Low-altitude unmanned aerial vehicle real-time aerial photography data fusion and safe transmission method based on state constraint

Abstract

This invention relates to the technical field and discloses a method for real-time aerial data fusion and secure transmission of low-altitude unmanned aerial vehicles (UAVs) based on state constraints. The method utilizes a 5G-A integrated sensing base station to simultaneously perform communication and radar sensing, extracting micro-Doppler features from the echo and determining the health level. Health characteristics such as UAV rotor speed fluctuations and body vibration are used as hard constraints. Downlink control information from the physical layer directly overwrites the UAV's uplink transmission parameters. The UAV's MAC layer uses a forced-mode gated video encoder output and layered encrypted transmission. Based on the UAV's state-aware forced constraint communication, secure communication transmission of UAV aerial data is achieved. Furthermore, multi-source spatiotemporal alignment and edge fusion at the base station edge nodes enable secure, real-time, and intelligent management of low-altitude data, improving the security, real-time performance, and reliability of low-altitude UAV data transmission.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) technology, and more specifically to a method for real-time aerial data fusion and secure transmission of low-altitude UAVs based on UAV operational state constraints. Background Technology

[0002] Low-altitude unmanned aerial vehicles (UAVs) are increasingly widely used in emergency command, land surveying, environmental monitoring, agricultural plant protection, and infrastructure inspection. Utilizing their onboard multi-source sensor fusion systems, they can achieve multi-dimensional data perception. Aerial data can be used for applications such as fire detection, vehicle counting, environmental monitoring, and water pollution assessment, playing a vital role. As a core information carrier, the real-time performance, security, and reliability of aerial data directly affect the effectiveness of mission execution. Especially in emergency scenarios such as disaster relief (earthquakes, floods, and forest fires), UAVs need to promptly transmit high-definition video, multispectral images, gas concentration, and other sensor data from the scene back to the command center to support rapid decision-making.

[0003] Currently, the transmission and processing of aerial photography data from low-altitude drones typically leverages the advantages of wide coverage and high bandwidth of cellular networks, allowing drones to connect to ground base stations as aerial users and achieve real-time data transmission. However, it should be noted that drones are constrained by factors such as weight, payload, battery, and power, and the transmission of multi-source sensor data, especially aerial photography data, is significantly limited by bandwidth. Although patent CN114245491A proposes a drone data transmission system based on a 5G network, which improves the data transmission rate by optimizing uplink scheduling and resource allocation and adjusting the coding and modulation methods.

[0004] Currently, the construction of 5G-A (5G-Advanced) base stations has introduced Integrated Sensing and Communication (ISAC) capabilities, enabling base stations to simultaneously perform communication and radar sensing to acquire information such as the location, speed, and heading of drones. Current mainstream research focuses on communication aspects such as sensing-assisted beam management and interference suppression. In existing solutions, sensing results are only reported to higher layers as reference information, and communication parameters are then adjusted via RRC signaling. The closed-loop latency typically exceeds 100 milliseconds. For high-speed flight or swarm drone scenarios, this latency cannot meet emergency response requirements. Furthermore, sensing information is not involved in the drone's uplink transmission strategy, lacking a cross-layer coupling design with a direct physical layer connection. Summary of the Invention

[0005] In view of the problems and defects of the existing technology, the purpose of this invention is to provide a method for real-time aerial photography data fusion and secure transmission of low-altitude UAVs based on 5G-A integrated sensing and health status mandatory constraints. The method extracts micro-Doppler features such as rotor speed fluctuation and body vibration of the UAV by the base station, determines the health level in real time and maps it to a mandatory transmission mode, and uses downlink control information to directly control the uplink parameters of the UAV, so as to achieve a hard constraint that prioritizes security over channel quality.

[0006] Simultaneously, by combining federated Kalman filtering with inertial navigation and perception data, the system achieves the fusion and optimal estimation output of multi-source data from UAVs, and directly deploys a lightweight AI model at the base station edge node for real-time preprocessing, thereby improving the security, real-time performance, and reliability of low-altitude UAV data transmission and processing.

[0007] According to a first aspect of the present invention, a method for real-time aerial photography data fusion and secure transmission of low-altitude unmanned aerial vehicles based on state constraints is proposed, comprising the following steps:

[0008] Step 1: Within the drone monitoring area, expand the airspace coverage angle of the 5G-A base station and set the optimized antenna downtilt angle, and configure dual-band networking to serve ground users and drones respectively;

[0009] Step 2: During flight, the UAV collects real-time sensing data through its onboard multimodal sensing system and receives OFDM sensing signals periodically transmitted by the base station. After being reflected by the UAV, the base station returns echo signals.

[0010] Step 3: The base station performs spectrum analysis on the echo signal returned by the UAV to extract the UAV's operation monitoring status, determines the health status based on the UAV's operation monitoring status, determines the UAV's forced transmission mode, and sends it to the UAV via DCI message;

[0011] Step 4: After receiving the forced transmission mode in the DCI message, the UAV performs layered encryption on the real-time sensing data and then dynamically transmits it to the base station under the forced transmission mode and CQI constraints.

[0012] Step 5: After decrypting the received encrypted data, the edge nodes on the base station side perform multi-source spatiotemporal alignment and fusion of the real-time sensing data, and use the deployed recognition model to perform real-time recognition of aerial video, outputting the detected events; and

[0013] Step 6: The edge nodes on the base station side assess the flight risk of the drone in real time based on the drone's location risk, speed risk, behavior risk, and health risk, and make adaptive adjustments and responses based on the risk values.

[0014] As an optional implementation, in step 3, the base station transmits OFDM sensing pulses at a period of 10ms and receives the echo signal returned after being reflected by the UAV.

[0015] Calculate the echo time delay and Doppler frequency shift through FFT transformation to obtain the distance and radial velocity, and solve the three-dimensional position of the UAV based on the intersection of the arrival angles of multiple base stations or the phase difference of multiple antennas of a single base station.

[0016] Perform STFT short-time Fourier transform on the echo signal, and extract the micro-Doppler features from the time-frequency spectrogram, including: the main rotor speed f_rot, the variance of speed fluctuation σ_rot, and the body vibration energy E_vib.

[0017] As an optional implementation, in step 3, based on the main rotor speed f_rot, the variance of speed fluctuation σ_rot, and the body vibration energy E_vib, and compare with the preset threshold according to the UAV model, determine the UAV health level H and the corresponding forced transmission mode:

[0018] First, according to the main rotor speed f_rot, the variance of speed fluctuation σ_rot, and the body vibration energy E_vib, determine the health assessment components H_rot, H_σ, H_vib corresponding to each feature respectively, and take the maximum value of the health assessment components corresponding to the three features as the UAV health level H: H = max(H_rot, H_σ, H_vib);

[0019] Then, determine the forced transmission mode according to the mapping relationship between the UAV health level H and the forced transmission mode:

[0020] H = 0: Normal mode, transmit data according to the CQI strategy;

[0021] H = 1: Restricted mode, allowing the transmission of status information and aerial video with a resolution less than or equal to 720p; the status information includes position data, speed data, and health data;

[0022] H = 2: Video prohibited mode, only allowing the transmission of status information;

[0023] H = 3: Heartbeat mode, switch to satellite relay to transmit heartbeat packets, including position information and health data.

[0024] As an optional implementation, in the normal mode, the UAV will measure the downlink channel quality index CQI in real time and feedback it to the base station side through the uplink, and the base station side will make a judgment according to the value of CQI:

[0025] If CQI > 15, maintain the 5G-A main channel and send the original encrypted data;

[0026] If 10 < CQI ≤ 15, start the compression mode, the video stream only sends I-frame key frames, discards P / B frames; the control signaling is sent in full;

[0027] If CQI ≤ 10, immediately switch to satellite relay channel to transmit heartbeat packets, and store aerial data locally on the drone, to be retransmitted after signal recovery.

[0028] As an optional implementation, step 4, which involves performing layered encryption on the real-time sensing data, includes:

[0029] Content-aware encryption strategies identify data types through deep packet inspection (DPI).

[0030] The aerial video stream is encrypted using a layered encryption method based on the Region of Interest (ROI). The critical areas are encrypted using SM4, while the remaining scenes are encrypted using lightweight stream encryption.

[0031] Control signaling is encrypted using the national cryptographic algorithm SM4.

[0032] As an optional implementation, the multi-source spatiotemporal alignment and fusion of real-time sensed data specifically includes the following process:

[0033] Convert inertial navigation data and base station sensing data to a unified ENU coordinate system;

[0034] Linear interpolation is performed on the base station sensing data based on the inertial navigation sampling time.

[0035] Using the UAV's spatial position and spatial attitude angle as state vectors, and a uniform velocity model as the system model, federated Kalman filtering is used to fuse inertial navigation data and base station sensing data to obtain the optimal estimated UAV state.

[0036] Select at least three ground control points with known geographic coordinates that are not on a straight line. By minimizing the sum of squared reprojection errors of all control points, transform the coordinates of each pixel in the aerial image to geographic coordinates to generate an orthorectified image.

[0037] As an optional implementation, federated Kalman filtering is used to fuse inertial navigation data with base station sensing data to obtain the optimal estimated UAV state, including:

[0038] Define the state vector X as: X=[X,Y,Z,V] X V Y V Z ] T Where X, Y, Z, V X V Y V Z These represent the spatial position coordinate components and velocity components of the UAV, respectively.

[0039] Within two adjacent inertial navigation sampling intervals Δt, assuming the UAV moves at a constant speed, determine the state transition matrix F;

[0040] The noise covariance matrix Q is determined based on the power spectral density during the movement of the UAV.

[0041] A measurement model with a main filter and sub-filter structure is adopted. The first sub-filter processes inertial navigation data and independently performs time update and measurement update of standard Kalman filtering each time inertial navigation data is received. The second sub-filter processes base station sensing data and independently performs time update and measurement update of standard Kalman filtering each time base station sensing data is received. The two sub-filters use the same state transition matrix F and noise covariance matrix Q.

[0042] The optimal state estimate of the UAV is obtained by combining the estimation results of the two sub-filters by weighting the covariance using the main filter.

[0043] As an optional implementation, in step 6, the base station edge node calculates the UAV's location risk Rpos, speed risk Rvel, behavioral risk Rbeh, and health risk Rhel according to a set period, and then performs a fusion assessment of the UAV's flight risk.

[0044] The drone position risk Rpos is determined by the drone's current position and the no-fly zone (NFZ) boundary: if the drone's current position is within the NFZ boundary, then return Rpos=3; otherwise, traverse each edge of the NFZ polygon and calculate the shortest distance from a point to a line segment, taking the minimum value d as the shortest distance from the drone to the no-fly zone boundary.

[0045] Rpos=0, d>100;

[0046] Rpos=1, 50<d≤100;

[0047] Rpos=2, 10<d≤50;

[0048] Rpos=3, d≤10;

[0049] The speed risk Rvel is determined based on the ratio of the current speed v to the area speed limit:

[0050] Rvel=min(3,max(0,((v / v max )-1)×2));

[0051] Among them, v max This indicates the maximum permissible speed in the area as defined by the airspace classification.

[0052] The behavioral risk Rbeh is set to predict the drone's trajectory over the next 5 seconds based on the PINN model to assess whether the current behavior is abnormal, specifically including:

[0053] If any point in the predicted trajectory within the next 5 seconds enters a no-fly zone or is less than 5 meters away, Rbeh increases by 2; if the predicted speed exceeds 1.5v... max If the predicted acceleration exceeds a, then Rbeh increases by 1; max If so, Rbeh increases by 0.5;

[0054] Finally, sum the three indices and truncate them to [0,3].

[0055] The health risk Rhel value is taken as the drone's health level H, with a range of [0,3].

[0056] Assess drone flight risk by integrating location risk (Rpos), speed risk (Rvel), behavioral risk (Rbeh), and health risk (Rhel).

[0057] Risk=w1·Rpos+w2·Rvel+w3·Rbeh+w4·Rhel;

[0058] Among them, w1, w2, w3, and w4 are the weight coefficients corresponding to the drone's location risk Rpos, speed risk Rvel, behavioral risk Rbeh, and health risk Rhel, respectively, with values ​​ranging from 0 to 1, and their sum is 1.

[0059] As an optional implementation, in step 6, adaptive control and response are performed based on the risk value, including:

[0060] Based on the drone flight risk obtained from the fusion assessment, the risk level and corresponding response actions are determined:

[0061] (1) Risk 0, <0.5, Action: Record only, no action taken;

[0062] (2) Risk 1, 0.5~1.5, Action: Send a notification to the command center and do not interfere with the drone;

[0063] (3) Risk 2, 1.5~2.5, Action to be taken: Send an alarm to the command center and the drone operator to alert them;

[0064] (4) Risk 3, ≥2.5, Action: Send a forced return command via downlink control information and send it to the command center;

[0065] Upon receiving a forced return command, the drone's flight control system immediately aborts the current mission and executes the preset return procedure.

[0066] Compared with existing technologies, the significant advantages of the state-constrained real-time aerial photography data fusion and secure transmission method for low-altitude UAVs of the present invention are as follows:

[0067] (1) By leveraging 5G-A's integrated sensing capabilities and base station edge processing, the dynamic coupling of UAV trajectory perception and data transmission is achieved, improving transmission efficiency and success rate, especially suitable for low-bandwidth scenarios;

[0068] (2) Using a layered encryption mechanism, the computational overhead can be significantly reduced by 40% compared to the traditional AES algorithm, while meeting the requirements for low-altitude data confidentiality;

[0069] (3) By using a spatiotemporal alignment engine with federated Kalman filtering, soft constraints and compensations of road features are fused to reduce fusion errors and support precise geographic information applications.

[0070] The foregoing and other aspects, embodiments, and features of the teachings of the present invention will be more fully understood from the following description in conjunction with the accompanying drawings. Other additional aspects of the invention, such as features and / or beneficial effects of exemplary embodiments, will become apparent from the following description or may be learned through practice of specific embodiments according to the teachings of the present invention. Attached Figure Description

[0071] Figure 1 This is a flowchart illustrating a method for real-time aerial photography data fusion and secure transmission of low-altitude unmanned aerial vehicles based on state constraints, according to an embodiment of the present invention.

[0072] Figure 2 This is a schematic diagram of the process of performing multi-source spatiotemporal alignment and fusion of real-time sensing data on the base station side according to an embodiment of the present invention.

[0073] Figure 3 This is a schematic flowchart of the joint Kalman fusion filtering according to an embodiment of the present invention. Detailed Implementation

[0074] To better understand the technical content of the present invention, specific embodiments are described below in conjunction with the accompanying drawings.

[0075] Various aspects of the invention are described in this disclosure with reference to the accompanying drawings, which illustrate numerous illustrative embodiments. The embodiments of this disclosure are not necessarily intended to encompass all aspects of the invention. It should be understood that the various concepts and embodiments described above, as well as those described in more detail below, can be implemented in any of many ways, because the concepts and embodiments disclosed herein are not limited to any particular implementation. Furthermore, some aspects of the invention disclosed may be used alone or in any suitable combination with other aspects of the invention disclosed.

[0076] Referring to the accompanying drawings, the method for real-time aerial photography data fusion and secure transmission of low-altitude UAVs based on state constraints according to an embodiment of the present invention includes the following steps:

[0077] Step 1: Within the drone monitoring area, expand the airspace coverage angle of the 5G-A base station and set the optimized antenna downtilt angle, and configure dual-band networking to serve ground users and drones respectively;

[0078] Step 2: During flight, the UAV collects real-time sensing data through its onboard multimodal sensing system and receives OFDM sensing signals periodically transmitted by the base station. After being reflected by the UAV, the base station returns echo signals.

[0079] Step 3: The base station performs spectrum analysis on the echo signal returned by the UAV to extract the UAV's operation monitoring status, determines the health status based on the UAV's operation monitoring status, determines the UAV's forced transmission mode, and sends it to the UAV via DCI message;

[0080] Step 4: After receiving the forced transmission mode in the DCI message, the UAV performs layered encryption on the real-time sensing data and then dynamically transmits it to the base station under the forced transmission mode and CQI constraints.

[0081] Step 5: After decrypting the received encrypted data, the edge nodes on the base station side perform multi-source spatiotemporal alignment and fusion of the real-time sensing data, and use the deployed recognition model to perform real-time recognition of aerial video, outputting the detected events; and

[0082] Step 6: The edge nodes on the base station side assess the flight risk of the drone in real time based on the drone's location risk, speed risk, behavior risk, and health risk, and make adaptive adjustments and responses based on the risk values.

[0083] Therefore, the state-constrained real-time aerial photography data fusion and secure transmission method for low-altitude UAVs of this invention, with a 5G-A integrated sensing base station as the core, overcomes the traditional strategy of prioritizing communication and supplementing with sensing, where sensing results are only used as a reference and do not interfere with communication. This invention proposes a secure communication transmission method for UAV aerial photography data based on sensing-forced constraint communication. The 5G-A integrated sensing base station simultaneously completes communication and radar sensing, extracts micro-Doppler features from the echo and determines the health level (0~3), and uses health characteristics such as UAV rotor speed fluctuations and body vibration as hard constraints. The uplink transmission parameters of the UAV are directly overwritten through physical layer downlink control information (DCI). Accordingly, the UAV MAC layer controls the output of the forced mode gated video encoder and performs layered encrypted transmission. Based on the UAV state-aware forced constraint communication, secure communication transmission of UAV aerial photography data is achieved. Furthermore, at the base station edge node, multi-source spatiotemporal alignment and edge fusion are used to achieve secure, real-time, and intelligent governance of low-altitude data.

[0084] In step 1, the installation location of the base station can be determined based on the actual boundary of the monitoring area (such as disaster area, farmland, or sea area) and the minimum circle coverage algorithm using a list of polygon vertex coordinates, ensuring full coverage. A 128-channel 5G-A integrated sensing base station is deployed, using two independent carriers (100MHz carriers, labeled carrier_ground and carrier_uav) in the 3.5GHz band. One carrier is used for ground user communication, and the other is dedicated to low-altitude UAV communication. Ground user equipment can only access carrier_ground, and UAV terminals can only access carrier_uav. In the base station scheduler, 70% of the time-frequency resources are reserved for carrier_uav to ensure an uplink rate of no less than 50Mbps, achieving ground-to-air frequency isolation and avoiding mutual interference.

[0085] As an optional implementation, the monitoring area is divided into circular coverage units with a radius of 500 meters. The minimum circle coverage algorithm is used: the center of the circle with the smallest radius that can cover all vertices is found and used as the first candidate location for the base station. The area already covered by this circle is removed from the polygon, and the above process is repeated until full coverage is achieved, resulting in a list of base station coordinates (lat_i, lon_i).

[0086] Furthermore, based on the low-altitude UAV flight altitude H (controlled at 300m according to air traffic control and flight requirements) and the base station mounting height h (30m), combined with the preset maximum coverage distance d, the base station antenna downtilt angle θ = arctan((Hh) / d) is calculated, and the vertical beamwidth of the airspace coverage angle is set to ±24°, covering a range of 30° to -18° (relative to the horizontal plane). This ensures that when the UAV flies at an altitude of 300 meters, the uplink rate is improved and the communication latency is controlled within 50ms.

[0087] Therefore, by optimizing the deployment of 5G-A integrated sensing base stations, the low-altitude network and the ground user network are logically completely isolated, avoiding resource contention and ensuring that the drone can obtain stable and low-latency uplink and downlink links throughout the entire flight, thus achieving secure and stable transmission of sensing and control signals.

[0088] The drone performs pre-set inspection tasks, flying automatically according to a pre-set flight plan and route. It simultaneously acquires aerial images, environmental sensor data, and its own precise position, speed, and heading data through its onboard equipment. The aforementioned onboard equipment includes, but is not limited to: high-resolution multispectral cameras (for vegetation and water analysis), thermal imagers (for fire or personnel detection), and gas sensors (for hazardous gas monitoring).

[0089] During flight, the UAV continuously acquires imagery (e.g., multispectral cameras continuously capture images at 0.5-second intervals) and sensor data. The sensor data includes: the inertial navigation system outputting the UAV's 3D position (x_ins, y_ins, z_ins) and velocity (vx, vy, vz) every 0.01 seconds; and the thermal imager and gas sensors outputting values ​​every 0.1 seconds. Each data point is accompanied by a current UTC timestamp and BeiDou positioning coordinates, and the data is transmitted to the base station via uplink.

[0090] Meanwhile, the ground-based 5G-A integrated sensing base station serves as both a communication and sensing terminal, transmitting sensing signals and receiving echoes reflected from the drone.

[0091] In the example of this invention, the base station transmits an OFDM sensing pulse (period 10ms) and receives the echo signal returned after being reflected by the UAV. Then, the echo delay τ and Doppler frequency shift f_d are calculated using FFT, leading to further calculations:

[0092] Distance R = cτ / 2; c represents the speed of light;

[0093] Radial velocity v_r = f_d·λ / 2, λ = 0.0857m.

[0094] As an example, the three-dimensional position (X_5G, Y_5G, Z_5G) of the drone is calculated by using the intersection of the angles of arrival of multiple base stations, which serves as the drone position data sensed by the base stations.

[0095] Simultaneously, the base station performs a short-time Fourier transform (STFT) on the echo with a window length of 256 points and an overlap rate of 50%, and extracts micro-Doppler features from the time-spectrum graph.

[0096] Main rotor speed f_rot (Hz): peak frequency;

[0097] Rotational speed fluctuation variance σ_rot: the variance of 10 consecutive measurements;

[0098] Vibration energy of the body E_vib: average power in the 100~500Hz frequency band.

[0099] Based on this, and by comparing the main rotor speed f_rot, speed fluctuation variance σ_rot, and airframe vibration energy E_vib, as well as the preset thresholds for the UAV model, the health level H of the UAV and the corresponding forced transmission mode are determined.

[0100] First, based on the main rotor speed f_rot, the speed fluctuation variance σ_rot, and the airframe vibration energy E_vib, the health assessment components H_rot, H_σ, and H_vib corresponding to each feature are determined respectively. The maximum value of the three health assessment components corresponding to the three features is taken as the health level H of the UAV: ​​H=max(H_rot,H_σ,H_vib).

[0101] Then, based on the mapping relationship between the drone's health level H and the forced transmission mode, the forced transmission mode is determined:

[0102] H=0: Normal mode, data is transmitted according to the CQI policy;

[0103] H=1: Restricted mode, allowing the transmission of status information and aerial video with a resolution of less than or equal to 720p; the status information includes location data, speed data, and health data;

[0104] H=2: Video mode is disabled; only status information can be transmitted.

[0105] H=3: Heartbeat mode, switches to satellite relay transmission of heartbeat packets, including location information and health data.

[0106] After the UAV health level H is calculated and obtained by the sensing and processing unit (FPGA) on the base station side, downlink control information (DCI) is immediately generated. M_force is encoded using a 2-bit reserved field. The DCI is sent through the physical downlink control channel in the next time slot (within 1ms).

[0107] The MAC layer of the UAV's communication module parses the downlink control information (DCI) and gates the video encoder output according to the M_force value:

[0108] M_force=DisableVideo (H=2): Physically disconnects the video stream channel;

[0109] M_force=Heartbeat Only (H=3): Reduces the frequency of status information transmission from 10Hz to 1Hz.

[0110] In the example of this invention, after the UAV receives M_force from the DCI, a gating switch is added before the video encoder:

[0111] When M_force=2, the gating is closed, video data does not enter the encoder, and only status information is packaged.

[0112] When M_force=3, control status information is transmitted at a low frequency (e.g., from 10 times per second to 1 time per second), and the satellite relay is switched to transmit heartbeat packets containing location information and health data.

[0113] When M_force = 0 or 1, it is allowed to transmit aerial video information and status information. Thus, the secure dynamic transmission of sensing data is carried out based on the safe and healthy state of the drone and the transmission channel quality CQI.

[0114] After the sensing processing unit on the base station side obtains the H value, it does not pass through the high-level protocol stack. After the H value is sent to the drone in the next time slot (within 1 ms) through the downlink control information (DCI) of the physical layer, the communication module on the drone side analyzes the forced mode field in the DCI at the MAC layer, immediately shields the output of the video encoder, and only allows specified data types to pass through. The entire process does not pass through the flight control system or the RRC layer. Through the tight coupling of the physical layer and the MAC layer, the ultra-low latency control ≤ 1 ms is achieved.

[0115] In an optional embodiment, in the normal mode, the drone will measure the downlink channel quality index CQI in real time and feedback it to the base station side through the uplink. The base station side makes a judgment based on the value of CQI:

[0116] If CQI > 15, maintain the 5G-A main channel and send the original encrypted data;

[0117] If 10 < CQI ≤ 15, start the compression mode. The video stream only sends I-frame key frames and discards P / B frames; the control signaling is sent in full;

[0118] If CQI ≤ 10, immediately switch to the satellite relay channel to transmit heartbeat packets, such as Beidou short messages, at a rate of 500 bps, which is only enough to send alarms and positions. At the same time, store the aerial photography data locally on the drone and retransmit it after the signal recovers.

[0119] In the embodiment of the present invention, the data transmission adopts a dual-channel redundancy design. The main channel is the 5G-A network, and the secondary channel is satellite communication to ensure reliable transmission in complex terrains.

[0120] As an optional implementation manner, in step 4, the operation of performing hierarchical encryption on the real-time sensing data includes:

[0121] Based on the content-aware encryption strategy, identify the data type through deep packet DPI detection:

[0122] Adopt region of interest (ROI) hierarchical encryption for the aerial photography image video stream, where the key area is encrypted using SM4, and the remaining scenes adopt lightweight stream encryption;

[0123] Use the national secret SM4 algorithm to encrypt the control signaling.

[0124] As an optional implementation manner, the content-aware monitoring and identification include:

[0125] Perform deep packet detection (DPI) on the packet header of the data packet:

[0126] If the target port is 554 (RTSP) and the payload is an H.264 start code, it is determined to be a video stream;

[0127] If the target port is 5000 and the data length is fixed at 128 bytes, it is determined to be control signaling;

[0128] The rest are ordinary images.

[0129] Further analysis of the video stream: Use a lightweight object detection model (such as YOLO-tiny) to extract regions of interest (ROIs), such as faces, license plates, and flames in the scene, and record the coordinates of the bounding boxes of the ROIs.

[0130] Furthermore, layered encryption processing is performed based on the content-aware recognition results:

[0131] For control signaling, the ECB mode of the national cryptographic SM4 algorithm is directly used for encryption, with a key length of 128 bits.

[0132] For video streams: macroblocks within the ROI area are encrypted using SM4; non-ROI areas are encrypted using RC4 stream encryption (the key is changed every 5 seconds).

[0133] For ordinary imagery: if the imagery's geographic coordinates fall within a preset sensitive area (e.g., a restricted area), then SM4 encryption is used; otherwise, lightweight PRINCE encryption is used (halving the number of rounds and reducing overhead by 60%).

[0134] As mentioned above, based on the health status and the downlink channel quality index (CQI) measured in real time by the drone, the base station determines the encrypted secure transmission method by judging the H value and CQI: the main channel is 5G-A, and the backup channel is satellite (such as BeiDou short message). Automatic switching occurs when CQI ≤ 10 or H = 3.

[0135] In an optional embodiment, the edge node on the base station side receives encrypted uplink data from the UAV and decrypts it using a shared key (SM4 or a corresponding lightweight algorithm) to obtain inertial navigation data and aerial image data. Combined with sensing data from the base station itself (including the UAV's position (x_5G, y_5G, z_5G) and timestamp t_5G in the base station coordinate system), the fused high-precision UAV state sequence is output through multi-source spatiotemporal alignment and fusion. This sequence includes position (X_fused, Y_fused, Z_fused) and velocity (VX_fused, VY_fused, VZ_fused) in groups of 0.01 seconds, as well as geocoded aerial imagery.

[0136] In this example, two types of data streams are obtained after decryption: inertial navigation data stream and aerial image stream.

[0137] Inertial navigation data stream: frequency 100Hz, each stream contains t_INS (GPS time, millisecond precision), (x_INS, y_INS, z_INS) (body coordinate system position with takeoff point as origin, in meters), and (vx_INS, vy_INS, vz_INS) (velocity, in meters / second).

[0138] Aerial image stream: 2Hz frequency, each stream contains t_cam, image matrix I(u,v), camera intrinsic parameters (focal length, distortion coefficient), and the approximate GPS coordinates of the drone at the time of shooting (lat_cam,lon_cam,alt_cam).

[0139] Since the inertial navigation data has already been transmitted by the drone via an encrypted uplink, the base station does not need additional sensing capabilities. The base station's sensing data is generated independently and does not depend on the drone's upload.

[0140] As a specific example, such as Figure 2 As shown, the multi-source spatiotemporal alignment and fusion processing of real-time sensing data includes the following steps:

[0141] Convert inertial navigation data and base station sensing data to a unified ENU coordinate system;

[0142] Linear interpolation is performed on the base station sensing data based on the inertial navigation sampling time.

[0143] Using the UAV's spatial position and spatial attitude angle as state vectors, and a uniform velocity model as the system model, federated Kalman filtering is used to fuse inertial navigation data and base station sensing data to obtain the optimal estimated UAV state.

[0144] Select at least three ground control points with known geographic coordinates that are not on a straight line. By minimizing the sum of squared reprojection errors of all control points, transform the coordinates of each pixel in the aerial image to geographic coordinates to generate an orthorectified image.

[0145] In the method of this invention, data from different sources are converted to the same Cartesian coordinate system for subsequent alignment and fusion processing. In this example, the East-North-Up (ENU) coordinate system with the center point of the monitoring area as the origin is selected, and the unit is meters.

[0146] The absolute geographic coordinates (latitude, longitude, altitude) of the UAV takeoff point are known to be (lat0, lon0, alt0), recorded by the ground station. The inertial navigation system (INS) provides (x_INS, y_INS, z_INS) as the northeast-sky coordinates relative to the takeoff point (assuming the INS has completed initial alignment). Taking the takeoff point as the ENU origin (region center) as an example, we have:

[0147] XINS=xINS;

[0148] YINS = yINS;

[0149] ZINS = zINS;

[0150] It should be understood that if the takeoff point is not in the center of the area, a corresponding translation can be performed.

[0151] Furthermore, coordinate transformation is performed on the base station sensing data, including:

[0152] Given the geographic coordinates (lat_BS, lon_BS, alt_BS) of each base station, and combining them with the local coordinate system output by the base station sensing, with the base station antenna phase center as the origin (usually x_5G points due east, y_5G points due north, and z_5G is vertically upward (i.e., in the ENU direction)), the base station sensing coordinates are transformed to the regional ENU coordinates through the following translation:

[0153] X_5G = x_5G + (E_BS - E_center);

[0154] Y_5G = y_5G + (N_BS - N_center);

[0155] Z_5G = z_5G + (U_BS - U_center);

[0156] Where (E_BS,N_BS,U_BS) and (E_center,N_center,U_center) are the coordinates of the base station and the regional center in the global ENU (calculated by latitude and longitude).

[0157] Among them, (X_5G,Y_5G,Z_5G) is the unified northeast celestial coordinate system, with the origin at the center of the region, X pointing east, Y pointing north, and Z pointing upward. All coordinate units are meters.

[0158] Based on coordinate alignment, since the inertial navigation sampling interval Δt_INS=0.01s and the base station sensing sampling interval Δt_5G=0.1s (because the base station sends sensing pulses with a period of 10ms, and multi-pulse accumulation is used in actual operation with an effective output rate of 10Hz), it is necessary to interpolate the base station sensing data to each inertial navigation time t_INS(k).

[0159] In the example of this invention, linear interpolation is used:

[0160] For any target time t, find the sensing times t1 and t2 of the two base stations before and after the target, with corresponding sensing positions (X_5G(t1), Y_5G(t1), Z_5G(t1)) and (X_5G(t2), Y_5G(t2), Z_5G(t2)), where t1≤t≤t2), then:

[0161] X_5G(t)=X_5G(t1)+[(X_5G(t2)-X_5G(t1)) / (t2-t1)]·(t-t1);

[0162] Similarly, the same processing is applied to the Y and Z directions:

[0163] Y_5G(t)=Y_5G(t1)+[(Y_5G(t2)-Y_5G(t1)) / (t2-t1)]·(t-t1);

[0164] Z_5G(t)=Z_5G(t1)+[(Z_5G(t2)-Z_5G(t1)) / (t2-t1)]·(t-t1).

[0165] Furthermore, combined Figure 3 As shown, using the UAV's spatial position and spatial attitude angle as state vectors, and a uniform velocity model as the system model, a federated Kalman filter is used to fuse inertial navigation data and base station sensing data to obtain the optimal estimated UAV state. The specific process includes the following steps:

[0166] Define the state vector X as: X=[X,Y,Z,V] X V Y V Z ] T Where X, Y, Z, V X V Y V Z These represent the spatial position coordinate components and velocity components of the UAV, respectively.

[0167] Within two adjacent inertial navigation sampling intervals Δt (Δt=0.01s), assuming the UAV moves at a constant velocity, the state transition matrix F is (a 6×6 matrix):

[0168] ;

[0169] The two inertial navigation sampling intervals Δt define the time step; for example, 0.01 seconds. The 3×3 block in the upper left corner is the identity matrix, indicating that the position remains unchanged when there is no velocity. The 3×3 block in the upper right corner is the diagonal matrix Δt· I _3 represents the positional change caused by velocity; the 3×3 block in the lower right corner is the identity matrix, indicating that the velocity remains constant in the model prediction (i.e., the uniform velocity assumption).

[0170] Process noise (unknown acceleration during UAV flight) is introduced into the model. Let the power spectral density of the acceleration noise be q. Based on the power spectral density q during UAV motion, the process noise covariance matrix Q is determined after discretization for a continuous-time system.

[0171] ;

[0172] In this example, q is an empirical value, q=0.1m² / s³, which can be adjusted according to the maximum acceleration of the UAV, and represents the spectral density of the acceleration random walk;

[0173] Then, a measurement model with a main filter and sub-filters is adopted. The sub-filters are used to perform fusion filtering on the two independent measurement sources, inertial navigation and base station sensing, respectively. The main filter receives the estimation results of the two sub-filters and performs optimal fusion.

[0174] As a specific example, the first sub-filter processes inertial navigation data and independently performs time and measurement updates of standard Kalman filtering each time inertial navigation data is received; the second sub-filter processes base station sensing data and independently performs time and measurement updates of standard Kalman filtering each time base station sensing data is received. Both sub-filters use the same state transition matrix F and noise covariance matrix Q.

[0175] The estimation results of the two sub-filters are combined by weighting the covariance using the main filter:

[0176] ;

[0177] ;

[0178] In the formula, P 5G and P INS X represents the covariance matrices of the base station sensing data and the inertial navigation data, respectively. The inverses of the covariance matrices are the information matrices estimated by the base station sub-filter and the inertial navigation sub-filter, respectively, which represent the confidence level of the estimation; 5G and X INS Do not perform state estimation for base station sensing data and inertial navigation data.

[0179] Therefore, the fused information matrix is ​​the sum of the two information matrices, that is, the fusion of their respective confidence levels. The fused state is the weighted average of the two information matrices according to their confidence levels.

[0180] For inertial navigation data, since the inertial navigation system directly outputs the position and velocity, the observation matrix H_INS is determined to be a 6×6 identity matrix.

[0181] H_INS=I_3;

[0182] The measurement noise covariance matrix R_INS is set according to the accuracy of the inertial navigation device; taking a medium-precision MEMS inertial navigation system as an example:

[0183] RINS=diag(σ pos 2 ,σ pos 2 ,σ pos 2 ,σ vel 2 ,σ vel 2 ,σ vel 2 );

[0184] Where σ_pos=0.1m represents the position standard deviation, and σ_vel=0.05m / s represents the velocity standard deviation.

[0185] For base station sensing measurements, since base station sensing provides location (three-dimensional) but not velocity, the observation matrix H_5G uses a 3×6 matrix:

[0186] ;

[0187] The measurement noise covariance matrix R_5G depends on the sensing mode:

[0188] (1) Single base station (angle + distance only) mode: σ pos ≈0.5m, therefore R_5G=diag(0.25,0.25,0.25);

[0189] (2) Multi-base station convergence mode (≥3 base stations): σ pos ≈0.1m, therefore R_5G}=diag(0.01,0.01,0.01).

[0190] Based on this, the first sub-filter and the second sub-filter independently run standard Kalman filtering to perform time updates and measurement updates (when there is new inertial navigation data and new base station sensing data after interpolation).

[0191] Finally, based on the fusion results of the main filter, at each inertial navigation sampling time (i.e. every 0.01s), the main filter receives the estimation results of the two sub-filters and performs optimal fusion by weighting the results according to the covariance.

[0192] For example, considering only the position component, ignoring velocity, and assuming the three position axes are independent:

[0193] Let P_INS = diag(0.01, 0.01, 0.01) (standard deviation 0.1 meters);

[0194] P_5G=diag(0.25,0.25,0.25) (standard deviation 0.5 meters);

[0195] Then P_INS -1 =diag(100,100,100), P_5G -1 =diag(4,4,4).

[0196] Summing yields diag(104,104,104), and inversely, P_fused≈diag(0.00962,0.00962,0.00962), meaning the standard deviation is approximately 0.098 meters. The accuracy of the fused sample is superior to that of any single sensor.

[0197] If the inertial navigation system estimates the position at 245.30m and the base station estimates it at 245.45m, then the calculated position based on the aforementioned fusion is approximately 245.38m, which is between the two and closer to the inertial navigation system with higher confidence.

[0198] Furthermore, for aerial imagery, fine correction is performed using ground control points (GCPs). At least three ground control points with known geographic coordinates that are not collinear are selected. By minimizing the sum of squared reprojection errors of all control points, the coordinates of each pixel in the aerial imagery are transformed to geographic coordinates, generating an orthorectified image. Specifically, this includes:

[0199] Several ground control points are pre-deployed within the monitoring area, such as painted cross marks, road marking corner points, etc. Each control point has known precise geographic coordinates (X_gcp, Y_gcp), and its pixel coordinates (u_gcp, v_gcp) can be manually or automatically identified in aerial images. At least 3 non-collinear control points are required, and generally 5 to 10 are used to improve accuracy.

[0200] Assuming there is an affine transformation between the image and the geographic coordinates, then:

[0201] X = a·u + b·v + e;

[0202] Y = c·u + d·v + f;

[0203] Where (u,v) are pixel coordinates (origin at the top left corner, unit is pixels), (X,Y) are geographic coordinates (unit is meters), and the six parameters (a,b,c,d,e,f) need to be solved;

[0204] Then, based on the least squares method, for each control point GCP, the positional relationship is written in matrix form by combining the above affine transformation, pixel coordinates and geographic coordinates. By minimizing the sum of squared reprojection errors of all control points, each parameter (a,b,c,d,e,f) is obtained.

[0205] In a further optional embodiment, when the number of control points is insufficient or unevenly distributed (e.g., in rural or mountainous areas), to avoid deformation of non-control point areas caused by control point correction, road features can be added as soft constraints. A lightweight semantic segmentation network (such as U-Net) can be used to extract binary road maps from aerial images, and then the road vector lines of the area can be obtained from the local geographic information system to determine the geographic coordinates. Thus, in the least squares process, for each pixel j (M in total) belonging to a road in the image, it is mapped to geographic coordinates (X_j, Y_j) = T(u_j, v_j) through the current transformation parameter T. The distance from this point to the nearest GIS road curve is calculated. For example, the Euclidean distance between the point and all line segments of the road curve is calculated, and the minimum distance is taken as the shortest Euclidean distance from the point to the curve. This incorporates the road input into the total loss, and the sum of squared reprojection errors of the control points L is included. point Road loss L road The weighted average (multiplied by a weighting coefficient of 0.3-0.4) is used as the total loss of least squares, thereby correcting the situation where large-area deformation can occur when only a few points are used for correction when control points are sparse, and improving the geometric accuracy of the entire image.

[0206] The solution process can be iteratively optimized based on LM or gradient descent, using existing technologies.

[0207] In an optional embodiment, the average distance error from the control point to each road pixel is calculated. If it is greater than a preset threshold (e.g., 0.5 meters), it indicates that more control points are needed for more accurate pixel coordinate fusion.

[0208] Finally, at the edge nodes on the base station side, multi-source spatiotemporal alignment fusion is used to obtain the optimal estimated UAV state and the corrected image (transformed to the geographic coordinate system, with pixels corresponding to coordinates). The UAV state is in the form of the fused UAV trajectory, that is, the spatial position coordinate information (X_fused, Y_fused, Z_fused) and the corresponding velocity information (VX_fused, VY_fused, VZ_fused) at each inertial navigation sampling time.

[0209] In an embodiment of the present invention, we use a recognition model deployed at the edge node of the base station to perform real-time recognition of aerial video and output detection events.

[0210] As an optional example, the identification model can use existing detection models, especially lightweight models, such as fire detection models based on MobileNetV2, vehicle technology models based on CSRNet lightweight, NDWI water pollution index models based on multispectral input, personnel detection models (YOLO-tiny), smoke detection models, etc.

[0211] The model identifies and outputs corresponding detection events, such as event alarms and keyframe thumbnails.

[0212] As an optional implementation, in step 6, the base station edge node calculates the UAV's location risk Rpos, speed risk Rvel, behavioral risk Rbeh, and health risk Rhel according to a set period, and then performs a fusion assessment of the UAV's flight risk.

[0213] The drone position risk Rpos is determined by the drone's current position and the no-fly zone (NFZ) boundary: if the drone's current position is within the NFZ boundary, then return Rpos=3; otherwise, traverse each edge of the NFZ polygon and calculate the shortest distance from a point to a line segment, taking the minimum value d as the shortest distance from the drone to the no-fly zone boundary.

[0214] Rpos=0, d>100;

[0215] Rpos=1, 50<d≤100;

[0216] Rpos=2, 10<d≤50;

[0217] Rpos=3, d≤10;

[0218] The speed risk Rvel is determined based on the ratio of the current speed v to the area speed limit:

[0219] Rvel=min(3,max(0,((v / v max )-1)×2));

[0220] Among them, v max This indicates the maximum permissible speed in the area as defined by the airspace classification.

[0221] The behavioral risk Rbeh is set to predict the drone's trajectory over the next 5 seconds based on the PINN model to assess whether the current behavior is abnormal, specifically including:

[0222] If any point in the predicted trajectory within the next 5 seconds enters a no-fly zone or is less than 5 meters away, Rbeh increases by 2; if the predicted speed exceeds 1.5v... max If the predicted acceleration exceeds a, then Rbeh increases by 1; max If so, Rbeh increases by 0.5;

[0223] Finally, sum the three indices and truncate them to [0,3].

[0224] The health risk Rhel value is taken as the drone's health level H, with a range of [0,3].

[0225] Assess drone flight risk by integrating location risk (Rpos), speed risk (Rvel), behavioral risk (Rbeh), and health risk (Rhel).

[0226] Risk=w1⋅Rpos+w2⋅Rvel+w3⋅Rbeh+w4⋅Rhel;

[0227] Among them, w1, w2, w3, and w4 are the weight coefficients corresponding to the drone's location risk Rpos, speed risk Rvel, behavioral risk Rbeh, and health risk Rhel, respectively, with values ​​ranging from 0 to 1, and their sum is 1.

[0228] It should be understood that when planning a mission, the command center writes the scenario type (integer encoding, 0~4) into the UAV mission file, and the UAV attaches this tag when uploading the mission plan. The base station edge nodes parse the tag and automatically load the corresponding weight vector. The weight vector remains unchanged during mission execution.

[0229] If no scene label is received, the base station can make online inferences based on the following characteristics: if the flight altitude is >200m, it is presumed to be a surveying scene, and the weight of Rpos is increased; if the flight speed is >20 m / s, it is presumed to be an emergency scene, and the weight of Rbeh is increased; if there are a large number of high-density no-fly zones in the flight path planning, the weight of Rpos is increased.

[0230] As examples, in routine infrastructure inspection tasks, the weights of w1, w2, w3, and w4 are allocated as follows: 0.4, 0.05, 0.35, and 0.2. In emergency command, the weights are allocated as follows: 0.3, 0.1, 0.4, and 0.2. In land surveying, the weights are allocated as follows: 0.5, 0.1, 0.2, and 0.2. In urban logistics inspection, the weights are allocated as follows: 0.5, 0.1, 0.2, and 0.2.

[0231] As an example, adaptive control and response based on risk values ​​include:

[0232] Based on the drone flight risk obtained from the fusion assessment, the risk level and corresponding response actions are determined:

[0233] (1) Risk 0, <0.5, Action: Record only, no action taken;

[0234] (2) Risk 1, 0.5~1.5, Action: Send a notification to the command center and do not interfere with the drone;

[0235] (3) Risk 2, 1.5~2.5, Action to be taken: Send an alarm to the command center and the drone operator to alert them;

[0236] (4) Risk 3, ≥2.5, Action: Send a forced return command via downlink control information and send it to the command center;

[0237] Upon receiving a forced return command, the drone's flight control system immediately aborts the current mission and executes the preset return procedure.

[0238] In an optional embodiment, the edge node on the base station side can package and report each event with a risk level ≥1 to the command center. After receiving the report, the command center can display the drone aerial photography data and risk alarm information in real time.

[0239] In a further embodiment, based on aerial video of each edge node, risk alarm information can be output, and combined with the spatiotemporal attention weight heatmap of each edge node (location), the weight ratio of high-risk nodes and time periods can be presented. The identified high-risk results can then be traced back.

[0240] While the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the invention. Those skilled in the art can make various modifications and refinements without departing from the spirit and scope of the invention. Therefore, the scope of protection of the present invention shall be determined by the claims.

Claims

1. A method for real-time aerial photography data fusion and secure transmission of low-altitude UAVs based on state constraints, characterized in that, Includes the following steps: Step 1: Within the drone monitoring area, expand the airspace coverage angle of the 5G-A base station and set the optimized antenna downtilt angle, and configure dual-band networking to serve ground users and drones respectively; Step 2: During flight, the UAV collects real-time sensing data through its onboard multimodal sensing system and receives OFDM sensing signals periodically transmitted by the base station. After being reflected by the UAV, the base station returns echo signals. Step 3: The base station performs spectrum analysis on the echo signal returned by the UAV to extract the UAV's operation monitoring status, determines the health status based on the UAV's operation monitoring status, determines the UAV's forced transmission mode, and sends it to the UAV via DCI message; Step 4: After receiving the forced transmission mode in the DCI message, the UAV performs layered encryption on the real-time sensing data and then dynamically transmits it to the base station under the forced transmission mode and CQI constraints. Step 5: After decrypting the received encrypted data, the edge node on the base station side performs multi-source spatiotemporal alignment and fusion of real-time sensing data, and uses the deployed recognition model to perform real-time recognition of aerial video and output detection events; as well as Step 6: The edge nodes on the base station side assess the flight risk of the drone in real time based on the drone's location risk, speed risk, behavior risk, and health risk, and make adaptive adjustments and responses based on the risk values.

2. The method for real-time aerial photography data fusion and secure transmission of low-altitude UAVs based on state constraints according to claim 1, characterized in that, In step 1, the downtilt angle of the base station antenna is calculated based on the flight altitude H of the low-altitude UAV and the mounting height h of the base station, and the vertical beamwidth of the airspace coverage angle is set to ±24°, covering a range of 30° to -18°.

3. The method for real-time aerial photography data fusion and secure transmission of low-altitude UAVs based on state constraints according to claim 1, characterized in that, In step 3, the base station transmits OFDM sensing pulses at a period of 10ms and receives the echo signals returned after being reflected by the UAV. The echo delay and Doppler frequency shift are calculated by FFT transformation to obtain the distance and radial velocity. The three-dimensional position of the UAV is calculated based on the intersection of the angle of arrival of multiple base stations or the phase difference of multiple antennas of a single base station. The echo signal was subjected to STFT short-time Fourier transform, and micro-Doppler features were extracted from the time-spectrum diagram, including: main rotor speed f_rot, speed fluctuation variance σ_rot, and airframe vibration energy E_vib.

4. The method for real-time aerial photography data fusion and secure transmission of low-altitude UAVs based on state constraints according to claim 3, characterized in that, In step 3, based on the main rotor speed f_rot, the speed fluctuation variance σ_rot, and the airframe vibration energy E_vib, and by comparing with the preset thresholds for the UAV model, the UAV health level H and the corresponding forced transmission mode are determined. First, based on the main rotor speed f_rot, the speed fluctuation variance σ_rot, and the airframe vibration energy E_vib, the health assessment components H_rot, H_σ, and H_vib corresponding to each feature are determined respectively. The maximum value of the three health assessment components corresponding to the three features is taken as the health level H of the UAV: ​​H=max(H_rot,H_σ,H_vib). Then, based on the mapping relationship between the drone's health level H and the forced transmission mode, the forced transmission mode is determined: H=0: Normal mode, data is transmitted according to the CQI policy; H=1: Restricted mode, allowing the transmission of status information and aerial video with a resolution of less than or equal to 720p; the status information includes location data, speed data, and health data; H = 2: Prohibit video mode, only allow the transmission of status information; H = 3: Heartbeat mode, switch the satellite relay to transmit heartbeat packets, including position information and health data.

5. The method for real-time aerial photography data fusion and secure transmission of low-altitude UAVs based on state constraints according to claim 3, characterized in that, In the normal mode, the drone will measure the downlink channel quality index CQI in real time and feedback it to the base station side through the uplink. The base station side will make a judgment based on the value of CQI: If CQI > 15, maintain the 5G-A main channel and send the original encrypted data; If 10 < CQI ≤ 15, start the compression mode. The video stream only sends I-frame key frames and discards P / B frames; the control signaling is sent in full volume; If CQI ≤ 10, immediately switch to the satellite relay channel to transmit heartbeat packets, and at the same time store the aerial photography data locally on the drone and retransmit it after the signal recovers.

6. The method for real-time aerial photography data fusion and secure transmission of low-altitude UAVs based on state constraints according to any one of claims 1-5, characterized in that, In step 4, the operation of hierarchical encryption of real-time perception data includes: Based on the content-aware encryption strategy, identify the data type through deep packet inspection DPI: Adopt region of interest ROI hierarchical encryption for the aerial photography image video stream, where the key area is encrypted using SM4, and the rest of the scenes use lightweight stream encryption; 7. The method for real-time aerial photography data fusion and secure transmission of low-altitude UAVs based on state constraints according to any one of claims 1-5, characterized in that, Use the national standard SM4 algorithm to encrypt the control signaling. The multi-source spatio-temporal alignment and fusion of the real-time perception data specifically includes the following processes: Convert the inertial navigation data and base station perception data to a unified ENU coordinate system; Based on the inertial navigation sampling time as a reference, perform linear interpolation on the base station perception data; Use the drone's spatial position and spatial attitude angle as the state vector, use the uniform motion model as the system model, and adopt the federated Kalman filter to fuse the inertial navigation data and the base station perception data to obtain the optimal estimated state of the drone; 8. The method for real-time aerial photography data fusion and secure transmission of low-altitude UAVs based on state constraints according to claim 7, characterized in that, Select at least 3 ground control points with known geographical coordinates that are not on the same straight line, and transform each pixel coordinate of the aerial photography image to the geographical coordinate by minimizing the sum of the squared reprojection errors of all control points to generate an orthorectified image. Define the state vector X as: X = [X, Y, Z, V X ,V Y ,V Z ] T ; wherein X, Y, Z, V X ,V Y ,V Z represent the spatial position coordinate components and velocity components of the UAV respectively; Adopt the federated Kalman filter to fuse the inertial navigation data and the base station perception data to obtain the optimal estimated state of the drone, including: Within the adjacent two inertial navigation sampling intervals Δt, assume that the drone is moving at a constant speed and the speed remains unchanged, and determine the state transition matrix F; Based on the power spectral density q during the drone's movement, determine the process noise covariance matrix Q; Adopt the main filter and sub-filter structure measurement model. The first sub-filter processes the inertial navigation data and independently executes the time update and measurement update of the standard Kalman filter every time inertial navigation data is received; the second sub-filter processes the base station perception data and independently executes the time update and measurement update of the standard Kalman filter when the base station perception data is received. The two sub-filters use the same state transition matrix F and noise covariance matrix Q; 9. The method for real-time aerial photography data fusion and secure transmission of low-altitude UAVs based on state constraints according to claim 1, characterized in that, Through the main filter, the estimation results of the two sub-filters are weighted and combined according to the covariance to obtain the optimal estimated state of the drone. In step 6, the base station edge node calculates the drone position risk Rpos, speed risk Rvel, behavior risk Rbeh, and health risk Rhel at a set period, and evaluates the drone flight risk after fusion: The drone position risk Rpos is determined by the drone's current position and the no-fly zone (NFZ) boundary: if the drone's current position is within the NFZ boundary, then return Rpos=3; otherwise, traverse each edge of the NFZ polygon and calculate the shortest distance from a point to a line segment, taking the minimum value d as the shortest distance from the drone to the no-fly zone boundary. Rpos=0, d>100; Rpos=1, 50<d≤100; Rpos=2, 10<d≤50; Rpos=3, d≤10; The speed risk Rvel is determined based on the ratio of the current speed v to the area speed limit: Rvel=min(3,max(0,((v / v max )-1)×2)); Among them, v max This indicates the maximum permissible speed in the area as defined by the airspace classification. The behavioral risk Rbeh is set to predict the drone's trajectory over the next 5 seconds based on the PINN model to assess whether the current behavior is abnormal, specifically including: If any point in the predicted trajectory within the next 5 seconds enters a no-fly zone or is less than 5 meters away, Rbeh increases by 2; if the predicted speed exceeds 1.5v... max If the predicted acceleration exceeds a, then Rbeh increases by 1; max If so, Rbeh increases by 0.5; Finally, sum the three indices and truncate them to [0,3]. The health risk Rhel value is taken as the drone's health level H, with a range of [0,3]. Assess drone flight risk by integrating location risk (Rpos), speed risk (Rvel), behavioral risk (Rbeh), and health risk (Rhel). Risk=w1·Rpos+w2·Rvel+w3·Rbeh+w4·Rhel; Among them, w1, w2, w3, and w4 are the weight coefficients corresponding to the drone's location risk Rpos, speed risk Rvel, behavioral risk Rbeh, and health risk Rhel, respectively, with values ​​ranging from 0 to 1, and their sum is 1.

10. The method for real-time aerial photography data fusion and secure transmission of low-altitude UAVs based on state constraints according to claim 1, characterized in that, Step 6, which involves adaptive control and response based on risk values, includes: Based on the drone flight risk obtained from the fusion assessment, the risk level and corresponding response actions are determined: (1) Risk 0, <0.5, Action: Record only, no action taken; (2) Risk 1, 0.5~1.5, Action: Send a notification to the command center and do not interfere with the drone; (3) Risk 2, 1.5~2.5, Action to be taken: Send an alarm to the command center and the drone operator to alert them; (4) Risk 3, ≥2.5, Action: Send a forced return command via downlink control information and send it to the command center; Upon receiving a forced return command, the drone's flight control system immediately aborts the current mission and executes the preset return procedure.

Images