Unmanned aerial vehicle data security encryption and remote transmission management system and method, and medium

Abstract

The invention discloses an unmanned aerial vehicle data security encryption and remote transmission management system and method and a medium, and relates to the field of unmanned aerial vehicle data security protection and remote transmission. The intelligent sensing and edge computing module integrates multiple sensors, and adopts a heterogeneous architecture and a YOLOv8 model; the dynamic encryption processing module adopts SM4, chaos and SM9 cryptographic encryption, and supports three sensitive levels; the self-organizing destroy-resistant communication network module constructs a LORAMesh network and is matched with a phased-array antenna to switch an unmanned aerial vehicle cluster relay; the physical unclonable key management module uses PUF features and a block chain smart contract; the anti-interference receiving and intelligent reconstruction module is provided with a metamaterial antenna and 3D-CNN reconstruction; the active defense and situation awareness module deploys a GAN honeypot and CNN monitoring. The method is high in encryption speed, safe in key management, high in interference resistance, low in delay rate and capable of intelligently processing data, reinforcing hardware safety and reducing power consumption, and provides efficient and safe support for related fields.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) data security protection and remote transmission technology, and in particular to a UAV data security encryption and remote transmission management system, method and medium. Background Technology

[0002] The application of drones in fields such as power line inspection, emergency rescue, and geological exploration is becoming increasingly widespread. However, data security and remote transmission capabilities have become key bottlenecks restricting the industry's development. Existing drone systems are vulnerable to eavesdropping, tampering, or interference in complex electromagnetic environments. Traditional encryption algorithms, such as AES, consume significant computing resources on mobile terminals, resulting in insufficient real-time performance and difficulty in dealing with quantum computing threats. Simultaneously, issues such as low synchronization accuracy of multi-sensor data and poor stability of heterogeneous network transmission often lead to missing or distorted inspection data, affecting the accuracy of equipment status assessment. For example, in non-line-of-sight scenarios such as mountainous areas or urban canyons, communication links between drones and ground stations are prone to interruption. Traditional relay solutions rely on fixed infrastructure, lacking deployment flexibility and generally resulting in low data transmission success rates.

[0003] Furthermore, existing key management solutions mostly employ static keys or centralized distribution models, posing a high risk of key leakage and failing to meet the security requirements of dynamic drone flights. For example, when the inspection area spans different security level zones, the lack of an automatic key update mechanism makes it difficult to meet security standards such as the Cybersecurity Classified Protection 2.0. At the data processing level, weak edge computing capabilities necessitate the transmission of large amounts of raw data back to the cloud, increasing the transmission load and potentially causing missed fault warnings due to network latency. For instance, traditional drone detection of power line cracks relies on manual analysis, with the cycle from data acquisition to fault identification taking several hours, which cannot meet real-time maintenance needs.

[0004] Insufficient intelligence is also a significant shortcoming of existing technologies. Traditional inspection systems lack the ability to deeply fuse and analyze multimodal data, making it difficult to extract effective features from massive amounts of data. For example, the spatiotemporal registration error between optical images and millimeter-wave radar point clouds often exceeds 10 centimeters, resulting in insufficient defect location accuracy and a high rate of missed detections. At the same time, model training relies on a large amount of labeled data, has weak generalization ability in novel defect recognition scenarios, and requires frequent manual intervention, thus hindering the improvement of the automation level of UAV inspections. Summary of the Invention

[0005] The present invention proposes a UAV data security encryption and remote transmission management system, method and medium to solve the problems mentioned in the prior art.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a UAV data security encryption and remote transmission management system, method, and medium, comprising: Intelligent sensing and edge computing module: integrates inertial navigation system, optical camera, millimeter wave radar; adopts CPU+FPGA+NPU heterogeneous architecture, built-in YOLOv8 model to annotate defects in inspection images in real time, and compresses model size through knowledge distillation to adapt to UAV terminal; Dynamic encryption processing module: It uses the national standard SM4 algorithm to encrypt data in blocks, with a dynamic key generation cycle; it applies a chaotic encryption algorithm to images, supports three levels of sensitivity encryption, and superimposes SM9 identifier encryption on top-secret data; Self-organizing and resilient communication network module: Constructs a LORAMesh self-organizing network that supports concurrent nodes; relay nodes are equipped with phased array antennas, optimize paths through reinforcement learning, and automatically switch to UAV swarm relay when satellites are interrupted; Physically Unclonable Key Management Module: Utilizes RF device process deviations to generate PUF features as a key entropy source, manages the key lifecycle based on blockchain smart contracts, automatically updates the key when entering a sensitive area, and verifies it through zero-knowledge proofs; Anti-interference reception and intelligent reconstruction module: The ground station is equipped with a metamaterial anti-interference antenna and a blind source separation algorithm to recover data; 3D-CNN is applied to reconstruct encrypted point clouds and generate digital twin models in real time; Distributed secure storage and federated learning module: It adopts IPFS distributed storage, and sensitive data is divided into 128KB micro-blocks and encrypted with AES-256; federated learning supports edge-cloud collaborative training and homomorphic encrypted gradient updates; Active defense and situational awareness module: Deploys a GAN honeypot system to release fake links, monitors electromagnetic features in real time through CNN, and triggers adaptive frequency hopping and encryption protocol upgrades.

[0007] Furthermore, it also includes: Photonic crystal filter module: The optical camera lens integrates a photonic crystal filter to suppress sunlight interference and monitor the reflectivity of vegetation bands.

[0008] Furthermore, it also includes: Bayesian Network Intelligent Decision Module: Real-time analysis of link quality and drone energy consumption, dynamically adjusting compression ratio and transmission power.

[0009] Furthermore, the intelligent sensing and edge computing module is equipped with a spatiotemporal synchronization unit, BeiDou-3 short message timing, control of multi-sensor timestamp errors, and RTK-GNSS correction of spatial coordinates.

[0010] Furthermore, the dynamic encryption processing module is equipped with an adaptive encryption rate mechanism, using SM4 for text data and the NTRU algorithm for video streams.

[0011] Furthermore, the distributed secure storage and federated learning module employs data sandbox isolation technology to create an independent encrypted container for each task, prohibiting cross-task access and enabling audit log traceability.

[0012] Furthermore, it also includes: Hardware-level security hardening module: 3D printed metal-ceramic shell with built-in electromagnetic shielding layer; CNC precision-carved integrated PCB board with anti-detection design for key chips.

[0013] Furthermore, it also includes: Multidimensional data intelligent acquisition and feature enhancement steps: The UAV synchronously acquires 4K imagery, radar point cloud, and inertial navigation data along the flight path; the edge computing unit reconstructs the imagery and voxel filters the point cloud through ESRGAN to generate fused data. Dynamic key negotiation and hierarchical encryption steps: The UAV and the ground station negotiate the initial key through the Diffie-Hellman protocol, and generate the working key by combining PUF features; the data is hierarchically encrypted, with public level CRC-32 check, confidential level SM4-CTR, and top secret level SM9+ chaotic encryption; Anti-interference multipath transmission and intelligent relay steps: Data is fragmented using the MPTCP protocol and transmitted in parallel via satellite, 5G, and LoRa; relay nodes optimize paths using the Q-Learning algorithm for relaying drone swarms in non-line-of-sight scenarios; Joint decryption and security audit steps: The ground station reconstructs the key through Shamir threshold signature, and verifies the data consistency after SM4 decryption and MD5 verification; the audit system uses the Apriori algorithm to analyze access behavior, and abnormal operations trigger link blocking, generate evidence logs, and initiate the key destruction process.

[0014] Furthermore, multi-dimensional data acquisition incorporates GAN synthesis of rare scene data to optimize model generalization ability, and metamaterial reflectors are used to construct relay surfaces for interference-resistant transmission.

[0015] Furthermore, a computer-readable storage medium is characterized in that it stores a computer program, and the processor executes the program using CPU+GPU+FPGA heterogeneous acceleration.

[0016] Compared with existing technologies, the beneficial effects of this invention are: At the data security level, the dynamic encryption processing module employs the national standard SM4 algorithm combined with a chaotic encryption cascade mechanism to implement differentiated protection for data of different sensitivity levels. The encryption speed for confidential data reaches 500Mbps, a significant improvement over traditional solutions. Automated key lifecycle management is achieved through blockchain smart contracts, shortening the key update cycle and effectively resisting man-in-the-middle attacks and quantum computing threats. The physically unclonable key management module utilizes process deviations in radio frequency devices to generate a key entropy source, combined with zero-knowledge proof technology, elevating the security of the key generation and verification process to the hardware level, meeting the encryption requirements of high-security scenarios.

[0017] In terms of communication transmission, the self-organizing, resilient communication network improves data integrity in non-line-of-sight scenarios through LORAMesh relay collaboration with drone swarms. Multi-link aggregation technology enables dynamic switching and bandwidth optimization between satellite, 5G, and LORA, reducing transmission latency and ensuring priority transmission of real-time data such as fault warning information. The combination of metamaterial anti-interference antennas and blind source separation algorithms improves the signal-to-interference ratio, maintaining data recovery rates even in environments with strong electromagnetic interference, thus solving the communication interruption problem of traditional solutions in scenarios such as industrial areas and high-voltage power grids.

[0018] In terms of intelligence and energy efficiency optimization, the intelligent sensing and edge computing modules achieve a real-time inference speed of 60FPS for the YOLOv8 model through a heterogeneous computing architecture. Defect detection automatically identifies typical defects such as wear on transmission line hardware and cracks in insulators, improving detection efficiency. The application of federated learning and GAN data augmentation technologies improves monthly accuracy and reduces data annotation costs. Simultaneously, hardware-level security hardening and low-power design reduce the overall power consumption of the drone, extending continuous operation time. Combined with distributed storage and edge computing, this reduces cloud transmission load and improves endurance and real-time processing capabilities in complex scenarios.

[0019] In summary, this invention constructs a secure and reliable system covering the entire process of "collection-encryption-transmission-storage-analysis," which not only solves the core pain points of traditional UAVs in terms of data security, communication reliability, and intelligence level, but also provides efficient and secure technical support for intelligent operation and maintenance in the power, energy, and other fields, and has significant industrial application value and market competitiveness. Attached Figure Description

[0020] Figure 1 This is a schematic block diagram of a UAV data security encryption and remote transmission management system proposed in this invention; Figure 2 This is a schematic block diagram of a method for data security encryption and remote transmission management of unmanned aerial vehicles (UAVs) proposed in this invention. Figure 3 A schematic diagram illustrating the relationship between topology reconfiguration time and the number of nodes in a self-organizing network; Figure 4 A diagram illustrating the efficiency comparison of dynamic encryption rate adaptive mechanisms; Figure 5 A diagram showing the comparison of communication network transmission delay before and after optimization; Figure 6 This is a diagram illustrating the comparison of anti-interference capabilities; Figure 7 This is a diagram comparing energy consumption and efficiency. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," and "counterclockwise," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0023] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified. Furthermore, the terms "installed," "connected," and "linked" should be interpreted broadly; for example, they may refer to a fixed connection, a detachable connection, or an integral connection; they may refer to a mechanical connection or an electrical connection; they may refer to a direct connection or an indirect connection through an intermediate medium; and they may refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances. The invention will now be described in further detail with reference to the accompanying drawings.

[0024] Reference Figures 1 to 7 A data security encryption and remote transmission management system, method, and medium for unmanned aerial vehicles (UAVs) includes: Intelligent Sensing and Edge Computing Module: Integrates multi-source heterogeneous sensors and computing architecture to construct a unified edge node. The high-precision inertial navigation system employs MEMSIMU and satellite navigation fusion, outputting data at 100Hz. Kalman filtering corrects attitude angle errors to ≤0.1°, providing a spatial reference for inspection. The 8K optical camera integrates AI phase detection autofocus, and a depth estimation network combined with a focal length-resolution curve completes 0.5-100m depth-of-field focusing within 30ms, ensuring clear imaging in complex scenes. The 77GHz millimeter-wave radar uses FMCW technology, analyzing echoes using an FFT algorithm to achieve a point cloud density of 100 points / m². 2 It accurately perceives environmental contours. The heterogeneous computing architecture includes an ARM Cortex-A78 (2.8 GIPS), a Xilinx FPGA (latency ≤10ns), and a Cambricon NPU (128 TOPS). The FPGA embeds image feature extraction and other algorithms, improving efficiency by 10 times; the NPU accelerates YOLOv8 inference, enabling real-time defect annotation at 60 FPS. The model undergoes knowledge distillation and compression by 40% (to 480MB), and after transfer learning, crack detection accuracy is ≤0.1mm, with a false detection rate ≤1%. Multi-source data spatiotemporal synchronization error is ≤1ms, system power consumption is ≤15W, and battery life is improved by 25%, supporting scenarios such as UAV power line inspection and driving the upgrade of intelligent detection.

[0025] Dynamic encryption processing module: For image data (such as inspection images), it introduces Logistic chaotic mapping encryption, with the formula being x. n+1 =μ·x n ·(1−x n ), where x n Let μ be the chaotic sequence value of the nth iteration, μ be the chaotic control parameter, and n be the iteration number. During encryption, the pixel value is normalized and then XORed with the chaotic sequence. This leverages the initial value sensitivity and pseudo-randomness of the chaotic system to achieve anti-cracking capabilities. The algorithm achieves an encryption speed of 200MB / s with GPU acceleration, and an 8K image encryption latency of ≤20ms. It supports three sensitivity levels: ordinary data uses the national standard SM4 algorithm (128-bit key, throughput ≥500Mbps); confidential data is overlaid with RBAC tags; and top-secret data is forcibly encrypted with SM9 identifier encryption (256-bit key, based on the elliptic curve discrete logarithm problem) on top of SM4 or chaotic encryption. Through an encryption algorithm scheduler, an automatic matching strategy is performed based on data type (text / image) and sensitivity level, combined with a hardware acceleration engine (FPGA implementation of SM4 round function parallelism) to ensure low latency requirements for real-time data transmission (overall encryption latency ≤100μs).

[0026] Self-organizing, resilient communication network module: Constructs a LoRaMesh network architecture, with relay nodes integrating phased array antennas (beam switching time 10μs), and optimizes the transmission path using reinforcement learning, as shown in the formula. Where R(s,a) is the reward value for performing action a in state s, T(s,a) is the transmission delay, interference(s,a) is the interference intensity, and α and β are weighting coefficients. The reinforcement learning agent collects network status (delay, interference, node load) in real time and dynamically adjusts the beam direction and relay path, achieving a 35% reduction in transmission delay (compared to static paths). It supports 100+ concurrent nodes, with a single node covering 5km. When satellite communication is interrupted, it triggers drone swarm relay (D2D communication), with topology reconstruction time ≤5s, providing "wide coverage, high efficiency, and strong resilience" communication support for emergency rescue, field inspection, and other scenarios.

[0027] Physically Unclonable Key Management Module: This module utilizes manufacturing process variations in radio frequency devices (such as RFID chips and millimeter-wave transceivers) to generate Physically Unclonable Function (PUF) features. After stimulus-response acquisition and wavelet transform quantization, an 80-bit high-entropy feature is extracted. Uniqueness is ensured through a Challenge-Response Pair (CRP) mechanism; for the same device with different stimuli and different devices with the same stimuli, the Hamming distance is ≥64 bits and ≥40 bits respectively, giving the key entropy source resistance to physical cloning. Smart contracts are deployed on a consortium blockchain to automatically manage key generation, distribution, updating, and destruction. When a device enters a sensitive area via GPS positioning and geofencing, and the last key update is more than one minute old, the smart contract automatically triggers an update. Using a Key Derivation Function (KDF), combined with PUF features, blockchain random numbers, and an update counter, a new 256-bit session key is derived and synchronized to network nodes via the PBFT consensus mechanism, ensuring key consistency. Zero-knowledge proof (zk-SNARKs) technology is used to verify key legitimacy, with a processing time of ≤100ms. The device generates proof based on the current key and PUF features. The verifier only needs to use the pre-registered PUF reference features to complete the legitimacy verification without obtaining the complete key, which is both efficient and ensures the "zero-knowledge" security attribute.

[0028] Anti-interference reception and intelligent reconstruction module: Integrating metamaterial antennas, blind source separation, and 3D-CNN technology, an anti-interference and intelligent reconstruction system is constructed. The ground station is equipped with a metamaterial anti-interference antenna, and the interference suppression characteristics are described by the transmission matrix, with the formula S(ω)=M(ω)⋅S free (ω), where S(ω) are the antenna scattering parameters after loading the metamaterial, and M(ω) is the transmission matrix of the metamaterial element. free(ω) represents the scattering parameters of the antenna in free space. The metamaterial unit adopts an "open resonant ring + metal wire" structure. By adjusting the unit size, phase cancellation and amplitude attenuation of interference signals are achieved, suppressing bandwidth coverage of 2GHz (e.g., 1GHz-3GHz) and ensuring the quality of useful signal reception. Independent component analysis (ICA) is used to separate blind sources, recover the interfered data, and improve the signal-to-interference ratio by 25dB. For UAV communication scenarios, interfered images and telemetry data can be separated. For encrypted point clouds, 3D-CNN is used to achieve intelligent reconstruction with an error ≤2%. The 3D-CNN adopts an encoder-decoder architecture, where the encoder extracts the spatial features of the point cloud, and the decoder reconstructs a high-precision point cloud.

[0029] Distributed Secure Storage and Federated Learning Module: This module integrates decentralized storage, cryptography, and collaborative learning technologies to build a data security system. It adopts the IPFS distributed storage architecture, dividing sensitive data into 128KB micro-blocks, encrypting them with AES-256-GCM, adding authentication tags and initialization vectors to ensure integrity, and storing them with redundancy level 3 on geographically dispersed, credit-filtered IPFS nodes. A Distributed Hash Table (DHT) is used to maintain storage locations, mitigating the risks of data leakage and single points of failure. In the federated learning phase, it supports edge-cloud collaborative training. Homomorphic encryption based on the RLWE problem is used to encrypt model gradients, achieving "data usable but not visible," with computational efficiency loss controlled within 15%. Through optimization using the FedAvg algorithm, high-quality data devices are selected for participation, the learning rate is dynamically adjusted, and L2 regularization is introduced to suppress overfitting. Model accuracy is improved by 12% compared to plaintext centralized training, adapting to sensitive data scenarios such as the Industrial Internet and smart healthcare, helping to break down "data silos" and achieve secure collaborative training.

[0030] Active defense and situational awareness module: Deploys GAN honeypots to lure attackers, with the generation probability derived from the adversarial game formula. , where P decoy Let G(z) be the probability that a link is identified as a real link, G(z) be the pseudo-link feature generated by the generator, and G(·) be the discrimination probability. The GAN honeypot contains a generator and a discriminator. The generator, based on the Transformer architecture, learns the features of real network links and generates pseudo-links (with a cosine similarity ≥ 0.9 to real links). The discriminator uses a CNN classifier with a false positive rate ≤ 10%. Electromagnetic environment features are monitored in real time using CNN, achieving an eavesdropping detection rate ≥ 95%. The CNN model uses a ResNet-18 architecture and, with GPU acceleration (inference speed ≥ 100 frames / s), enables real-time monitoring of eavesdropping behavior. Upon detection of eavesdropping, adaptive frequency hopping and encryption protocol upgrades are triggered, with a frequency hopping rate of 1000 hops / s and a protocol upgrade from SM4 to SM9. The combination of frequency hopping and protocol upgrades increases the difficulty for attackers to crack the code by 10%.30 times.

[0031] This invention also includes the following modules: Photonic Crystal Filter Module: An integrated photonic crystal filter is used in the optical camera lens to construct a spectral modulation mechanism. Addressing the optical needs of vegetation monitoring scenarios, it achieves dual optimization of solar interference suppression and vegetation signal enhancement. Based on photonic bandgap theory, the filter utilizes a periodic arrangement of silicon dioxide and silicon (refractive indices 1.46 and 3.42, respectively) to construct a one-dimensional photonic crystal structure (20 layers). Calculations using the transfer matrix method show that the thicknesses of the high and low refractive index layers are approximately 39 nm and 90 nm, respectively, resulting in high transmission (≥95%) in the 532 nm band and precise control of the full width at half maximum (FWHM) at 5 nm. This effectively suppresses interference in the solar spectrum, such as blue light (400-500 nm) and red light (700-800 nm). The filter is fabricated using a vacuum coating and precision alignment process. Electron beam evaporation coating is performed on a quartz glass substrate (thickness 0.5mm, flatness ≤λ / 10, λ=632.8nm). The film thickness is monitored in real time by a crystal oscillator (accuracy ±0.1nm) and in-situ annealing is performed to eliminate internal stress. After embedding the lens spacer (aluminum alloy material, blackened surface treatment), the optical axis is ensured to coincide by a three-coordinate alignment platform (positioning accuracy ±5μm). After integration, the effective focal length change of the lens is ≤0.5%, and the distortion rate increase is ≤0.1%. The spectral control effect is ensured by calibration with a spectral analyzer. In vegetation monitoring scenarios, multiple optimizations were achieved to suppress sunlight interference, increase the reflectivity of blue light in the 400-500nm range to 80%, and reduce the transmittance of infrared light in the 700-800nm ​​range to 10%; enhance vegetation reflectivity, increasing the transmittance of the lens in the 532nm band from 80% to 95%, and increasing the intensity of vegetation reflected light signal by 30%; improve the image signal-to-noise ratio, increasing it from 20dB to 38dB at noon on a sunny day, and reducing noise particles by 70%.

[0032] This invention also includes the following modules: Bayesian Network Intelligent Decision Module: Constructs link quality assessment through multi-dimensional data collection. A broadband RF front-end (2.4 / 5.8GHz dual-band) integrates an LNA (20dB gain, 1.5dB noise figure), samples RSSI at 100Hz, and uses a moving average filter (window 10) to ensure an error ≤ ±2dB. Based on CRC-32 checksum, it calculates the bit error rate in real time (accuracy ±0.01%, latency ≤100ms). A Bayesian network link quality node is constructed, using RSSI and bit error rate as input, and outputs a quality level (Excellent, Good, Medium, Poor) through prior probability and CPT inference, achieving an accuracy ≥92% and an inference latency ≤50ms. Battery parameters are collected through the BMS, the EKF algorithm estimates the SOC (error ≤3%), and the remaining battery power is predicted by fusing LSTM and Bayesian networks (error ≤5%), dynamically adjusting communication parameters and load power consumption. Based on Bayesian inference, resources are dynamically scheduled. Inputting link quality, remaining battery power, and task priority, the compression ratio is intelligently adjusted (selectable from 5:1 to 20:1, adjustment latency ≤200ms, accuracy ≥95%). Combined with a path loss model to predict communication distance, the transmission power is dynamically adjusted (10mW-1W, step size ≤10mW, response ≤100ms) to ensure effective communication over 50km. After each adjustment, the effect is verified through link reassessment and energy consumption re-prediction to ensure adaptability. In actual tests, RSSI error ≤±2dB, bit error rate accuracy ±0.01%, remaining battery power prediction error ≤5%, communication power consumption reduced by 20%-30% in complex environments, and bit error rate ≤0.1% for critical tasks, driving the upgrade of UAV communication towards "dynamic intelligent adaptation".

[0033] In this invention, the intelligent sensing and edge computing module is equipped with a spatiotemporal synchronization unit, integrating BeiDou-3 short message timing and RTK-GNSS positioning to construct a high-precision spatiotemporal reference. BeiDou timing is achieved by a dual-mode receiver (B1I and B2a frequencies) with a high-gain antenna capturing signals, amplified by an LNA, demodulated by a baseband chip, and using a two-way time transfer model to integrate carrier phase and pseudorange measurements. Combined with ionospheric and tropospheric corrections, the calculated time deviation accuracy is ≤50ns. The local high-stability crystal oscillator is tamed, and the TCXO's hold-over mode is activated when satellites are blocked, ensuring a deviation ≤100ns within one hour, and guaranteeing a multi-sensor timestamp error <1μs. The RTK-GNSS uses a rover-base station architecture. The base station broadcasts RTCM3.2 differential data (1Hz), and the rover uses the LAMBDA algorithm to resolve carrier phase ambiguity (success rate ≥99.5%). Combined with ephemeris and differential data, the horizontal accuracy reaches 2cm and the vertical accuracy 5cm. Coordinates are synchronized to the sensors via PPS pulses (rising edge accuracy ≤50ns), and the local coordinate system is transformed using the Bursa-Taylor seven-parameter transformation, with multi-sensor position deviation ≤1cm. The unit ensures spatiotemporal consistency through timestamp synchronization (PPS-triggered acquisition, time alignment error ≤500ns) and spatial coordinate and attitude fusion (EKF correction, camera field of view error ≤0.1°). The monitoring module collects errors in real time, automatically calibrating when thresholds are exceeded (time >1μs, space >5cm). In actual measurements, the BeiDou timing accuracy is 50ns, and the multi-sensor time error is <1μs; the RTK horizontal accuracy is 2cm, and the spatial deviation is ≤1cm. Even in complex environments, it maintains time accuracy ≤100ns and spatial accuracy ≤5cm, supporting the upgrade of IoT sensing to "precise spatiotemporal collaboration" and empowering scenarios such as smart water conservancy.

[0034] In this invention, the dynamic encryption processing module employs an adaptive encryption rate mechanism. Text encryption uses SM4, accelerated by FPGA hardware, with an 8-stage pipeline processing 64-bit data blocks in parallel, theoretically achieving a throughput of 1Gbps. Combined with PCIe 3.0 x8 bandwidth and memory latency optimization, it stably reaches 500Mbps. CCM mode is adopted, encrypting while verifying integrity, with a latency ≤50μs, and supporting dynamic key updates (time / data volume dual factors). DPI identifies the text stream, prioritizing hardware resource scheduling, and initiating software fallback when exceeding a threshold, with a scheduling latency ≤1ms. Video stream encryption uses NTRU, with the parameter set cropped to N=251, reducing latency to 10ms and ensuring quantum-resistant security. For the video NAL unit, a block-by-block XOR operation is used in stream cipher mode, achieving a throughput of 200Mbps, adapting to 1080P@30fps. A traffic classifier identifies the video stream, using polling + priority queue scheduling, triggering QoS degradation during congestion, ensuring a latency ≤10ms. Overall efficiency improvements are achieved through business classification and resource pooling. DPI+SVM classification achieves an accuracy of ≥99.5% and latency of ≤500μs. A hardware resource pool is built, and EDF algorithm scheduling reduces allocation latency to ≤2ms. Compared to traditional solutions, text throughput is increased by 5 times, video latency is reduced by 80%, and overall efficiency is increased by 45%. Real-world testing with multiple concurrent services shows text latency ≤50μs and video latency ≤10ms, with resource utilization increasing by 30%. This achieves a balance between security, real-time performance, and efficiency, driving the upgrade of encryption technology towards "customization and adaptability."

[0035] In this invention, the distributed secure storage and federated learning modules employ data sandbox isolation technology. When creating independent encrypted containers, a dedicated runtime environment is built for each task, leveraging the six Linux namespaces. Seccomp restricts system calls while granting necessary permissions; eBPF dynamically monitors processes, resulting in an abnormal behavior trigger rate of less than 0.1%. The storage layer uses LUKS encryption paired with the AES-256-XTS algorithm. Keys are managed throughout the entire lifecycle by the Hardware Security Module (HSM), and hardware acceleration via the kernel CryptoAPI achieves an encryption throughput exceeding 500MBps and a latency ≤100μs. Third-party testing shows a cross-container memory and network access success rate ≤0.1%, with a stable isolation level of 99.9%. Cross-task access control is based on the ABAC model. A unique UUID is generated upon task startup, and an access token is obtained via the OAuth2.0 protocol. The policy engine analyzes the matching degree between task and resource attributes in real time, directly intercepting illegal cross-task access (matching degree ≤10%). Policy updates utilize the GitOps process, with an effective latency ≤100ms. Each access generates an audit log, recording nanosecond-level timestamps, task identity, and other information. Unauthorized access triggers an alarm, notifying the administrator within 1 second. Audit tracing relies on the PTPv2 protocol, with server-to-server time synchronization accuracy ≤100ns. The kernel uses the RDTSC instruction to record nanosecond-level operation timestamps (error ≤50ns). Logs are stored in a consortium blockchain, with the BFT-SMRT consensus algorithm ensuring integrity and a 100% tamper detection rate. The tracing platform supports searching by task UUID, time, and other dimensions, visually reconstructing operation sequences, analyzing latency ≤100ms, and accurately locating data operation nodes. Multi-task parallel testing (using 100 encrypted containers) verifies a 99.9% cross-container access interception rate, a policy matching accuracy ≥99.9%, and controllable nanosecond-level timestamp errors, providing precise evidence for data security incidents. This supports multi-tenant scenarios in cloud computing and edge computing, driving data isolation from "coarse partitioning" to "task-level precise protection."

[0036] This invention also includes the following modules: Hardware-level safety hardening module: 3D-printed metal-ceramic shell, using a 7:3 metal (aluminum alloy) and ceramic (alumina) composite, printed by SLM and hot-pressed (600℃, 50MPa), eliminating porosity (≤0.5%), with a thermal conductivity of 1.5W / m・K. Honeycomb internal structure (2mm pore size, 0.5mm wall thickness) with thermally conductive silicone (3W / m・K), increasing heat dissipation area by 30%, operating temperature ≤60℃. Anodizing (10μm layer thickness) provides corrosion resistance, salt spray test ≥480h, SLM printing accuracy ±0.1mm, five-axis CNC precision machining (Ra≤0.8μm) ensures compatibility. Built-in triple shielding of "metal mesh + absorbing material + metal foil": copper mesh (200 mesh, 0.1mm), ferrite (0.5mm, absorption rate ≥90%), and aluminum foil (0.05mm), bonded with conductive adhesive. Shielding ≥60dB for 10kHz-1GHz, ≥80dB for 1GHz-18GHz. Embedded molding ensures bonding strength (peeling ≥5N / mm), and grounding resistance ≤10mΩ. The PCB is an 8-layer HDI, Any-layer interconnect, with trace width / spacing ≤30μm / 30μm, 1000Trace / mm, and SI simulation optimized differential pairs (impedance ±5%). CNC precision-carved (accuracy ±10μm) irregular holes, plasma cleaning for contamination removal, thermal cycling (-40℃-85℃, 500 cycles), and vibration (5-2000Hz, 30g) testing, with solder joint failure ≤1% and wiring breakage ≤0.1%. The key chip (BGA package) features a self-destruct mechanism with anti-detection. A 10μm detection circuit is integrated around the substrate perimeter. Probe contact triggers capacitive / resistive change (≥10%). Within 100ns, the bridge wire melts the gold wire (25μm) or releases hydrofluoric acid microcapsules (50μm) to etch the active layer. After self-destruction, all functions are lost and data recovery is difficult. A security register is linked to the main controller for backup destruction, ensuring protection in extreme security scenarios. Testing and verification: The casing withstands impact ≥50J; tamper-proof opening ≤1 time; electromagnetic shielding at high frequencies ≥80dB; chip self-destruct response <100ns; normal function in harsh environments ≥99%, adapting to high-security scenarios and achieving a leap from "passive protection" to "active defense."

[0037] This invention also includes the following steps: Multidimensional data intelligent acquisition and feature enhancement steps: The UAV synchronously acquires 4K imagery (60fps), radar point cloud (100,000 points / s) and inertial navigation data along the flight path; the edge computing unit reconstructs the imagery (magnified 4 times) and voxel-filtered point cloud (95% retention rate) through ESRGAN to generate fused data with semantic tags (such as "tower-crack-coordinates").

[0038] Dynamic key negotiation and hierarchical encryption steps: The UAV and the ground station negotiate the initial key through the Diffie-Hellman protocol and generate the working key by combining PUF features; Data hierarchical encryption: public level CRC-32 check, confidential level SM4-CTR (key updated every 10 minutes), top secret level SM9+ chaotic encryption (independent key for each packet).

[0039] Anti-interference multipath transmission and intelligent relay steps: Data is fragmented using the MPTCP protocol (1MB per fragment) and transmitted in parallel via satellite (priority 1), 5G (priority 2), and LoRa (priority 3); relay nodes optimize paths using the Q-Learning algorithm (path loss reduced by 28%), and relay drone clusters in non-line-of-sight scenarios (delay 50ms).

[0040] Joint decryption and security audit steps: The ground station reconstructs the key through Shamir threshold signature, and verifies the data consistency after SM4 decryption and MD5 verification; the audit system analyzes access behavior using the Apriori algorithm, and abnormal operation triggers: link blocking (response 10ms), generation of evidence log, and initiation of key destruction process.

[0041] In this invention, multidimensional data acquisition utilizes Conditional Generative Adversarial Networks (CGANs) to synthesize rare scene data (such as rainstorms). The generator, based on U-Net, generates 256×256 rainstorm images through 5 layers of deconvolution and spectral normalization. The discriminator uses PatchGAN for block-by-block discrimination, optimized by Wasserstein distance. The conditional constraint module incorporates meteorological parameters to generate multidimensional rainstorm variant data (intensity, raindrop morphology, etc.), expanding to the 100,000 level. The synthesized data is mixed with real monitoring data (3:7), and the model is fine-tuned through transfer learning. The measured classification accuracy is improved by 22%, and the regression generalization error is reduced by 17% (rainfall prediction from 0.8mm to 0.66mm). To prevent interference during transmission, a metamaterial reflector is introduced, with the unit being a "metal resonator + dielectric substrate" (10mm×10mm period), a cross-shaped slit (0.2mm wide, 3mm deep), and the substrate selected is Rogers4350B. By leveraging geometric diffraction and genetic algorithms, phase adjustment is achieved using 100×100 units to realize a broadband bandwidth of 2.4-5.8GHz with a reflection efficiency of 90%. A 2m×2m relay plane (40,000+ units) with a tilt adjustment mechanism adapts to moving targets. In urban canyon testing, coverage expanded from 1km to 3km, with an intensity increase of 15dB (-85dBm to -70dBm); under multipath interference, narrowband / broadband interference suppression ratios reached 25dB / 18dB. The synergy of these two technologies, with GAN supplementing data and metamaterials ensuring transmission, improves response by 30% and task efficiency by 25% in scenarios such as intelligent transportation, contributing to reliable "edge-cloud" operation and promoting intelligent perception's adaptation to complex environments.

[0042] In this invention, the computer-readable storage medium adopts an NVMe-over-Fabrics architecture, integrating a 3D NAND flash array (10TB QLC) with a custom controller. Modular program partitioning storage (SLC for the core library, QLC for the cache) and dual-copy redundancy + CRC-64 checksum ensure reliability. The program is layered according to "algorithm-scheduling-driver," supporting UAV data security encryption and remote transmission management methods, and leveraging OpenCL to achieve heterogeneous acceleration via CPU+GPU+FPGA. Task splitting and mapping are based on computational characteristics. The CPU (Xeon Platinum) manages serial control (e.g., data sharding), the GPU (A100) uses CUDA for parallel encryption (1024 channels, throughput 10GB / s), and the FPGA has embedded logic circuits (round functions, etc., latency ≤10ns). Greedy algorithm + performance prediction dynamic scheduling, synchronization barriers ensure timing, and collaborative latency ≤1μs. Single-frame encryption latency <20ms, broken down into read (≤2ms), computation (≤15ms), and write-back (≤3ms), with a stable 17ms latency in 4K image testing. Power consumption optimization relies on DVFS (reducing GPU frequency and shutting down FPGA for idle operation under low load) and low-power storage (standby ≤5W). The average power consumption in 24-hour testing is 120W (traditional 170W), a 30% reduction, making it suitable for scenarios such as real-time encryption and edge computing, and providing an efficient environment for computationally intensive tasks.

[0043] I. Dynamic Encryption Processing Module Level 3 Sensitivity Level Classification Standard Public level: Drone location, flight status, and non-sensitive environmental imagery; Confidential: Power line inspection defects, pipeline coordinates, internal operation data; Top Secret: Images of classified areas, 3D point clouds of key facilities, keys and control commands.

[0044] Complete Implementation Parameters of Chaotic Encryption Logistic chaotic mapping: μ=4.0, initial value x0∈(0,1) and x0≠0.25, 0.5, 0.75; Number of iterations: ≥200, discard the first 100 iteration values ​​to eliminate transients; Pixel encryption: 8-bit image pixels are normalized to [0,1], and then XORed point by point with the chaotic sequence and then reversed and normalized.

[0045] Automatic matching of encryption policies and rules Text / command data: SM4-CTR mode, 128-bit key, updated every 10 minutes; Image / point cloud data: Chaotic encryption + SM4; Video stream: NTRU lattice cipher, parameter set N=251, q=128, p=3; Top Secret Data: SM9 identifier encryption + SM4 + chaotic three-level encryption.

[0046] Key dynamic generation threshold Time-triggered: Forced update after 15 minutes of continuous operation; Data volume trigger: Forced update when cumulative encrypted data ≥ 10GB; Location trigger: Update immediately upon entering a sensitive area of ​​the electronic fence.

[0047] II. Physically Unclonable Key Management Module Complete PUF Feature Extraction Process 64 sets of random stimuli were applied to the radio frequency chip, and the responses were collected. After db4 wavelet transform, the level 3 detail coefficients are quantized into an 80-bit binary string; Hamming distance constraint: different excitations on the same chip ≥ 64 bits, and the same excitations on different chips ≥ 40 bits.

[0048] Blockchain key management rules Smart contract trigger conditions: GPS location enters a sensitive area + more than 60 seconds since the last update; Key derivation: Working key = KDF(PUF feature + blockchain random number + update counter); Consensus mechanism: PBFT, effective when the number of nodes is ≥4 and the number of confirmations is ≥2f+1.

[0049] Zero-knowledge proof verification process The device generates zk-SNARKs proofs using the current key and PUF features; The verification end performs verification using pre-stored PUF reference features, without obtaining the plaintext key; Verification time ≤ 100ms, link disconnected immediately if verification fails.

[0050] III. Self-organizing and resilient communication network module LoRa Mesh networking parameters Frequency band: 470MHz-510MHz, bandwidth 125kHz, spreading factor SF=7-12; Single node coverage: 5km, supports ≥100 concurrent nodes; Topology reconfiguration: Switch to drone swarm relay within 5 seconds of satellite interruption.

[0051] Reinforcement learning path optimization parameters State space S: Delay, interference intensity, remaining node power, link quality; Operation space A: Beam pointing, relay node selection, transmit power; Reward function: R = α*(1 / T) - β*I, α = 0.7, β = 0.3; The learning rate is 0.01, the discount factor is 0.9, and the Q-table update cycle is 100ms.

[0052] Multi-link parallel transmission rules Link priority: Satellite > 5G > LoRa; MPTCP fragmentation: 1MB per fragment, dynamically allocated according to link quality; Non-line-of-sight scenarios: Enable drone D2D relay, one-hop latency ≤50ms.

[0053] IV. Anti-interference reception and intelligent reconstruction module Details of Metamaterial Antenna Implementation Unit structure: open resonant ring + metal wire, unit size 10mm×10mm; Interference suppression bandwidth: 1GHz-3GHz, suppression ratio ≥25dB; Power supply method: microstrip line power divider, beam switching ≤10μs.

[0054] Blind source separation and point cloud reconstruction Blind source separation algorithm: FastICA, number of iterations ≥ 1000, convergence threshold 1e-6; 3D-CNN reconstruction: Encoder-Decoder architecture, 3×3×3 convolutional kernels, stride 1; Reconstruction error: ≤2%, output point cloud density ≥100 points / m².

[0055] V. Intelligent Sensing and Edge Computing Module Heterogeneous computing task partitioning CPU: System control, data scheduling, protocol parsing; FPGA: Image feature extraction, SM4 hardware acceleration, latency ≤10ns; NPU: YOLOv8 inference, computing power 128 TOPS, frame rate 60 FPS.

[0056] YOLOv8 Compression and Deployment Knowledge distillation: Teacher model YOLOv8x, student model YOLOv8n; Compression ratio: 40%, model size: 480MB; Inspection targets: insulator cracks, hardware wear, and missing vibration dampers; accuracy ≤0.1mm; false detection rate ≤1%.

[0057] Spatiotemporal synchronization parameters BeiDou-3 timing: accuracy ≤50ns, multi-sensor timestamp error <1μs; RTK-GNSS: Horizontal deviation 2cm, elevation deviation 5cm, position deviation ≤1cm.

[0058] VI. Active Defense and Situation Awareness Module GAN honeypot implementation rules Generator: Transformer architecture, inputting real-world link features; Discriminator: ResNet18, false positive rate ≤10%; Pseudo-link characteristics: Cosine similarity to real links ≥ 0.9.

[0059] Anomaly detection and response CNN Electromagnetic Monitoring: Eavesdropping detection rate ≥95%, inference speed ≥100 frames / s; Trigger response: Frequency hopping 1000 hops / s, encryption protocol upgraded from SM4 to SM9; The attack difficulty has increased by 10. 30 times.

[0060] VII. Distributed Secure Storage and Federated Learning Module IPFS storage rules Data blocks: 128KB / block, encryption algorithm AES-256-GCM; Redundancy: 3 replicas; storage nodes are selected based on credit scores. Access control: ABAC permission model, isolated by task UUID.

[0061] Federated learning implementation Gradient encryption: Homomorphic encryption (RLWE), efficiency loss ≤15%; Optimization algorithm: FedAvg, L2 regularization coefficient 0.001; Model accuracy: 12% improvement compared to intensive training.

[0062] VIII. Complete and executable method and process Multidimensional data acquisition Simultaneous acquisition: 4K video at 60fps, radar point cloud at 100,000 points / s, inertial navigation at 100Hz; Image enhancement: ESRGAN 4x super-resolution; Point cloud processing: Voxel filtering at 0.1m resolution, with a retention rate of 95%.

[0063] Dynamic key negotiation DH protocol to negotiate the initial key; A 256-bit working key is generated by combining PUF features; Hierarchical encryption: Public level CRC32, Confidential level SM4, Top Secret level SM9+Chaos.

[0064] Joint decryption and auditing Key reconstruction: Shamir threshold (3,5), any 3 copies can be recovered; Auditing algorithm: Apriori, minimum support 0.05, minimum confidence 0.8; Abnormal response: If the chain is interrupted within 10ms, a blockchain evidence log will be generated and the key will be destroyed.

[0065] IX. Hardware-level security hardening implementation details Structure and shielding Shell: Aluminum alloy + alumina 7:3, SLM printing, porosity ≤0.5%; Electromagnetic shielding: ≥60dB for 10kHz-1GHz, ≥80dB for 1GHz-18GHz; Grounding resistance: ≤10mΩ.

[0066] Chip anti-detection self-destruct Detection ring: 10μm metal ring, probe contact trigger impedance change ≥10%; Response time: <100ns, if the gold wire is melted / the chip is etched, the data is unrecoverable.

[0067] 10. Test Methods and Verification Basis Encryption rate: FPGA-accelerated SM4, 500Mbps@CCM mode; Latency reduction: Hybrid link reduces latency by 35% compared to satellite-only transmission; Signal-to-interference ratio: Improved by 25dB with metamaterial antenna and blind source separation; Test environment: temperature -40℃~+60℃, electromagnetic interference 10V / m, outdoor non-line-of-sight.

[0068] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A data security encryption and remote transmission management system for unmanned aerial vehicles (UAVs), characterized in that, Includes the following modules: Intelligent sensing and edge computing module: integrates inertial navigation system, optical camera, millimeter wave radar; adopts CPU+FPGA+NPU heterogeneous architecture, built-in YOLOv8 model to annotate defects in inspection images in real time, and compresses model size through knowledge distillation to adapt to UAV terminal; Dynamic encryption processing module: It uses the national standard SM4 algorithm to encrypt data in blocks, with a dynamic key generation cycle; it applies a chaotic encryption algorithm to images, supports three levels of sensitivity encryption, and superimposes SM9 identifier encryption on top-secret data; Self-organizing and resilient communication network module: Constructs a LORAMesh self-organizing network that supports concurrent nodes; relay nodes are equipped with phased array antennas, optimize paths through reinforcement learning, and automatically switch to UAV swarm relay when satellites are interrupted; Physically Unclonable Key Management Module: Utilizes RF device process deviations to generate PUF features as a key entropy source, manages the key lifecycle based on blockchain smart contracts, automatically updates the key when entering a sensitive area, and verifies it through zero-knowledge proofs; Anti-interference reception and intelligent reconstruction module: The ground station is equipped with a metamaterial anti-interference antenna and a blind source separation algorithm to recover data; 3D-CNN is applied to reconstruct encrypted point clouds and generate digital twin models in real time; Distributed secure storage and federated learning module: It adopts IPFS distributed storage, and sensitive data is divided into 128KB micro-blocks and encrypted with AES-256; federated learning supports edge-cloud collaborative training and homomorphic encrypted gradient updates; Active defense and situational awareness module: Deploys a GAN honeypot system to release fake links, monitors electromagnetic features in real time through CNN, and triggers adaptive frequency hopping and encryption protocol upgrades.

2. The UAV data security encryption and remote transmission management system according to claim 1, characterized in that, Also includes: Photonic crystal filter module: The optical camera lens integrates a photonic crystal filter to suppress sunlight interference and monitor the reflectivity of vegetation bands.

3. The UAV data security encryption and remote transmission management system according to claim 1, characterized in that, Also includes: Bayesian Network Intelligent Decision Module: Real-time analysis of link quality and drone energy consumption, dynamically adjusting compression ratio and transmission power.

4. The UAV data security encryption and remote transmission management system according to claim 1, characterized in that, The intelligent sensing and edge computing module is equipped with a spatiotemporal synchronization unit, BeiDou-3 short message timing, control of multi-sensor timestamp errors, and RTK-GNSS correction of spatial coordinates.

5. The UAV data security encryption and remote transmission management system according to claim 1, characterized in that, The dynamic encryption processing module has an adaptive encryption rate mechanism, using SM4 for text data and NTRU algorithm for video streams.

6. The UAV data security encryption and remote transmission management system according to claim 1, characterized in that, The distributed secure storage and federated learning module employs data sandbox isolation technology, creating an independent encrypted container for each task, prohibiting cross-task access, and enabling audit log traceability.

7. The UAV data security encryption and remote transmission management system according to claim 1, characterized in that, Also includes: Hardware-level security hardening module: 3D-printed metal-ceramic shell with built-in electromagnetic shielding layer; CNC precision-carved integrated PCB board, key chip anti-detection design.

8. A method for secure encryption and remote transmission management of UAV data based on the system described in any one of claims 1-7, characterized in that, Includes the following steps: Multidimensional data intelligent acquisition and feature enhancement steps: The UAV synchronously acquires 4K imagery, radar point cloud, and inertial navigation data along the flight path; the edge computing unit reconstructs the imagery and voxel filters the point cloud through ESRGAN to generate fused data. Dynamic key negotiation and hierarchical encryption steps: The UAV and the ground station negotiate the initial key through the Diffie-Hellman protocol, and generate the working key by combining PUF features; the data is hierarchically encrypted, with public level CRC-32 check, confidential level SM4-CTR, and top secret level SM9+ chaotic encryption; Anti-interference multipath transmission and intelligent relay steps: Data is fragmented using the MPTCP protocol and transmitted in parallel via satellite, 5G, and LoRa; relay nodes optimize paths using the Q-Learning algorithm for relaying drone swarms in non-line-of-sight scenarios; Joint decryption and security audit steps: The ground station reconstructs the key through Shamir threshold signature, and verifies the data consistency after SM4 decryption and MD5 verification; the audit system uses the Apriori algorithm to analyze access behavior, and abnormal operations trigger link blocking, generate evidence logs, and initiate the key destruction process.

9. The method for secure encryption and remote transmission management of UAV data according to claim 8, characterized in that, Multidimensional data acquisition introduces GAN to synthesize rare scene data, optimizes the model's generalization ability, and uses metamaterial reflectors to construct relay surfaces for interference-resistant transmission.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the method described in any one of claims 8-9; the processor executes the program with the aid of CPU+GPU+FPGA heterogeneous acceleration.

Images