of coaxial dual rotors, including: Calculate the attitude error of a coaxial UAV; Based on the acquired roll rate, pitch rate, and yaw rate of the coaxial UAV, as well as the attitude error, the desired control torque to bring the coaxial UAV back to center is calculated. Based on the desired control torque, the UAV's physical rotational mass, roll rate, and pitch rate, the target value of the optimal rotational speed difference between the two rotors is obtained; Calculate the residual aerodynamic blade tilt angle based on the target value of the optimal speed difference between the two rotors; Based on the base sustaining speed, the target value of the optimal speed difference between the two rotors, and the residual aerodynamic blade tilt angle set by the coaxial UAV flight control system, the coaxial UAV return-to-center control command is generated.

3. The multi-optical fusion target detection method based on a high-altitude airdropped coaxial UAV according to claim 2, characterized in that, The desired control torque includes the desired roll torque. Desired pitch moment and desired yaw moment ; The target value for the optimal speed difference between the two rotors is as follows: in, The target value for the optimal speed difference between the two rotors. The roll rate is angular velocity. The pitch angular velocity, For the physical rotational mass of the drone, To prevent extremely small constants with a denominator of zero.

4. The multi-optical fusion target detection method based on a high-altitude airdropped coaxial UAV according to claim 2, characterized in that, The generated coaxial UAV homing control command includes: A first speed command is issued to the upper rotor drive motor of the coaxial UAV, and a second speed command is issued to the lower rotor drive motor of the coaxial UAV; wherein, the first speed command is the base maintenance speed plus half of the target value of the optimal speed difference between the two rotors; the second speed command is the base maintenance speed minus half of the target value of the optimal speed difference between the two rotors. Based on the residual aerodynamic blade tilt angle, a periodic pitch control command is issued to the tilting disk of the coaxial UAV.

5. The multi-optical fusion target detection method based on a high-altitude airdropped coaxial UAV according to claim 1, characterized in that, The multiple optical sensors include a short-wave infrared camera, a long-wave infrared camera, a hyperspectral camera, and a visible light camera; the optical axes of each optical sensor are parallel to each other and are arranged in an array symmetrically around the central vertical axis of the camera load in space; Based on the real-time absolute fall height, the confidence weights of multiple optical sensors are dynamically adjusted, including: when At the same time, the highest confidence level is assigned to both the short-wave infrared camera and the long-wave infrared camera; when At that time, the hyperspectral camera was assigned a dominant confidence level; when At that time, based on the logistic inverse function model, the confidence weight of the visible light camera jumps as the height decreases, until it takes over the task of confirming the fine texture of the target. in, This represents the absolute height of the fall in real time. These are the first, second, and third height thresholds, respectively. .

6. The multi-optical fusion target detection method based on a high-altitude airdropped coaxial UAV according to claim 5, characterized in that, Based on the aforementioned confidence weights, the multimodal feature data collected by each optical sensor are fused and judged to identify and continuously confirm the target during continuous descent. Simultaneously, the three-dimensional spatial position of the target relative to the coaxial UAV is calculated in real time, including: Image data acquired by short-wave infrared camera, long-wave infrared camera, hyperspectral camera and visible light camera are extracted respectively. Candidate targets in each image data are detected and identified, and the probability of target existence corresponding to each optical sensor is generated. Obtain the confidence weights of each optical sensor at the current altitude, multiply the target existence probability of each optical sensor by the corresponding confidence weight, sum them up, and calculate the fusion decision probability. When the fusion decision probability is greater than or equal to a preset decision threshold, the existence of the target is confirmed and continuously tracked and confirmed. From images of at least two optical sensors with known relative positions, the pixel coordinates of the same target matching point are extracted. Combined with the intrinsic and extrinsic parameter matrices of the optical sensors, the three-dimensional spatial position coordinates of the target relative to the coaxial UAV are calculated.

7. The multi-optical fusion target detection method based on a high-altitude airdropped coaxial UAV according to claim 6, characterized in that, Based on the real-time calculated 3D spatial position, the target's motion state is predicted, and the expected accompanying tracking position located above and behind the target is dynamically generated based on the prediction results, including: The instantaneous velocity vector of the target is obtained by performing differential calculations on the target's three-dimensional spatial position calculated at consecutive time points. The instantaneous velocity vector of the target is smoothed by an exponential moving average filter to generate a smoothed velocity vector of the target. Based on the smoothed velocity vector of the target, the motion state of the target is determined and the predicted motion direction of the target is determined; Based on the preset safe tracking distance and height offset vector, the desired accompanying tracking position located above and behind the target is dynamically generated on the reverse extension line of the target's predicted motion direction.

8. The multi-optical fusion target detection method based on a high-altitude airdropped coaxial UAV according to claim 7, characterized in that, The expected accompanying tracking position is as follows: in, To accompany the tracking position, The target's current position, The preset safe tracking distance, Let be the unit vector representing the predicted direction of motion of the target. This is the preset height offset vector.

9. The multi-optical fusion target detection method based on a high-altitude airdropped coaxial UAV according to any one of claims 1-8, characterized in that, Based on the desired tracking position and the current position of the coaxial UAV, flight control commands are generated to guide the coaxial UAV to continuously approach and track the target, including: Real-time calculation of the three-dimensional error vector between the desired tracking position and the current position of the coaxial UAV; Separate the horizontal and height components of the three-dimensional error vector; A proportional control law is used to convert the horizontal component into a horizontal speed command and the height component into a vertical speed command. The horizontal and vertical speed commands are sent to the UAV's flight control system to guide the UAV to the desired tracking position, thereby achieving continuous close-range tracking of the target.

10. The multi-optical fusion target detection method based on a high-altitude airdropped coaxial UAV according to claim 9, characterized in that, When the reconnaissance mission is completed and the coaxial UAV descends to a preset landing threshold, a landing and recovery process will be initiated, including: When the detection mission is completed and the coaxial UAV descends to the preset landing threshold, the downward-looking visible light camera is activated to collect ground images. Combined with the ground distance information obtained by the airborne laser ranging module, optical flow analysis is performed on the ground texture. Based on optical flow analysis and laser ranging information, a local three-dimensional elevation map of the area directly below the coaxial UAV is constructed. The surface flatness of the candidate landing points in the local three-dimensional elevation map is calculated, and the flat area with the smallest surface variance and no protruding obstacles is selected as the target landing point. Control the coaxial UAV to fly towards the target landing point, make the coaxial UAV descend vertically, and perform autonomous landing and recovery.

Images