Searching unmanned aerial vehicle capable of being rapidly folded and unfolded and unmanned aerial vehicle image processing method

Abstract

The invention relates to the technical field of unmanned aerial vehicle manufacturing, in particular to a search unmanned aerial vehicle capable of being rapidly folded and unfolded and an unmanned aerial vehicle image processing method.The unmanned aerial vehicle comprises a fuselage, wings and an empennage, and the fuselage carries unmanned aerial vehicle hardware equipment; the wings are symmetrically arranged on the left side and the right side of the middle-rear portion of the fuselage, each wing comprises a box-shaped main beam located at the maximum thickness position of the wing, a tung wood strip and a light wood strip are arranged on the front portion and the rear portion of each box-shaped main beam respectively, and end wing ribs are arranged on the left side and the right side of each box-shaped main beam respectively. The box type main beam, the paulownia wood strip, the light wood strip and the two end wing ribs jointly form a straight rectangular wing; motor cabins are arranged on the wings close to the fuselage; the empennage is connected with the motor cabin through a tail pipe, and the whole empennage is of an H-shaped plane frame structure and is folded and retracted above the fuselage from back to front. The folding and unfolding mechanism can be quickly folded and unfolded, is small in occupied space, and has good stability and relatively high aerodynamic efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicle manufacturing, and in particular to a search unmanned aerial vehicle that can be quickly folded and unfolded and an unmanned aerial vehicle image processing method. Background Art

[0002] Unmanned aircraft, or "drones" for short, are unmanned aircraft that are controlled by radio remote control equipment and self-contained program control devices. There is no cockpit on board, but it is equipped with autopilots, program control devices and other equipment. Personnel on the ground, on ships or at the remote control station of the mother aircraft use radar and other equipment to track, locate, remotely control, telemeter and transmit digital data to it. It can take off like a normal aircraft under radio remote control or be launched into the air with a booster rocket, or it can be carried into the air by the mother aircraft for flight.

[0003] When a drone is used for search missions, it carries a payload to perform the search mission, and its performance indicators must meet the mission requirements. Its payload is an image processing system that can intelligently search the target area and return the search results. However, current search drones have the defects of large size, large space, and difficulty in storage and transportation. In addition, due to inherent design defects, drones have poor flight stability and poor aerodynamic efficiency. Summary of the invention

[0004] The purpose of the present invention is to solve the problems existing in the prior art and to propose a search drone and a drone image processing method that can be quickly folded and unfolded. On the one hand, it can be quickly folded and unfolded and occupies a small space. On the other hand, through the ingenious design of the drone fuselage, wings and tail wing, it has good stability and high aerodynamic efficiency.

[0005] To achieve the above purpose, the technical solution adopted is:

[0006] A search drone that can be quickly folded and unfolded, the entire surface of the drone is covered with a layer of skin, including:

[0007] The fuselage, which carries the drone hardware;

[0008] The wings are symmetrically arranged on the left and right sides of the middle and rear part of the fuselage, the wings include a box-type main beam located at the maximum thickness of the wing, a paulownia wood strip and a balsa wood strip are respectively arranged at the front and rear sides of the box-type main beam, and end wing ribs are respectively arranged on the left and right sides of the box-type main beam, and the box-type main beam, the paulownia wood strip, the balsa wood strip and the two end wing ribs together form a straight rectangular wing; an electric motor cabin is arranged on the wing near the fuselage;

[0009] The tail wing is connected to the motor compartment through the tail pipe. The entire tail wing is an H-shaped plane frame structure and is folded from the back to the front and retracted above the fuselage.

[0010] According to the search drone that can be quickly folded and unfolded of the present invention, further, the fuselage includes a streamlined hood located at the front end and a fuselage wall panel connected to the hood with a buckle, two upper accommodating cavities and two lower accommodating cavities are formed in the fuselage wall panel, and both accommodating cavities are matched with hatches, and a downward window is opened at the front end of the hood; the hood is made of composite materials, and the fuselage wall panel is made of laminated boards.

[0011] According to the search drone that can be quickly folded and unfolded according to the present invention, further, the drone hardware equipment includes a flight control module, a positioning module, a communication module, a processing module, a vision module, a power module and an airspeed meter. The processing module remotely communicates with the ground station computer through the communication module; the communication module and the vision module are built into the hood, and the camera of the vision module extends outward from the window; the power module and the flight control module are respectively embedded in two upper accommodating cavities of the fuselage wall panel, the processing module and the airspeed meter are embedded in a lower accommodating cavity of the fuselage wall panel, and the positioning module is installed on the fuselage wall panel between the two upper accommodating cavities.

[0012] According to the search drone that can be quickly folded and unfolded according to the present invention, further, the wing adopts a CLARK-Y airfoil; a plurality of internal wing ribs are arranged through the box-type main beam between the paulownia wood strips and the balsa wood strips, and the outer wing ribs are spaced wider and gradually narrow inward; weight-reducing holes are opened on the end wing ribs and the inner wing ribs; the end wing ribs are made of plywood, and the inner wing ribs are made of balsa wood; the upper and lower plates of the box-type main beam are made of plywood, and the front and rear walls are made of balsa wood; the motor cabin is a rectangular box composed of plywood, with a motor arranged inside, the output shaft of the motor is connected to the propeller, and the propeller is located on the outer front side of the motor cabin.

[0013] According to the search drone that can be quickly folded and unfolded of the present invention, further, the balsa wood strips behind the box-type main beam are hingedly provided with ailerons, and the ailerons are made of balsa wood; a first rudder foot is provided at the bottom of the aileron, and a roll servo is provided at the bottom of the two internal wing ribs, and the roll servo is connected to the first rudder foot by a steel wire, and the first rudder foot is made of a plywood and glass fiber composite material.

[0014] According to the search drone that can be quickly folded and unfolded of the present invention, further, the tail end of the motor cabin is connected to the flip box, the upper end of the flip box is hinged to the motor cabin, and the lower end is connected to the motor cabin by a connecting buckle, and the flip box is made of laminate; the tail pipe is inserted into the interior of the flip box, and the tail pipe is made of a carbon fiber tube.

[0015] According to the search drone that can be quickly folded and unfolded of the present invention, further, the H-shaped tail includes a horizontal tail assembly and a vertical tail assembly vertically arranged on the left and right sides of the horizontal tail assembly; an elevator is hingedly arranged on the rear side of the horizontal tail assembly, a second rudder foot is arranged above the elevator, a pitch servo is arranged above the horizontal tail assembly, and the pitch servo is connected to the second rudder foot through a steel wire; a rudder is hingedly arranged on the rear side of the vertical tail assembly, a third rudder foot is arranged on the outside of the rudder, a yaw servo is arranged on the outside of the vertical tail assembly, and the yaw servo is connected to the third rudder foot through a steel wire; the second rudder foot and the third rudder foot are made of plywood and glass fiber composite materials; the horizontal tail assembly, the vertical tail assembly, the elevator and the rudder are all made of balsa wood.

[0016] According to the rapidly foldable and unfolded search drone of the present invention, further, the upper retraction angle of the tail wing is 11°, and the lower retraction angle is 25°; the cruising angle of attack of the search drone is 4°.

[0017] Furthermore, the present invention also proposes a method for processing drone images, which is implemented by using the above-mentioned fast folding and unfolding search drone, and the method comprises:

[0018] The image is collected through the visual module, and the interface of the processing module reads the image in real time and performs the following image processing;

[0019] Use the YOLOv8n model for target detection and select the target area;

[0020] Use template matching or residual neural network to identify the target.

[0021] According to the drone image processing method of the present invention, further, the specific process of template matching is: first prepare a number of templates, use a matching algorithm to calculate the pixel differences of the templates one by one each time the target is recognized, and the target category can be determined by sorting the size of pixel similarity, and the normalized correlation matching method is used for the pixel similarity score; the residual neural network adopts the ResNet50 model.

[0022] The beneficial effects achieved by adopting the above technical solution are:

[0023] 1. When the search drone needs to be stored after use, you only need to open the connection buckle between the flip box and the bottom of the motor cabin, and fold the tail wing, tail pipe and flip box from back to front through the hinge. The wings are also detachable structures, which effectively solves the problem of large size and large space occupied by the drone, and realizes the rapid folding of the drone for easy transportation.

[0024] 2. The main structure of the search drone of the present invention adopts a light wood design, the overall structure is light, the structural strength is sufficient, the manufacturing cost is low, and the processing technology is simple. The whole machine has a high degree of modularization, all parts can be quickly disassembled and replaced, and the complexity of maintenance is low.

[0025] 3. The tail wing adopts an H-shaped layout, which simplifies the analysis and design process, reduces the surface area, thereby reducing flight resistance, is light in weight and easy to manufacture.

[0026] The drone hood is designed to be streamlined to reduce shape resistance and airflow separation.

[0027] The wing is a straight rectangular wing, which is characterized by the consistent airfoil along the span direction, which is significantly simple to manufacture. In addition, the design provides excellent take-off and landing capabilities, maneuverability, and high lift-to-drag ratio and range efficiency. The wing adopts the CLARK-Y airfoil, which shows a large lift coefficient and a high lift-to-drag ratio. It performs well in generating lift and maintaining low drag, achieving flight economy and efficiency. The maximum lift-to-drag ratio of the CLARK-Y airfoil is relatively large and the curve is smooth, indicating that it can maintain good performance under various flight conditions. The CLARK-Y airfoil has a good angle of attack range, which allows it to maintain stable lift even at a large angle of attack and is not prone to stalling.

[0028] When the drone is subject to external interference, it can quickly return to its original balance state and has good stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings of the embodiments of the present invention, wherein the drawings are only used to illustrate some embodiments of the present invention, but not to limit all embodiments of the present invention thereto.

[0030] Figure 1 This is one of the structural schematic diagrams of the rapidly foldable and unfoldable search drone according to the first embodiment of the present invention;

[0031] Figure 2 This is a second structural schematic diagram of the rapidly foldable and unfoldable search drone according to the first embodiment of the present invention;

[0032] Figure 3 This is the third structural schematic diagram of the rapidly foldable and unfoldable search drone according to the first embodiment of the present invention;

[0033] Figure 4 is a side view of a search drone that is rapidly folded and unfolded according to the first embodiment of the present invention;

[0034] Figure 5 is a curve diagram of lift coefficient, drag coefficient and lift-to-drag ratio coefficient of five airfoils in Example 1 of the present invention;

[0035] Figure 6 is a graph of pitching moment coefficients of five airfoils according to the first embodiment of the present invention;

[0036] Figure 7 is a wing pressure distribution and streamline diagram of the UAV according to the first embodiment of the present invention;

[0037] Figure 8 is a wing spanwise load distribution diagram of the first embodiment of the present invention;

[0038] Fig. 9 is a two-dimensional airfoil force distribution diagram of embodiment 1 of the present invention;

[0039] Fig.10 It is the YOLOv8 model network structure of the second embodiment of the present invention;

[0040] Fig.11 is a dataset image of Embodiment 2 of the present invention;

[0041] Fig.12 is a flow chart of the transfer learning method of the second embodiment of the present invention;

[0042] Fig.13 is the target detection result of Example 2 of the present invention;

[0043] Fig.14 is the ROI area for incoming digital recognition in the second embodiment of the present invention;

[0044] Fig.15 This is the perspective transformation principle under the specified mapping of the second embodiment of the present invention;

[0045] Fig.16 This is the template matching process of the second embodiment of the present invention;

[0046] Fig.17 This is the ResNet50 network structure of the second embodiment of the present invention.

[0047] Figures 1 to 4 The meanings of the serial numbers are:

[0048] 101. head cover, 102. fuselage wall panel, 103. upper accommodating cavity, 104. lower accommodating cavity, 105. window;

[0049] 201. Box-type main beam, 202. Paulownia strip, 203. Light wood strip, 204. End rib, 205. Motor cabin, 206. Internal rib, 207. Lightening hole, 208. Motor, 209. Aileron, 210. First rudder foot, 211. Rolling servo, 212. Turning box;

[0050] 301. Tail pipe, 302. Horizontal tail assembly, 303. Vertical tail assembly, 304. Elevator, 305. Second rudder foot, 306. Pitch servo, 307. Rudder, 308. Third rudder foot, 309. Yaw servo. DETAILED DESCRIPTION

[0051] The following will be combined with the drawings of specific embodiments of the present invention to clearly and completely describe the exemplary scheme of the embodiment of the present invention. Unless otherwise defined, the technical terms or scientific terms used in the present invention should be the common meanings understood by people with ordinary skills in the field.

[0052] Embodiment 1

[0053] like Figures 1 to 4 As shown, this embodiment discloses a search drone that can be quickly folded and unfolded. The entire surface of the drone is covered with a layer of heat shrinkable film skin, which makes it beautiful in appearance and waterproof and moisture-proof. The search drone mainly includes three parts: a fuselage, wings and tail wing; the fuselage is equipped with drone hardware equipment. There are two wings, which are symmetrically arranged on the left and right sides of the middle and rear part of the fuselage. A motor compartment is provided on each wing close to the fuselage. The tail is connected to the motor compartment through a tail pipe. The entire tail is an H-shaped plane frame structure and can be folded from the back to the front and retracted above the fuselage, occupying a small area and convenient for transportation. The tail is arranged in an H-shape. This layout simplifies the analysis and design process, reduces the surface area, thereby reducing flight resistance, is light in weight, and is easy to manufacture.

[0054] The fuselage includes a streamlined hood located at the front end and a fuselage wall panel connected to the hood with a buckle. The streamlined hood can reduce shape resistance and airflow separation; a downward window is provided at the front end of the hood. The hood can be removed from the fuselage wall panel to facilitate the maintenance and replacement of the communication module and the visual module. The fuselage wall panel is a rectangular box body, and two upper accommodating chambers and two lower accommodating chambers are formed in the fuselage wall panel. Both upper accommodating chambers are matched with hatches, which is convenient for opening the hatches to repair and replace the internal equipment. One lower accommodating chamber is a spare compartment, and the other lower accommodating chamber is a closed structure. An installation hole for the positioning module is reserved between the two upper accommodating chambers. Preferably, the hood is made of composite material, the fuselage wall panel is made of layered panels, and the fuselage wall panel is a hollow structure to reduce the weight of the whole machine.

[0055] Furthermore, the hardware equipment of the drone includes a flight control module, a positioning module, a communication module, a processing module, a vision module, a power module and an airspeed meter. The processing module communicates remotely with the ground station computer through the communication module to transmit images and data, thereby realizing communication between the ground and the aerial drone equipment. The communication module and the vision module are built into the hood, and the camera of the vision module extends outward from the window to ensure that it has a good shooting field of view. The power module and the flight control module are respectively embedded in the two upper accommodating cavities of the fuselage wall panel; the processing module and the airspeed meter are embedded in a lower accommodating cavity of the fuselage wall panel, and the positioning module is installed on the mounting hole reserved in the fuselage wall panel. In order to meet the performance indicators of the drone, the hardware selection is carried out under the guidance of these indicators, and some of the hardware selections are as follows.

[0056] Flight control module: Hexing Cube Orange+ flight control, using STM32H757 Dual Core M7+M4 processor, supports SBUS remote control signal. Equipped with ICM42688IMU, MS5611 barometer sensor. Supports Ardupilot (4.2.3 or above) and PX4 firmware. With 14 PWM IO, four GPS or serial port interfaces, two Debug interfaces, two I2C interfaces, and two CAN FD interfaces. Can work with JETSON TX1, Raspberry Pi, etc. The ground station supports Mission Planner and QGroundControl. High degree of freedom, mature firmware, conducive to secondary development and aircraft parameter adjustment.

[0057] Positioning module: Here3+ positioning module adopts high-precision GNSS chip, cooperates with PTK base station, supports real-time dynamic differential positioning technology, positioning accuracy can reach 0.025 meters, and navigation refresh rate can reach 8HZ.

[0058] Communication module: Mini Homer image and data transmission link, transmission distance in open areas can reach 1200 meters; serial port supports TCP server, TCP client, UDP mode, UDP multicast, data transmission delay as low as 30 milliseconds; standard IEEE802.11ah protocol, maximum transmission bandwidth 11M; frequency band is Sub1G band. The ground end interface is network port: RJ45 serial port: GH1.25 5P. The sky end interface is network port 2: GH1.25 4P, serial port 2: GH1.25 3P, supports up to 8 networking modes, and the performance meets the needs.

[0059] Processing module: Allspark-Orin NX AI edge computer. It has a built-in NVIDIA Jetson Orin NX module, providing up to 100TOPS floating-point computing power. Allspark-Orin NX has a built-in 128GB SSD solid-state drive and a 1024-core NVIDIA Ampere architecture GPU with 32 Tensor Cores, 8 cores v8.2 64-bit CPU 2MB L2 + 4MB L3 CPU, supports 3S (11.1 volts) to 6S (22.2 volts) power supply.

[0060] Vision module: A8mini gimbal camera from Siyi Technology, equipped with 1 / 1.7-inch Sony image sensor, fixed focus, 8 million effective pixels. Video output signal interface is Ethernet port, HDMI, CVBS (AV) control signal input mode is S.Bus, UART, network port UDP, output mode is S.Bus, supports Mavlink protocol and direct communication with Ardupilot flight control. Flight control reads gimbal attitude, direction, pitch, and roll parameters through the serial port, with an accuracy of up to 0.1°.

[0061] Considering the endurance and motor output power, the battery used is 4s 2600mAh 25c lithium-ion polymer battery. The actual thrust-to-weight ratio of the search drone is 1.192, and the wing load is calculated to be 53g / dm 2 . Through experiments, the battery power can meet the needs.

[0062] The wing includes a box-type main beam located at the maximum thickness of the wing, that is, near 25% of the chord length, which is easy to manufacture. A paulownia wood strip and a balsa wood strip are respectively arranged at the front and rear of the box-type main beam, and end wing ribs are respectively arranged on the left and right sides of the box-type main beam; the box-type main beam, the paulownia wood strip, the balsa wood strip and the two end wing ribs together form a straight rectangular wing. In this embodiment, the motor cabin divides the wing into two parts, the wing span between the fuselage wall panel and the motor cabin is shorter, and the end wing ribs are fixed to the fuselage wall panel and the motor cabin respectively by bolts; the wing part outside the motor cabin has a longer span, and the end wing ribs are installed on the motor cabin through a quick-release structure, which is convenient for quick installation and removal of the outer wing; the quick-release structure can be made of PLA material, and the quick-release structure includes but is not limited to the form of inserting the end wing rib into the slot of the fuselage wall panel and quickly locking it with a lock buckle.

[0063] Depend on Fig. 9 It can be seen that when the drone flies at an angle of attack of 4°, the pressure distribution diagram shows that the leading edge of the wing is subject to the greatest force. The leading edge of the wing is subject to the greatest force when flying against the wind, so polished Paulownia wood strips are used to make the leading edge to enhance the strength of the leading edge so that the wing will not be deformed by force.

[0064] The width of the box-type main beam is 16mm. As the aircraft needs to be more maneuverable, the maximum overload of 4G is taken according to experience. It can be calculated that the load concentration on the wing is about 47N / m; the bending moment of the dangerous section is about 39N·m. The beam thickness d = 2mm, the beam height h = 31mm, to simplify the calculation, the effect of the beam thickness on the stress distribution can be ignored. At this time, the maximum positive stress of the dangerous section is about 38MPa, which is lower than the stress limit of the laminate, indicating that the design scheme is feasible in theory.

[0065] Preferably, the wing adopts a CLARK-Y airfoil. In addition to the end ribs, a plurality of internal ribs are provided through the box-type main beam between the paulownia strips and the light wood strips. The wing torque increases from the outside to the inside, so the outer ribs are spaced wider during the arrangement of the ribs and gradually narrow inward. In order to reduce the weight, weight-reducing holes are provided on the end ribs and the internal ribs, which can reduce the total weight by about 30%. The ribs are mainly made of 2mm plywood and light wood, the end ribs use reinforced 2mm plywood to increase the strength, and the internal ribs are selected for light wood. The box-type main beam adopts a composite structure, the upper and lower plates use 2mm plywood, the front and rear walls use 2mm light wood, and the mortise and tenon structure is used between the components and glued with 502 glue. The failure form of the box beam is often the instability of the edge strip, so Kevlar wire is used to wrap the beam to form a whole, and the middle section is reinforced with carbon tubes to ensure the structural strength of the wing main beam. The design has strong bearing capacity and is torsion-resistant. The motor cabin is a rectangular box composed of layers, with a motor inside. The output shaft of the motor is connected to the propeller, which is located on the front side of the motor cabin. The motor uses Sunnysky's 2216 1100kv with APC1047 propeller. The pulling force of this propeller and motor can reach 1200g.

[0066] The light wood strips at the rear of the box-type main beam are hinged with ailerons. Ailerons are one of the main control surfaces used by the aircraft to control the roll movement of the fuselage. They appear in pairs and are made of light wood. The first rudder foot is set at the bottom of the aileron, and a roll servo is set between the bottoms of the two internal ribs (the ribs use reinforced ribs to support the roll servo). The roll servo is connected to the first rudder foot through steel wires. The first rudder foot is made of plywood and fiberglass composite materials. The output end of the roll servo is connected to the rocker arm, and the rocker arm is connected to the first rudder foot through steel wires. The rotation of the roll servo drives the rocker arm to rotate, which in turn drives the aileron to deflect and change the roll attitude of the aircraft. Under normal circumstances, the aileron area is 0.05 a / s<0.07, take s a / s=0.06, s represents the wing area; the total area of ​​the aileron is s a =2.3dm 2 Therefore, the aileron span is taken as 0.571m and the aileron chord length is taken as 0.042m.

[0067] ​The tail end of the motor cabin is connected to the flip box, the upper end of the flip box is hinged to the motor cabin through a hinge, and the lower end is connected to the motor cabin through a connecting buckle. The flip box is made of plywood. The tail pipe is inserted into the flip box and can be fixed by tightening bolts. The tail pipe is made of a carbon fiber tube.

[0068] The tail provides a small part of the lift for the aircraft, generates torque, and plays a balancing role. The H-type tail includes a horizontal tail assembly and a vertical tail assembly vertically arranged on the left and right sides of the horizontal tail assembly, and the two can be fixed by bonding. The horizontal tail assembly is arranged below the vertical tail assembly, which effectively enhances the longitudinal and lateral stability and handling performance of the aircraft. This configuration helps to improve heading control and provide good stability. The tail is not subjected to much force and the strength requirement is not high. In order to reduce the structural weight, a flat frame structure is adopted instead of a three-dimensional wing structure, which reduces the difficulty of production and reduces the working hours, saving costs. The bending moment at the root of the vertical tail is reduced by lowering the height of the vertical tail. The bending moment at the root of the vertical tail is related to the distance between the vertical tail and the center of gravity of the aircraft; lowering the height of the vertical tail can reduce the torque generated by external forces (such as crosswinds) acting on the vertical tail, thereby reducing the stress on the structure and reducing the strength requirements for the fixed part of the vertical tail.

[0069] The rear side of the horizontal tail assembly is hinged with an elevator, a second rudder foot is arranged above the elevator, a pitch servo is arranged above the horizontal tail assembly, and the pitch servo is connected to the second rudder foot through a steel wire. The rear side of the vertical tail assembly is hinged with a rudder, a third rudder foot is arranged outside the rudder, a yaw servo is arranged outside the vertical tail assembly, and the yaw servo is connected to the third rudder foot through a steel wire. The second rudder foot and the third rudder foot are both made of laminate and glass fiber composite materials. The horizontal tail assembly, the vertical tail assembly, the elevator and the rudder are all made of balsa wood. The control structure of the rudder surface is the same as the aileron above. The pitch servo controls the elevator to change the pitch attitude of the aircraft, and the yaw servo controls the rudder to change the yaw attitude of the aircraft.

[0070] Design the vertical tail area according to the tail capacity Among them C VT , b w , S w , L VT They represent the vertical tail capacity, the average aerodynamic chord length of the wing, the wing area, the tail arm of the vertical tail, and the horizontal tail area. C HT , b w , S w , L HT s respectively represent the tail capacity of the horizontal tail, the average aerodynamic chord length of the wing, the wing area and the tail arm of the horizontal tail, and the elevator control surface area s e =0.014m 2 , rudder surface area s r =0.01m 2 .

[0071] Preferably, a connecting and fixing box is provided at the bend where the horizontal tail assembly and the vertical tail assembly are connected. The connecting and fixing box is made of a layer plate. The tail pipe is inserted into the fixing box and fixed with a tightening bolt. By loosening the tightening bolt, the tail wing can be easily removed from the fuselage.

[0072] In order to reduce the pressure difference drag, the upper retraction angle of the UAV's tail wing is 11°, and the pressure difference drag is significantly reduced. In order to reduce the possibility of the tail rubbing the ground, the lower retraction angle is 25°.

[0073] This embodiment uses a straight rectangular wing. Although this design is not optimal in terms of aerodynamic efficiency because the wing tip has limited contribution to lift and cannot fully share the weight load of the drone, it has significant simplicity in manufacturing. The drone flies in a low-speed flight environment, and the XFLR5 software simulation verifies that the rectangular wing has similar performance to the trapezoidal wing under this condition. In addition, the design provides excellent take-off and landing capabilities, maneuverability, high lift-to-drag ratio and range efficiency, so a straight rectangular wing is used.

[0074] Take air density ρ = 1.225Kg / m 3 , viscosity coefficient u=1.79×10 -5 N·s / m 3 , take the cruising speed v = 12m / s, characteristic length D = 0.258m, calculate the Reynolds number Then at the Reynolds number R e =172961, the airfoil is selected. Since the airfoil directly affects the aerodynamic characteristics and efficiency of the UAV, various characteristics of the airfoil must be considered comprehensively to ensure that the UAV has excellent flight performance. By comparing the airfoils in the Profili software airfoil library, five typical airfoils Clark Y, S3024, NACA4412, Benedek 9404B and MH 114 are selected for comparison. Reynolds number R e = 172961, it is necessary to use Profili software to conduct a detailed analysis of different airfoils. By comparing five different airfoils (see Figure 5 ), the Clark Y airfoil shows a large lift coefficient and a high lift-to-drag ratio, which means that it performs well in generating lift and maintaining low drag, achieving flight economy and efficiency. The Clark Y airfoil also has a relatively large maximum lift-to-drag ratio and a smooth curve, indicating that it can maintain good performance under various flight conditions. The S3024 and NACA4412 lift-to-drag ratios are slightly greater than the Clark Y at small angles of attack, but the lift-to-drag ratio drops rapidly at large angles of attack, and the stability is poor, making it easy to stall due to boundary layer separation caused by increased angles of attack. The Clark Y airfoil has a better angle of attack range, which allows it to maintain stable lift even at larger angles of attack and is less likely to stall. Figure 6As shown in the figure, although the lift-to-drag ratio of the MH 114 airfoil is better than that of the Clark Y, it requires more balancing torque due to its large pitch moment coefficient. This feature may lead to a disadvantage in the control performance of the UAV and make it more difficult to change the pitch attitude of the UAV. The Clark Y airfoil has a smaller pitch moment, which can provide better maneuverability and easy balancing. These features make the Clark Y airfoil closer to the performance requirements of the aircraft.

[0075] Aerodynamic analysis using XFLR5 aerodynamic simulation analysis software shows that under the conditions of 12m / s incoming flow speed and 4 degrees angle of attack, the UAV Figure 7 As shown in the figure, the wing pressure distribution and streamlines of the drone show good aerodynamic performance. The low-pressure area of ​​the wing is very small, indicating that the aircraft has low lift loss. At the same time, the pressure distribution on the upper and lower surfaces of the wing is relatively uniform, which helps to maintain the stability of the aircraft and improve the range efficiency. The uniformity and continuity of the flow field shows that the airflow flows well around the leading edge of the wing, further confirming the superiority of the airfoil design, which is conducive to ensuring flight stability and improving maneuverability. There are obvious wingtip vortices at the wing tip of the straight wing, and the turbulence generated has little effect on the overall wing.

[0076] Figure 8 The load distribution of the wing along the span direction. The lift of a single wing gradually increases from the wing tip to the wing root, presenting a distribution pattern that is approximately trapezoidal. This indicates that the structure at the wing root needs to bear a greater load, while the load at the wing tip is lighter. This distribution can reduce the strength of the wing tip vortex, thereby improving the overall aerodynamic efficiency and reducing drag.

[0077] Through the curve of the lift coefficient (Cl), it can be seen that the lift coefficient of the drone increases with the increase of the angle of attack, which is in line with the principle of aerodynamics. The angle of attack is about -5° at the zero lift point, which means that when the angle of attack is lower than this value, the drone will begin to generate downward lift (i.e., descending force). The stall angle of attack is when the lift coefficient suddenly drops after reaching its maximum value and the drone loses lift. Because the accuracy of XFLR5 in calculating the high angle of attack state is limited. Therefore, it is difficult to obtain an accurate stall angle of attack. The stall angle of attack is when the lift coefficient suddenly drops after reaching its maximum value and the drone loses lift. Through the pitch moment characteristic curve, the pitch moment coefficient (Cm) of the drone decreases significantly with the increase of the angle of attack. This means that when the drone is subjected to external disturbances that increase the angle of attack, the drone will generate a natural restoring torque that helps it quickly return to the original equilibrium state. Therefore, this shows that the drone design has good longitudinal stability. Only when the angle of attack is about 1°, the pitch moment of the drone is close to zero, which further proves that the drone has excellent stability characteristics. The lift-to-drag ratio of a drone increases with the increase of the angle of attack. When it reaches the peak value, it means that the drone has the best aerodynamic efficiency in this state. The lift-to-drag ratio is the largest at an angle of attack of 4°, which means that this angle is ideal for cruising, providing the maximum lift-to-drag ratio, and is also the preferred cruising state in design. Therefore, the drone is designed to cruise at an angle of attack of 4°, at which the lift can reach the maximum.

[0078] The UAV is subjected to dynamic simulation analysis to evaluate its response and stability when subjected to external disturbances, including dynamic characteristics such as pitch, roll and yaw. Through simulation analysis, the natural frequency, damping ratio and various flight modes of the UAV can be identified. The short-period and roll damping modes of this UAV will disappear quickly due to large damping, while the long-period and Dutch roll damping modes are more obvious. The whole machine recovers quickly after being disturbed, and has good stability performance.

[0079] The parameters of each part of the UAV are shown in the table. Table 1 lists the size parameters of each part of the UAV, Table 2 lists the mass parameters of each part of the UAV, and the materials used in the structure of each part are shown in Tables 3, 4, and 5.

[0080] Table 1 UAV size parameters

[0081]

[0082] Table 2 UAV quality parameters

[0083]

[0084] Table 3 Wing structural materials

[0085]

[0086] Table 4 Fuselage structural materials

[0087]

[0088] Table 5 Tail group structural materials

[0089]

[0090] Embodiment 2

[0091] This embodiment provides a method for processing drone images, which is implemented by using the search drone in the first embodiment. The method includes:

[0092] Step S1: Image acquisition is performed through the visual module, and the interface of the processing module reads the image in real time and performs the following image processing.

[0093] Step S2: Use the YOLOv8n model to detect the target and select the target area.

[0094] Step S3: Identify the target using template matching or residual neural network.

[0095] The method is described in detail below by taking the identification of numbers on a target as an example.

[0096] The high-precision gimbal camera installed on the drone is used to collect images. The images are read in real time through the USB interface of the onboard computer. In addition, the gimbal's attitude parameters, yaw, pitch and roll, are read through serial communication. The onboard computer uses the above data for target detection and digital recognition.

[0097] Due to the problems of small target size, relatively blurry, uncertain direction, complex background, etc. in drone aerial images, and the hardware conditions of the drone flight platform limit the computing power of the onboard computer, it is not possible to use a model with a large number of parameters for detection. In order to balance the detection performance and computing power limitations, the version with the smallest number of parameters among several generations of YOLO series models was tested, and its performance comparison is shown in the following table:

[0098] Table 6 Performance comparison of some versions of YOLO series

[0099]

[0100] YOLOv8n is the lightest version, suitable for running in environments with limited hardware resources, such as mobile devices or edge computing devices. Despite its small number of parameters, it is still designed to maintain an acceptable level of accuracy, making it a fast real-time object detection solution.

[0101] The architecture of YOLOv8n consists of three parts (such as Fig.10As shown in the figure): backbone network, neck and detection head. The backbone network, i.e., the feature extraction module, uses a modified Darknet-53, which is constructed through convolutional layers and residual connections, and adds a C2f module. This module combines the concept of CSP (Cross Stage Partial) and the design principle of ELAN (Efficient Layer Aggregation Network) to improve gradient flow and path diversity, enhance the learning efficiency of the network and the integration of feature information. C2f optimizes the transfer of features and gradients through Split operations and dense skip-layer connections, alleviates the gradient vanishing problem, and enhances the capture ability of the feature layer. In addition, this module overlaps the convolutional layer and the C2f module four times to form four stages, each of which is followed by an SPPF (Spatial Pyramid Pooling Fusion) module similar to that in YOLOv5 to fix the size of the feature map.

[0102] The neck is responsible for feature fusion, using the C2f module to replace the traditional C3 module, and using the PAN (Path Aggregation Network) and FPN (Feature Pyramid Network) structures to achieve top-down and bottom-up fusion of features. In addition, unlike YOLOv5 and YOLOv6, YOLOv8 does not use 1×1 convolution before upsampling, but directly upsamples the features of different stages output by the backbone network.

[0103] In the head, YOLOv8n has been significantly modified from the Anchor-based coupling head to the Anchor-Free decoupling head. The new detection head removes the objectness branch and only includes a decoupled classification branch and a regression branch using DFL (Distribution Focal Loss). The detection head includes three detection heads that process feature maps of different sizes, specifically for identifying targets of various sizes.

[0104] like Fig.11 As shown in the figure, in order to train the network model, one thousand target images were taken from different backgrounds, different heights and different perspectives, and the target areas were marked. In order to make the model have better generalization performance, data enhancement was performed by cropping, scaling and flipping the original images, expanding the range of the data set, and thus improving the model training effect. The proportion of the target in the image is less than 10%, which meets the small target standard.

[0105] When conducting machine learning or deep learning experiments, correctly dividing the data set is a critical step, which can ensure that the model can be effectively trained and evaluated on different data. In this experiment, the images used were divided into training sets, validation sets, and test sets in a ratio of 7:2:1. In order to increase the training speed, the neural network uses the transfer learning method, such as Fig.12 As shown, the pre-trained weights YOLOv8n.pt are used to fine-tune the network in combination with a homemade dataset.

[0106] Network training on the NVIDIA Quadro GP100 graphics card can give full play to the performance of its 3584 CUDA cores and 16GB HBM2 high-bandwidth memory, up to 717GB / s memory bandwidth, PCI Express 3.0PCIE bus, 5.2TFLOPS FP64 performance, 10.3TFLOPS FP32 performance, and 20.7TFLOPS FP16 performance. It can improve training efficiency and shorten model training time.

[0107] After 100 iterations of training, the loss curve of the training process tends to be flat. Continuing training has little effect and is prone to overfitting, so the training is terminated.

[0108] The model correctly selects the targets in the image, and the targets are evenly distributed and discrete in the image. The target size is small, which meets the requirements of small target detection. The target detection results are as follows: Fig.13 shown.

[0109] In the field of target detection, commonly used evaluation indicators for measuring the characteristics of the model include precision, recall, F1 score, mean average precision (mAP), and IoU (Intersection over Union). This embodiment uses AP50 and AP95 as evaluation indicators of the target detection algorithm, which represent the average value of Average Precision when the IoU value is 0.5 and 0.95.

[0110] Table 7. Target detection algorithm performance

[0111]

[0112] Cut out the region of interest ROI (such as Fig.14 as shown) to improve the efficiency and accuracy of the digital recognition algorithm.

[0113] After the target image is segmented, the digital recognition on the target is performed, and the digital extraction is performed as follows: Specifically, through perspective transformation, the quadrilateral digital area is mapped to a square area of ​​a given size.

[0114] First, fit the pentagonal outer contour and quadrilateral inner contour features of the target through the convex hull approximation algorithm, and then obtain the corner points of the inner and outer contours according to the fitted inner and outer contour shapes. And sort them in counterclockwise order. Calculate the average value of the outer ring as the center point coordinate, calculate the distance from each point on the outer ring to the center point, and find the point with the farthest distance as the head position. Then in the code, determine the starting point of the outer ring according to the head position head: if the head position head is 4, take the first point in the outer ring as the upper left corner of the outer ring. Otherwise, take the point after the head position head plus one as the upper left corner of the outer ring (outer_ring[head+1][0]). After determining the upper left corner of the outer ring, calculate the distance from each point on the inner ring to the upper left corner of the outer ring, and find the closest point as the upper left corner of the inner ring. Then sort the four points of the inner ring in counterclockwise order. Finally, perform perspective transformation through cv2.getPerspectiveTransform according to the above corner point order, and map the quadrilateral digital area to a 100×100 square area, such as Fig.15 As shown. The area is equally divided into two 50×100 rectangular areas on the left and right, and the adaptive threshold is used for binary segmentation, and the final two rectangular areas are the ones digit area and the tens digit area.

[0115] The first consideration for digital recognition is to use the template method. Since the format, shape and size of digital characters in the target are specified, it is not necessary to recognize the digital characters, but only to compare the similarity with the digital set. Therefore, the template matching algorithm in the open source computer vision library (OpenCV) is used to perform character comparison to achieve digital recognition.

[0116] The template matching algorithm requires ten digital templates to be prepared in advance. Each time a digital recognition is performed, the matching algorithm is used to calculate the pixel differences of the ten digital templates one by one. The digital can be judged by sorting the pixel similarity. There are 6 methods for calculating pixel similarity scores, such as square difference, correlation matching method, normalized square difference, correlation coefficient and normalized correlation matching method. The higher the algorithm complexity, the more accurate the matching result. Taking into account the limited computing power of the onboard computer, in order to improve the speed of digital recognition and balance the requirements of speed and accuracy, the normalized correlation matching method was selected after actual testing. Fig.16 The specific algorithm is shown in the following formula: T represents the template image, I represents the image to be matched, and R represents the matching result of the cut template. The calculation process satisfies the following relationship, and the matching process is as follows:

[0117]

[0118] The algorithm was tested on 100 actual samples and took 6.8ms to recognize a single digit with an accuracy of 78%. In actual application, multiple frames of the same digit were recognized multiple times, the number of recognition results was sorted, and the result with the most number of times was selected as the final result. At this time, the accuracy rate can be close to 100%. At the same time, in order to be compatible with more application scenarios, the ResNet50 model in the residual network is used to recognize and classify the digits 0 to 9 for numbers with uncertain character formats and different shapes and sizes.

[0119] Digital recognition can also use deep neural networks. This embodiment uses ResNet50 (see Fig.17 ), ResNet50 is a deep neural network with 49 convolutional layers and 1 fully connected layer. Its core feature is the use of ID BLOCK×2 and CONV BLOCK as residual blocks, which are used to maintain and change the data dimension respectively. The second to fifth stages of the network are all residual blocks consisting of three convolutional layers, and it is calculated that there are 49 convolutional layers in total. The model includes convolution operations (CONV), batch normalization (BatchNorm), activation functions (ReLU), and two pooling operations (MAXPOOL and AvgPOOL).

[0120] The preprocessing of the image before processing includes changing it to the 224×224×3 size required by the model. After preprocessing and input into the neural network, the number of image channels is gradually expanded and finally converted into a vector through the Flatten operation, which is the input of the fully connected layer. Finally, the softmax classifier uses the output of the fully connected layer to provide the predicted probabilities of different categories.

[0121] Self-made aerial images are used as the data set for digital recognition. A total of 5,000 images of numbers 0 to 9 are segmented from aerial images under different angles, heights, speeds, postures, lighting conditions, thresholds, etc., and a ten-class data set is produced. The training set and the validation set are divided into a ratio of 9:1, and then the model training is carried out.

[0122] In the experiment, the learning rate was set to a fixed value of 0.0001, the batch size was 10, the training set data was 4648, the validation set data was 511, and the ResNet50 model was trained for 5 epochs with an accuracy of 94%. After 3 epochs, the error on the test set was close to zero. This shows that the model has fast convergence speed, short training time, high recognition accuracy, and effectively suppresses overfitting.

[0123] The confusion matrix, as the name implies, can intuitively show which of the 10 categories of numbers in the test set are misclassified, that is, which categories the model is confused about, where each row represents the true category and each column represents the predicted category. The results show that most of the numbers are located on the main diagonal and the numbers are correctly identified. Number 3 is the number with the worst performance of the model, and it is mistakenly predicted as number 8 the most (50 times). This may mean that numbers 3 and 8 are similar in some features, which makes it easier for the model to confuse them. It is necessary to further analyze the data of these two numbers to find out their distinguishing features in order to improve the model. Except for number 3, the accuracy of most numbers is above 0.9, and the model performs relatively well on these numbers. The accuracy of numbers 4, 7, and 8 reaches 1.0, indicating that the model has a good ability to distinguish these numbers, probably because their features are more obvious.

[0124] Finally, it should be noted that the above-described embodiments are only specific implementations of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The protection scope of the present invention is not limited thereto. Although the present invention is described in detail with reference to the above-described embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-described embodiments within the technical scope disclosed by the present invention, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A search drone that can be quickly folded and unfolded, characterized in that: The entire drone surface is covered with a layer of skin, including: The fuselage, which carries the drone hardware; The wings are symmetrically arranged on the left and right sides of the middle and rear part of the fuselage, the wings include a box-type main beam located at the maximum thickness of the wing, a paulownia wood strip and a balsa wood strip are respectively arranged at the front and rear sides of the box-type main beam, and end wing ribs are respectively arranged on the left and right sides of the box-type main beam, and the box-type main beam, the paulownia wood strip, the balsa wood strip and the two end wing ribs together form a straight rectangular wing; an electric motor cabin is arranged on the wing near the fuselage; The tail wing is connected to the motor compartment through the tail pipe. The entire tail wing is an H-shaped plane frame structure and is folded from the back to the front and retracted above the fuselage.

2. The fast-folding and unfolding search drone according to claim 1, characterized in that: The fuselage includes a streamlined hood located at the front end and a fuselage wall panel connected to the hood with a buckle, two upper accommodating cavities and two lower accommodating cavities are formed in the fuselage wall panel, and both accommodating cavities are matched with cabin covers. A downward window is opened at the front end of the hood; the hood is made of composite materials, and the fuselage wall panel is made of laminated boards.

3. The fast-folding and unfolding search drone according to claim 2, characterized in that: The hardware equipment of the UAV includes a flight control module, a positioning module, a communication module, a processing module, a vision module, a power module and an airspeed meter. The processing module remotely communicates with the ground station computer through the communication module; the communication module and the vision module are built into the head cover, and the camera of the vision module extends outward from the window; the power module and the flight control module are respectively embedded in the two upper accommodating cavities of the fuselage wall panel, the processing module and the airspeed meter are embedded in a lower accommodating cavity of the fuselage wall panel, and the positioning module is installed on the fuselage wall panel between the two upper accommodating cavities.

4. The fast-folding and unfolding search drone according to claim 1, characterized in that: The wing adopts a CLARK-Y airfoil; a plurality of internal wing ribs are arranged through the box-type main beam between the paulownia wood strips and the balsa wood strips, and the outer wing ribs are spaced wider and gradually narrower inward; weight-reducing holes are opened on the end wing ribs and the inner wing ribs; the end wing ribs are made of plywood, and the inner wing ribs are made of balsa wood; the upper and lower plates of the box-type main beam are made of plywood, and the front and rear walls are made of balsa wood; The motor cabin is a rectangular box composed of layer plates, and a motor is arranged inside. The output shaft of the motor is connected to a propeller, and the propeller is located on the outer front side of the motor cabin.

5. The fast-folding and unfolding search drone according to claim 4, characterized in that: The balsa wood strips at the rear of the box-shaped main beam are hingedly provided with ailerons, and the ailerons are made of balsa wood; A first rudder foot is arranged at the bottom of the aileron, and a roll servo is arranged at the bottom of two internal wing ribs. The roll servo is connected to the first rudder foot through a steel wire, and the first rudder foot is made of a laminate and glass fiber composite material.

6. The fast-folding and unfolding search drone according to claim 1, characterized in that: The tail end of the motor cabin is connected to the flip box, the upper end of the flip box is hinged to the motor cabin, and the lower end is connected to the motor cabin through a connecting buckle. The flip box is made of plywood; the tail pipe is inserted into the flip box, and the tail pipe is made of a carbon fiber tube.

7. The fast-folding and unfolding search drone according to claim 1, characterized in that: The H-type tail comprises a horizontal tail assembly and a vertical tail assembly vertically arranged on the left and right sides of the horizontal tail assembly; an elevator is hingedly arranged on the rear side of the horizontal tail assembly, a second rudder foot is arranged above the elevator, a pitch servo is arranged above the horizontal tail assembly, and the pitch servo is connected to the second rudder foot through a steel wire; a rudder is hingedly arranged on the rear side of the vertical tail assembly, a third rudder foot is arranged on the outside of the rudder, a yaw servo is arranged on the outside of the vertical tail assembly, and the yaw servo is connected to the third rudder foot through a steel wire; the second rudder foot and the third rudder foot are made of plywood and glass fiber composite materials; the horizontal tail assembly, the vertical tail assembly, the elevator and the rudder are all made of balsa wood.

8. The fast-folding and unfolding search drone according to claim 1, characterized in that: The upper retraction angle of the tail wing is 11°, and the lower retraction angle is 25°; the cruising angle of attack of the search UAV is 4°.

9. A method for processing drone images, characterized in that: The method is implemented by using the foldable and unfoldable search drone as described in any one of claims 1 to 8, and the method comprises: The image is collected through the visual module, and the interface of the processing module reads the image in real time and performs the following image processing; Use the YOLOv8n model for target detection and select the target area; Use template matching or residual neural network to identify the target.

10. The drone image processing method according to claim 9, characterized in that: The specific process of template matching is as follows: first prepare several templates, use the matching algorithm to calculate the pixel differences of several templates one by one each time the target is recognized, and the target category can be determined by sorting the size of pixel similarity. The normalized correlation matching method is used for the pixel similarity score; the residual neural network adopts the ResNet50 model.

Images