Unmanned aerial vehicle flight control method and system

Abstract

The invention belongs to the technical field of unmanned aerial vehicle flight control, and relates to an unmanned aerial vehicle flight control method and system. The method comprises the following steps: collecting a terahertz signal reflected by each electronic device in a target space and a test signal of multiple frequencies reflected by each electronic device; performing feature extraction on the terahertz signal to obtain radio frequency feature information of the electronic equipment; establishing an impedance spectrum characteristic matrix according to the test signals of the multiple frequencies; identifying the civil electronic equipment according to the radio frequency characteristic information and the impedance spectrum characteristic matrix; real-time positioning of the civil electronic equipment is obtained; generating a virtual potential field vector of the civil electronic equipment according to the real-time positioning; and controlling the flight attitude of the unmanned aerial vehicle according to the virtual potential field vector, so that the unmanned aerial vehicle responds to the equipment position change in real time to optimize the path for obstacle avoidance.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of unmanned aerial vehicle flight control, and particularly relates to an unmanned aerial vehicle flight control method and system. BACKGROUND

[0002] With the wide application of unmanned aerial vehicle technology in civil, industrial and public service fields, its autonomous flight control and safety obstacle avoidance capability in complex environments have become a core technical pain point. Especially in flight tasks involving densely populated areas, how to accurately identify the range of personnel activities and achieve safe avoidance is directly related to flight safety and task compliance.

[0003] Traditional unmanned aerial vehicle obstacle avoidance technology mainly relies on single sensing means such as visual sensors, laser radars or ultrasonic waves, and realizes path planning by constructing an environment three-dimensional model. However, such methods have significant limitations in complex scenes: visual sensors are easily affected by light changes, weather interference (such as fog, haze, dust), and obstructions, and are easily affected by visual blind areas in building shadow areas, night lightless environments or dense building groups, resulting in a significant decrease in the sensing accuracy of personnel activity areas; laser radars have certain penetration ability, but have low identification resolution for non-metallic objects, and are prone to misjudgment in multi-target mixed scenes, making it difficult to accurately distinguish between personnel and non-personnel targets.

[0004] In terms of personnel feature recognition, existing technologies are mostly based on image feature analysis, and personnel recognition is achieved by extracting human body contours, behavior postures and other visual features. However, such methods have two major problems: one is the risk of privacy protection, and high-resolution image acquisition easily involves the leakage of personal privacy information; the other is the lack of recognition robustness, and human features are easily disturbed by factors such as clothing obstruction, group gathering and posture changes, resulting in an identification accuracy of less than 60% in complex environments, which is difficult to meet the high-precision obstacle avoidance demand. At the same time, traditional methods mostly rely on static maps or pre-set no-fly zones for path constraint, and cannot respond to dynamic personnel movement in real time, which is prone to obstacle avoidance lag in dynamic personnel activity area scenes (such as rallies and community activities). SUMMARY

[0005] The purpose of the present application is to provide an unmanned aerial vehicle flight control method and system, which breaks through the environmental limitations and recognition blind areas of traditional sensing technology, constructs a personnel recognition mechanism that takes into account recognition accuracy, real-time performance and anti-interference performance, and deeply integrates it with unmanned aerial vehicle flight control to form an automated control link that responds dynamically.

[0006] The present application is realized by the following technical solutions: In a first aspect, a method for controlling the flight of a UAV is provided, comprising the following steps: collecting terahertz signals reflected by each electronic device in a target space and test signals of multiple frequencies; extracting features of the terahertz signals to obtain radio frequency feature information of the electronic device; establishing an impedance spectrum feature matrix according to the test signals of the multiple frequencies; identifying a civilian electronic device according to the radio frequency feature information and the impedance spectrum feature matrix; obtaining real-time positioning of the civilian electronic device; generating a virtual potential field vector of the civilian electronic device according to the real-time positioning; and controlling the flight attitude of the UAV according to the virtual potential field vector.

[0007] Further, the method further comprises the following steps: generating an entangled photon pair containing a probe photon and a reference photon; emitting the probe photon to the target space; collecting the probe photon reflected by each electronic device in the target space; obtaining a quantum state correlation degree between the reference photon and the reflected probe photon; marking noise quantum states with a quantum state correlation degree less than a correlation degree threshold in the reflected probe photon; generating an entangled state signal orthogonal to the noise quantum states; and using the entangled state signal to cancel noise signals in the terahertz signals.

[0008] In a second aspect, a UAV flight control system is provided, comprising: a first signal acquisition module for acquiring terahertz signals reflected by each electronic device in a target space; a second signal acquisition module for acquiring test signals of multiple frequencies reflected by each electronic device in the target space; a feature information extraction module for extracting features of the terahertz signals to obtain radio frequency feature information of the electronic device; a feature matrix establishment module for establishing an impedance spectrum feature matrix according to the test signals of the multiple frequencies; a civilian device identification module for identifying a civilian electronic device according to the radio frequency feature information and the impedance spectrum feature matrix; a real-time positioning acquisition module for obtaining real-time positioning of the civilian electronic device; a potential field vector acquisition module for generating a virtual potential field vector of the civilian electronic device according to the real-time positioning; and a flight attitude control module for controlling the flight attitude of the UAV according to the virtual potential field vector.

[0009] Compared with the prior art, the present application has the following advantages and beneficial effects: 1、The present application is based on the popularity of current civilian electronic equipment, and adopts the mode of identifying the characteristics of the electromagnetic signals reflected by the civilian electronic equipment to indirectly realize personnel identification, thereby avoiding the limitations of the prior art through image feature identification. Specifically, by synchronously collecting the reflection data of the terahertz signal and the multi-frequency test signal, the sensitivity of the terahertz signal to the hardware radiation characteristics is used to extract the equipment radio frequency characteristics, and the impedance spectrum feature matrix constructed by the multi-frequency reflection is combined to cross verify the equipment physical properties, forming a dual identification mechanism of "radio frequency characteristics + impedance spectrum characteristics", which not only enhances the anti-interference ability of complex environment by virtue of the strong penetration of terahertz signal and the multi-frequency noise reduction ability of impedance spectrum, but also distinguishes civilian electronic equipment from non-civilian equipment more accurately; on this basis, the position of the civilian electronic equipment is located in real time, and the position is strongly associated with the personnel activity area, thereby breaking through the visual blind area to realize dynamic perception of the core activity area of the personnel; further, a virtual potential field vector is generated based on the positioning data, the personnel associated area is converted into a dynamic "repulsive force field", and the unmanned aerial vehicle relies on the vector to realize millisecond-level smooth adjustment of the flight attitude, forming an automatic closed loop link of "signal perception-feature identification-positioning analysis-potential field decision-attitude control"; this process not only reduces the single feature identification misjudgment rate, but also can respond to the equipment position change in real time to optimize the path obstacle avoidance.

[0010] 2, The quantum state correlation of entangled photon pairs is used to construct a "detection-reference-correlation analysis" noise identification mechanism. By emitting detection photons to the target space and collecting reflected photons, the quantum state correlation degree is calculated combined with reference photons, which can accurately mark the noise quantum state with a correlation degree lower than the threshold value - this process relies on the inherent correlation of quantum state, breaks through the dependence of traditional noise filtering on signal intensity and frequency characteristics, and can distinguish effective signals (associated photons reflected by civilian electronic equipment) from environmental noise (such as electromagnetic interference and random scattered photons) from the quantum level, realizing accurate positioning and marking of noise. On this basis, an entangled state signal orthogonal to the noise quantum state is generated to cancel the noise, and by virtue of the physical characteristics of quantum state orthogonality, the noise can be targetedly eliminated without damaging the effective terahertz signal. This quantum domain noise cancellation method is more efficient than traditional filtering technology, can maximize the retention of device radio frequency feature information contained in the terahertz signal, solves the problem of effective signal being overwhelmed by noise in complex electromagnetic environment (such as mixed multi-device signals and strong interference area), and significantly improves the clarity and integrity of subsequent radio frequency feature extraction. BRIEF DESCRIPTION OF DRAWINGS

[0011] The drawings described herein are used to provide further understanding of the embodiments of the present application, constitute a part of the present application, and do not constitute a limitation on the embodiments of the present application. In the drawings: Figure 1 An unmanned aerial vehicle flight control process schematic diagram is provided for the embodiment 1 of the present application.

[0012] Figure 2 A virtual potential field vector visualization schematic diagram provided for the embodiment 1 of the present application.

[0013] Figure 3 A vector superposition generated target flight direction visualization schematic diagram of the unmanned aerial vehicle provided for the embodiment 1 of the present application. DETAILED DESCRIPTION

[0014] In order to make the objects, technical solutions and advantages of the present application clearer, further detailed description will be made to the present application combined with embodiments, the illustrative embodiments of the present application and the description thereof are only used to explain the present application, and do not limit the present application. The following described embodiments are part of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0015] In the following description, a large number of specific details are set forth in order to provide a thorough understanding of the present application. However, it is apparent to those skilled in the art that the present application can be implemented without the specific details. In other embodiments, in order to avoid confusion of the present application, well-known structures, materials or methods are not specifically described. The materials, instruments and reagents used in the following embodiments, etc. can be obtained from commercial channels if not otherwise specified. The technical means used in the embodiments, if not otherwise specified, are conventional means known to those skilled in the art.

[0016] In addition, the terms "first", "second" are only used for description purpose, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features. Therefore, the features limited by "first", "second" can explicitly or implicitly include one or more features. In the description of the present application, the meaning of "multiple" is two or more, unless otherwise specifically limited.

[0017] Embodiment 1: provide a method for controlling the flight of an unmanned aerial vehicle, comprising Figure 1 The following steps are shown: Step 1: collect the terahertz signal reflected by each electronic device in the target space and the test signal of multiple frequencies.

[0018] With the acceleration of urbanization and the popularity of consumer electronics, smart phones, tablets, smart watches and other consumer electronics have become an essential part of modern residents' daily life. Their distribution characteristics are strongly correlated with the areas of personnel activities. On the one hand, when the internal circuit of an electronic device is working, its chips, antennas, capacitors and other components will naturally generate electromagnetic signals inherent to the device due to current changes. Terahertz radar uses the penetration of 0.3-3 THz frequency band, which can directly penetrate the non-metallic shell (such as plastic, glass) of the electronic device, accurately collect these original electromagnetic signals generated inside the electronic device, and the terahertz signals reflected by the electronic device carry the signal characteristics inside the electronic device, which are collected by the terahertz radar, creating conditions for obtaining the real signal characteristics inside the device.

[0019] On the other hand, when the wireless transmission module of an electronic device (such as the radio frequency antenna of a mobile phone or the communication module of a smart watch) is working, the impedance of its circuit system will change uniquely with factors such as working frequency, transmission power and component characteristics. Impedance is the ratio of voltage to current in a circuit, reflecting the impedance and energy conversion characteristics of the circuit, and is a physical property inherent to the device hardware. This method uses a vector network analyzer (parameter index: 100 kHz-20 GHz frequency band, impedance measurement accuracy ±0.5%) to collect the impedance characteristics of the wireless transmission module of the electronic device. The vector network analyzer transmits test signals of different frequencies to the electronic device, calculates the impedance values (including resistance and reactance components) at different frequencies according to the reflection and transmission parameters of the measured signals in the module, and finally generates the impedance change curve of the device within a wide frequency band, i.e. "impedance spectrum".

[0020] Therefore, before this step, the terahertz signal and the test signal of multiple frequencies need to be continuously transmitted from multiple directions in the target space to detect the electronic device in the target space.

[0021] Step 2: Eliminate environmental electromagnetic noise interference in the terahertz signal through quantum entangled state signal detection.

[0022] The reflected terahertz signal of an electronic device is usually weak (≤-110 dBm) and can be easily overwhelmed by environmental noise in a complex electromagnetic environment. To eliminate noise interference in weak signals, this embodiment achieves the following steps: Step 2.1: Generate an entangled photon pair containing a probe photon and a reference photon, and transmit the probe photon to the target space.

[0023] Quantum entanglement refers to the "non-local correlation" of the quantum state formed by two or more quantum particles (such as photons) - when the state of one particle is measured, the state of the other particle is instantly determined, and is not affected by distance or environmental interference. The present embodiment generates entangled photon pairs through the quantum radar module, emits one of the photons as a "probe photon" into the target environment, and retains the other photon as a "reference photon". Electromagnetic noise in the environment (such as interference signals from industrial equipment and 5G base stations) belongs to classical electromagnetic signals, and its mechanism of action is to interfere with the probe signal through energy superposition, but it cannot affect the correlation of the quantum entangled state - that is, noise can only interfere with the state of the probe photon, but cannot change the original entanglement relationship between the reference photon and the probe photon. Therefore, even if the probe photon is contaminated by noise during transmission, the quantum state of the reference photon remains "pure", providing an absolute benchmark for distinguishing between "true signals" and "noise interference" in the future.

[0024] Step 2.2: Collect the reflected probe photons from each electronic device in the target space, obtain the quantum state deviation between the reference photon and the reflected probe photon, and mark the noise quantum state with a quantum state deviation > deviation threshold in the reflected probe photon.

[0025] After the probe photon interacts with the target device (or environment), it carries the target signal characteristics (such as phase shift and polarization change caused by terahertz band signals emitted by the circuit) back, while superimposing environmental noise. Based on the correlation of quantum entanglement, the quantum state of the reference photon and the probe photon should strictly match when there is no interference. However, the quantum state deviation caused by noise has random and irregular characteristics (classical noise cannot form stable correlation), while the true signal of the electronic device affects the probe photon through electromagnetic interaction, and the deviation caused by it presents regularity related to the device hardware (such as specific phase patterns emitted by the chip). Therefore, by calculating the quantum state deviation (such as phase difference fluctuation or polarization angle) between the probe photon and the reference photon, we obtain the quantum state deviation caused by the true signal of the target device and the quantum state deviation caused by noise interference.

[0026] Further, the reference photon and the returned probe photon are jointly measured by the quantum state analyzer, and the quantum state deviation between them is calculated. If the quantum state with a quantum state deviation > deviation threshold is the noise quantum state. For example, calculate the phase difference band of the reference photon and the returned probe photon, if the phase difference band of a certain signal in the returned probe photon > 2°, then this signal can be determined as being interfered by environmental noise.

[0027] Step 2.3: Generate an entangled state signal orthogonal to the noise quantum state, and use the entangled state signal to cancel the noise signal in the terahertz signal.

[0028] The electromagnetic noise in the environment can be regarded as a "noise quantum state" with specific quantum state characteristics at the quantum level. This step generates an entangled state signal orthogonal to the noise quantum state through a quantum entangled state generator. In quantum mechanics, "orthogonal" means that there is no overlapping component between the two quantum states, and they are independent of each other and cannot interfere with each other. For example, if the polarization direction of the noise quantum state is horizontal, the polarization direction of the generated entangled state signal is vertical (orthogonal relationship); if the noise quantum state has a specific phase distribution, the entangled state signal is formed by phase modulation to form a phase mode orthogonal to it. This orthogonal characteristic ensures that the newly generated entangled state signal will not be "polluted" by noise, providing a "clean" reference basis for subsequent signal purification.

[0029] The collected terahertz signal and the generated orthogonal entangled state signal are subjected to quantum interference measurement, and the orthogonal principle is used to make the noise quantum state and the entangled state signal cancel each other out, while the quantum state of the real signal is retained because it is not orthogonal to the entangled state.

[0030] Step 3: Extracting features of the terahertz signal to obtain the radio frequency characteristic information of the electronic device.

[0031] Using the signal processing and visualization functions of the terahertz radar, the terahertz signal collected by the terahertz radar is converted into a two-dimensional frequency spectrum image (horizontal axis is frequency, vertical axis is signal intensity). The two-dimensional frequency spectrum image is input into the trained ResNet model. The ResNet model extracts features from the frequency spectrum image layer by layer through multiple residual convolution layers. The shallow network captures basic morphological features such as peak and valley of the frequency spectrum, and the deep network excavates subtle fluctuations of the frequency spectrum. These features correspond to the model of the chip, the material and structure of the antenna, the wiring method of the circuit board, and other hardware physical properties.

[0032] By processing and identifying the terahertz band signal reflected by the internal circuit of the electronic device, unique and stable characteristic information of the device can be extracted, and the collection of these characteristic information constitutes the "radio frequency fingerprint" of the device. Just like everyone's fingerprint has uniqueness and can be used for identity recognition, the "radio frequency fingerprint" of the electronic device also has uniqueness and stability. The "radio frequency fingerprint" generated based on terahertz waves can accurately distinguish different electronic devices. Even if they are the same type of device, due to factors such as production process and use wear, their "radio frequency fingerprints" will also be different. In the unmanned aerial vehicle personnel identification scene, the unmanned aerial vehicle can determine whether the electronic device is a civilian electronic device by identifying the "radio frequency fingerprint" of the electronic device, thereby providing reliable identification basis for subsequent positioning and obstacle avoidance decision-making.

[0033] Before that, the ResNet model needs to be trained, which is the key to realizing the feature extraction of terahertz signals. By training the ResNet model, the residual learning advantage is effectively utilized to achieve high-precision recognition in tasks such as terahertz spectrum image classification, while having strong generalization ability and being able to adapt to equipment identification needs in complex environments. The method for training the ResNet model is: 1. Constructing a dataset (1) Data source: Collect samples (such as terahertz spectrum images), covering all categories involved in the task (such as mobile phones, watches, earphones), with a sample size of ≥3000 for each category to ensure class balance (avoiding a certain class of samples accounting for more than 70%).

[0034] (2) Data annotation: Label the samples (such as "mobile phone - model A", "watch - model B"), which can be annotated manually combined with automated tools (such as Label Studio) to improve efficiency.

[0035] (3) Data division: Divide into training set (model learning), validation set (parameter tuning), and test set (final evaluation) in the ratio of 7:2:1, and keep the class distribution consistent (such as mobile phones accounting for 60% in the training set, and the validation set and test set should also be close to 60%).

[0036] 2. Construction and configuration of ResNet model Select the appropriate ResNet architecture (such as ResNet-18 / 34 / 50) according to the task requirements and make targeted adjustments.

[0037] (1) Model selection Since this method needs to identify the fine-grained spectral features of electronic devices from terahertz signals, ResNet-50 / 101 architecture is selected to extract subtle features using a deeper network.

[0038] (2) Network structure adaptation Input layer: Change the original 3-channel input of ResNet to 1-channel for processing single-channel two-dimensional spectrum images, and adjust the parameters of the first convolutional layer (in_channels=1).

[0039] Output layer: Modify the output dimension of the last fully connected layer according to the number of electronic device categories (such as 3 for 3 categories), and use the Softmax activation function to output class probabilities.

[0040] Residual block optimization: Use the "1x1 convolution dimension reduction → 3x3 convolution feature extraction → 1x1 convolution dimension increase" structure for key residual blocks, and use shortcut connection (jump connection) to alleviate gradient disappearance and ensure that the deep network can be trained.

[0041] (3) Initialization parameters Convolutional layer weights are initialized with Kaiming Normalization (adapted to ReLU activation function), and fully connected layer weights are initialized with Xavier initialization; the gamma of Batch Normalization layer is set to 1 and the beta is set to 0 to ensure the stability of the initial distribution.

[0042] 3. Training parameters and strategy settings (1) Core parameter configuration Optimizer: Prefer Adam (adaptive learning rate), initial learning rate is set to 0.001, and weight decay (L2 regularization) is set to 1e-5 (to suppress overfitting).

[0043] Learning rate scheduling: Use cosine annealing or piecewise decay (such as halving the learning rate every 30 epochs), to avoid oscillation caused by high learning rate or slow convergence caused by low learning rate.

[0044] Batch Size: Set according to GPU memory (e.g. 32 for 12GB memory), too small will cause large gradient fluctuations, too large will occupy too many resources.

[0045] Training rounds (Epoch): Initially set to 100-200, combined with early stopping mechanism (e.g. stop if validation accuracy does not improve for 10 consecutive rounds).

[0046] (2) Loss function Similarly, based on fine-grained recognition, this embodiment selects the Center Loss function to narrow the distance between features of the same class.

[0047] 4. Training model (1) Training process Forward propagation: input enhanced samples, extract features through convolutional layers and residual blocks, and finally output prediction results through fully connected layers.

[0048] Backpropagation: Calculate the loss between the predicted value and the true label, update the network parameters through gradient descent (e.g. Adam), and update the mean and variance of the Batch Normalization layer.

[0049] Iterative optimization: evaluate performance on the validation set after each training round, and adjust the learning rate or early stop based on the validation results.

[0050] (2) Key indicator monitoring Loss: The training set Loss should gradually decrease and stabilize, if the fluctuation is too large, check BatchSize or learning rate; Accuracy: The accuracy of the training set and the validation set should be improved simultaneously. If the accuracy of the validation set decreases, overfitting may occur, and regularization should be strengthened. Confusion Matrix: Analyze misclassified samples (e.g., mistaking earphones for watches) and supplement samples or adjust the model accordingly.

[0051] 5. Model Evaluation and Optimization (1) Test Set Evaluation After training, evaluate the model performance on an independent test set. Key indicators include: Accuracy: The overall classification accuracy should be ≥95% (complex tasks can be appropriately reduced). Precision and Recall: Ensure the reliability of identifying each device (e.g., earphone recall rate ≥90%). Robustness Test: Verify the model's stability in interference scenarios such as noise and occlusion (performance degradation should be ≤10%).

[0052] (2) Model Optimization For classes with high error rates, supplement samples for fine-tuning (learning rate reduced to 1 / 10 of the initial value). Use Transfer Learning: Initialize weights based on a pre-trained ResNet (e.g., a model trained on ImageNet), only adjust the output layer and some residual blocks, accelerate convergence, and improve performance.

[0053] Step 4: Establish an impedance spectrum feature matrix based on multiple frequency test signals.

[0054] Step 4.1: Obtain the impedance value of each test signal.

[0055] Based on the explanation in Step 1, use a vector network analyzer to collect the impedance characteristics of the wireless transmission module of the electronic device. The vector network analyzer is an instrument that can accurately measure the characteristics of radio frequency and microwave circuits within a wide frequency range. In this embodiment, its working frequency range is 100kHz-20GHz, and the impedance measurement accuracy is ±0.5%. When working, it sends test signals of different frequencies to the wireless transmission module of the electronic device, measures the reflected and transmitted signals, calculates the impedance values of the electronic device at each frequency point, and presents these impedance values as a curve with respect to frequency, i.e., the impedance variation curve. Due to differences in internal circuit design, component characteristics, etc., the impedance variation curves of different devices are different. For example, a mobile phone and a smart watch will have significantly different impedance variation trends and values within the same frequency range due to differences in functionality and circuit complexity.

[0056] Step 4.2: Organize all impedance values into a matrix form to obtain the impedance spectrum feature matrix.

[0057] The impedance values collected at different frequencies are arranged in a matrix form to construct an impedance spectrum feature matrix. The arrangement of the matrix can be: each row represents a frequency point, the elements in a row correspond to the impedance values (including resistance, reactance, etc.) of the frequency point, each column represents an electronic device, and the elements in a column correspond to the impedance values of the electronic device at different frequencies. The impedance spectrum feature matrix comprehensively records the impedance characteristics of each electronic device within a specific frequency range, and is a kind of "electrical fingerprint" of the electronic device. For example, by measuring multiple mobile phones of the same model and arranging their impedance data into an impedance spectrum feature matrix, the common characteristics of the impedance characteristics of the mobile phones of the same model can be clearly seen, and subtle differences between individuals can also be found.

[0058] Step 5: Identify the civilian electronic device according to the radio frequency feature information and the impedance spectrum feature matrix.

[0059] This embodiment uses a support vector machine (SVM) to identify the civilian electronic device from the radio frequency feature information and the impedance spectrum feature matrix. The radio frequency signal features and impedance spectrum features of different devices (such as mobile phones, routers, microwave ovens, and Bluetooth earphones) differ significantly, and these differences constitute the "discriminative features" for classification. By fusing the radio frequency and impedance spectrum features, the SVM can learn the joint distribution rules of the two types of features, and then distinguish different device types. The specific method is as follows: Similarly, before using the SVM model to classify and identify the civilian electronic device, the SVM model needs to be trained. The method for training and optimizing the SVM model is as follows: 1. Data collection and preprocessing (1) Radio frequency signal collection Use a spectrum analyzer, software-defined radio (such as USRP, HackRF) to collect the radio frequency signals of the device when it is working, and record the time domain or frequency domain data (the sampling rate needs to cover the working frequency band of the device, such as 2.4GHz, 5GHz, etc. civilian frequency band). Remove environmental noise through filtering (such as low-pass filtering), and extract the effective signal of the device working period (avoid Idle state interference) through signal interception.

[0060] (2) Impedance spectrum data collection Use an impedance analyzer to measure the impedance spectrum of the device within a specific frequency range (such as 1kHz~100MHz), and obtain the matrix of impedance modulus and phase with respect to frequency. Preprocessing: remove measurement noise through smoothing processing (such as moving average), and normalize the impedance spectrum (eliminate device individual differences or measurement environment influence).

[0061] 2. Feature extraction and fusion Extract high-discriminative features from the preprocessed raw data and perform multi-modal fusion.

[0062] (1) RF feature extraction The features of the RF signal need to reflect the "communication / working characteristics" of the device, and the key features include: Frequency domain features: center frequency, bandwidth, peak frequency, power spectral density (PSD), spectral entropy (measuring the uniformity of spectral distribution), harmonic components, etc. (extracted by Fourier transform, short-time Fourier transform (STFT)).

[0063] Modulation domain features: modulation method (such as ASK, FSK, PSK, QAM), symbol rate, modulation depth, etc. (extracted by constellation analysis, cyclic spectral density).

[0064] Time domain features: signal duration, pulse width, frequency hopping period (for frequency hopping devices), envelope peak value, etc.

[0065] (2) Impedance spectrum feature extraction The impedance spectrum reflects the "circuit characteristics" of the device, and the key features include: Characteristic frequency point parameters: resonance frequency (frequency point with minimum / maximum impedance modulus), cutoff frequency, impedance modulus and phase at characteristic frequency.

[0066] Trend features: slope of impedance modulus with frequency (such as difference in slope between low and high frequency bands), phase change rate, curvature extreme points of impedance spectrum curve.

[0067] Statistical features: mean, variance, peak number, energy concentration area of impedance spectrum (quantified by integral or entropy value).

[0068] (3) Feature fusion The RF features and impedance spectrum features are concatenated into a joint feature vector, which is used as the input of the SVM: Assuming the RF feature dimension is D1, the impedance spectrum feature dimension is D2, and the fused feature dimension is D1+D2.

[0069] Feature selection: select features strongly related to device type by ANOVA, MI, or RFE, and remove redundant features (reduce dimensionality to avoid overfitting).

[0070] 3. Feature preprocessing To improve the classification effect of SVM, the fused features need to be standardized.

[0071] Normalization: map the feature values to the [0, 1] interval, suitable for impedance spectrum and other features with clear ranges.

[0072] Standardization: convert the features to a distribution with mean 0 and variance 1, suitable for RF power and other features with large dynamic range.

[0073] 4. SVM model training and optimization (1) Data set division Divide the samples into a training set (70%), a validation set (15%), and a test set (15%), wherein: the training set is used for SVM model parameter learning; the validation set is used for hyperparameter tuning (such as kernel function selection, regularization coefficient); and the test set is used to evaluate the final generalization ability of the model.

[0074] (2) Selection of SVM core parameters The performance of SVM is highly dependent on parameters, and the following parameters need to be optimized: Kernel function: select RBF kernel (radial basis function), which is suitable for nonlinear feature scenarios (RF and impedance spectrum features usually have nonlinear correlation), and is the preferred kernel function for civilian device identification.

[0075] Regularization coefficient (C): controls the model complexity and fault tolerance, the larger C is, the stronger the model fitting on the training set (easier to overfit); the smaller C is, the stronger the fault tolerance (easier to underfit).

[0076] Kernel parameter (γ): for RBF kernel, controls the sample influence range, the larger γ is, the smaller the single sample influence range (the more complex the model).

[0077] Train the model on the training set based on the optimal parameters, and evaluate the model stability through cross-validation (such as 5-fold cross-validation).

[0078] Through accurate collection of RF and impedance spectrum data, extraction of discriminative features with physical meaning, and combination of the advantages of SVM kernel function and parameter tuning, efficient classification of different civilian electronic devices is realized.

[0079] Step 6: Obtain the real-time positioning of the civilian electronic device.

[0080] This embodiment uses the triangular positioning method to obtain the real-time positioning of the civilian electronic device. The specific method is: 1. Obtain the signal collection point position Extract the coordinates of three signal collection points from the GPS positioning data, denoted as A ( x A , y A ), B ( x B , y B ) and C ( x C , y C )。

[0081] 2. Obtain the distance between the target electronic device and the signal collection points According to the time when the terahertz imaging radar emits terahertz waves t 1. And the time when the terahertz waves reflected by the electronic device are collected t 2. Calculate the distance from the target electronic device to each collection point d A 、 d B and d C The distance calculation formula is d = c ×( t 2- t 1) / 2, c where c is the speed of light.

[0082] 3. Solve the coordinates of the target electronic device Given the coordinates of three collection points A ( x A , y A ), B ( x B , y B ) and C ( x C , y C ) and the distance from the target electronic device to each collection point d A 、 d B and d C , establish the equation group to solve the coordinates of the target electronic device x , y ): (1) Establish the distance equation The expression of the distance equation is: .

[0083] (2) Linearization solution Square both sides of the equation group and eliminate the quadratic terms to convert it into a linear equation group: , Solve the linear equation group by matrix operation to get the coordinates of the target electronic device x , y .

[0084] Step 7: Generate a virtual potential field vector of the civil electronic device according to real-time positioning.

[0085] In the concept of "force interaction" in the physical field, the position of the electronic device carried by the person is abstracted as a "repulsive field source", and a virtual "virtual potential field vector" is constructed around the UAV. The position of each identified civilian electronic device is regarded as a field source, and the "repulsive force" of the field source changes with the distance, that is, the closer the distance, the stronger the repulsive force, and the farther the distance, the weaker the repulsive force (similar to the intensity distribution law of electric field or magnetic field); the vector direction of each point in the vector diagram represents the direction of the resultant force on the UAV at that position, and the vector size represents the intensity of the resultant force. Through this visualization method, the complex obstacle avoidance decision is converted into the calculation of "force balance and direction", and the area to be avoided by the UAV and the optimal moving path are intuitively reflected.

[0086] Based on the concept of force interaction, the generation method of the virtual potential field vector of the civilian electronic device is: Step 7.1: Establish a three-dimensional coordinate system of the target space with the signal collection point as the origin.

[0087] Step 7.2: According to the real-time positioning, mark the coordinate points of each civilian electronic device in the three-dimensional coordinate system.

[0088] Step 7.3: Obtain the vector direction and vector size of each coordinate point.

[0089] Wherein, the vector direction is the direction of the coordinate point pointing to the origin, and the vector size is the reciprocal of the distance of the coordinate point from the origin. After visualization processing, the virtual potential field vector established is as shown in Figure 2 . Figure 2 Wherein, M point, N point and P point respectively represent the real-time positions of the civilian electronic devices in the target space, and the collection point is located at the origin O of the coordinate system; the vector directions of M point, N point and P point are represented by dashed arrows; 1 / d M is the vector size of M point, that is, the repulsive force of M point on the collection point O; 1 / d N is the vector size of N point, that is, the repulsive force of N point on the collection point O; 1 / d P is the vector size of P point, that is, the repulsive force of P point on the collection point O.

[0090] Step 8: Control the flight attitude of the UAV according to the virtual potential field vector.

[0091] Step 8.1: Superimpose the vector size and vector direction of all coordinate points to obtain the target flight direction of the UAV.

[0092] When there are multiple targets, the total repulsive force on the UAV is the vector superposition of the repulsive forces of all field sources, and the safest obstacle avoidance path is determined by calculating the resultant force direction. For reference, Figure 3First, the vector of point M is superimposed with the vector of point N. In this embodiment, point N and its vector arrow are translated to obtain point N' and its vector arrow. After translation, they are superimposed with the vector of point M (drawing the direction and magnitude of the resultant force). The superimposed vector is Figure 3 Then, the dotted arrow is translated and superimposed with the vector of point P to finally obtain the target flight direction of the UAV. Figure 3 Indicated by the solid arrows.

[0093] Step 8.2: Control the flight attitude of the drone based on its current flight direction and target flight direction.

[0094] The core of drone flight attitude control is to dynamically adjust the aircraft's attitude (pitch, roll, and yaw) by comparing the deviation between the current flight direction and the target flight direction, ensuring that the drone flies stably and accurately along the target trajectory. In conjunction with the obstacle avoidance scenario of civilian electronic equipment, the specific method and steps are as follows: 1. Flight direction data collection and analysis (1) Core parameter definition Analyze the real-time motion direction of the drone into a three-dimensional direction vector through sensor data , where: the horizontal direction (xy plane) reflects the flight direction of the drone projected on the ground (such as 30° north by east); the vertical direction (z axis) reflects the climbing or descending trend (positive for ascending, negative for descending). The target flight direction of the drone is defined as a three-dimensional vector .

[0095] (2) The accelerometer and gyroscope are used to measure the angular rate and acceleration of the aircraft in real time, and the current attitude angle (pitch angle θ, roll angle φ, yaw angle ψ) is analyzed.

[0096] 2. Calculation of direction angle deviation (1) Calculate the yaw deviation (horizontal deviation). Calculate the difference between the current direction and the target direction. xy Angle between planes ,like >180° or <-180°, the deviation range is normalized to [-180°, 180°] by adding or subtracting 360°, which facilitates control direction selection (such as direct steering for small angle deviations and step-by-step adjustment for large angle deviations).

[0097] (2) Calculate the pitch deviation (vertical deviation). z Deviation in the axial direction A positive value indicates that the current climb angle is insufficient and the pitch angle needs to be increased; a negative value indicates that the pitch angle needs to be decreased (or increased for descent).

[0098] (3) Set the deviation threshold Set the maximum allowable deviation threshold (e.g. horizontal deviation ≤ 5°, vertical deviation ≤ 3°), when the deviation is less than the threshold, it is determined that the UAV is flying stably in the target direction and does not need to be adjusted significantly; when the threshold is exceeded, the attitude control algorithm is triggered and the control amount is dynamically adjusted according to the deviation (the larger the deviation, the larger the adjustment).

[0099] 3. Flight attitude control Calculate the required attitude adjustment amount based on the deviation, and convert it to motor control signals through the flight control system.

[0100] (1) Proportional-Integral-Derivative (PID) control algorithm.

[0101] PID structure is a classic closed-loop control algorithm framework in the field of industrial control, full name is Proportional-Integral-Derivative control. It combines the current error, historical cumulative error and error trend of the system to calculate the control amount to adjust the controlled object, so that the system output is stable at the target value.

[0102] Take horizontal yaw control as an example, the core formula is: . Where, Kp is the proportional term coefficient, which directly outputs the control amount according to the current deviation (e.g. 5% motor differential speed when the deviation is 5°), quickly reducing the deviation; Ki is the integral term coefficient, which accumulates long-term small deviations (e.g. 2° deviation for 2°), eliminates static error and ensures that the deviation tends to 0 in steady state; Kd is the derivative term coefficient, which suppresses overshoot according to the rate of change of the deviation (e.g. reduces speed in advance when the deviation increases rapidly), avoiding violent shaking. Kp, Ki, Kd Need to be determined according to the characteristics of the system.

[0103] The vertical pitch control uses the PID structure, and the pitch angle PID parameters are adjusted separately (usually Kp less than the yaw control to avoid vertical oscillation).

[0104] (2) Attitude angle and motor output mapping Convert the control amount output by PID into specific attitude angle adjustment instructions.

[0105] Yaw angle adjustment: achieved by left and right motor differential speed (e.g. left motor speed up, right motor speed down, UAV turns clockwise), the differential speed is directly proportional to , and the maximum turning angle speed is limited to 30° / s (to avoid body shaking).

[0106] Pitch angle adjustment: change the pitch of the body by differential speed of front and rear motors (e.g. slow down the front motor and speed up the rear motor, the body will tilt up and increase the pitch angle, which will increase the lift and realize climbing), the pitch angle adjustment range is limited to [-10°, 15°] (out of range will easily lose control).

[0107] Roll angle assistance: small angle deviation (<5°) in the horizontal direction can be smoothly steered by fine-tuning the roll angle (±3°), reducing the energy consumption and shock of yaw adjustment.

[0108] (3) Dynamic parameter self-adaptation When flying at low speed (<5m / s), reduce the proportional coefficient of PID Kp , reduce the adjustment sensitivity and ensure stability.

[0109] When flying at high speed (>10m / s), increase Kp and Kd , improve response speed and avoid deviation accumulation.

[0110] In the emergency scenario of obstacle avoidance, temporarily increase the response priority of the deviation threshold, allow larger amplitude of attitude adjustment (such as the maximum yaw angle speed is increased to 50° / s), and quickly avoid the target.

[0111] Embodiment 2: corresponding to embodiment 1, the embodiment provides a flight control system of an unmanned aerial vehicle, comprising: A first signal acquisition module is configured to acquire terahertz signals reflected by each electronic device in a target space. A second signal acquisition module is configured to acquire test signals of multiple frequencies reflected by each electronic device in the target space. A feature information extraction module is configured to extract features of the terahertz signals to obtain radio frequency feature information of the electronic device. A feature matrix establishment module is configured to establish an impedance spectrum feature matrix according to the test signals of multiple frequencies. A civilian device identification module is configured to identify a civilian electronic device according to the radio frequency feature information and the impedance spectrum feature matrix. A real-time positioning acquisition module is configured to acquire real-time positioning of the civilian electronic device. A potential field vector acquisition module is configured to generate a virtual potential field vector of the civilian electronic device according to the real-time positioning. A flight attitude control module is configured to control a flight attitude of the unmanned aerial vehicle according to the virtual potential field vector.

[0112] Further, the feature information extraction module comprises: A spectrum image conversion unit is configured to convert the terahertz signals into a two-dimensional spectrum image. The feature information extraction unit is configured to input the two-dimensional spectrum image into a ResNet model, and output radio frequency feature information of the electronic device; the radio frequency feature information comprises a peak value, a valley value and a fluctuation of the spectrum.

[0113] Further, the feature matrix establishing module comprises: The impedance value acquisition unit is configured to acquire impedance values of each test signal. The feature matrix generation unit is configured to arrange all the impedance values into a matrix form to obtain an impedance spectrum feature matrix; one row of the impedance spectrum feature matrix corresponds to one frequency point, and one column of the impedance spectrum feature matrix corresponds to one electronic device.

[0114] Further, the potential field vector acquisition module comprises: The coordinate system establishing unit is configured to establish a three-dimensional coordinate system of the target space with the signal acquisition point as the origin; The coordinate point marking unit is configured to mark coordinate points of each civil electronic device in the three-dimensional coordinate system according to real-time positioning. The vector acquisition unit is configured to acquire a vector direction and a vector size of each coordinate point; the vector direction is a direction of the coordinate point pointing to the origin, and the vector size is an inverse of a distance of the coordinate point from the origin.

[0115] Further, the flight attitude control module comprises: The flight direction acquisition unit is configured to superimpose the vector sizes and the vector directions of all the coordinate points to obtain a target flight direction of the unmanned aerial vehicle; The flight attitude control unit is configured to control a flight attitude of the unmanned aerial vehicle according to a current flight direction and the target flight direction of the unmanned aerial vehicle.

[0116] The system further comprises: The probe photon emission module is configured to generate an entangled photon pair containing a probe photon and a reference photon, emit the probe photon to the target space, and collect the probe photon reflected by each electronic device in the target space; The quantum state correlation degree acquisition module is configured to acquire a quantum state correlation degree between the reference photon and the reflected probe photon; The noise quantum state marking module is configured to mark noise quantum states with a quantum state correlation degree less than a correlation degree threshold in the reflected probe photon; The entangled state signal generation module is configured to generate an entangled state signal orthogonal to the noise quantum state; The noise signal cancellation module is configured to cancel noise signals in the terahertz signal by using the entangled state signal.

[0117] Further, the first signal acquisition module and the probe photon emission module are coaxial and arranged at the bottom of the unmanned aerial vehicle.

[0118] It should be understood that the terms "system," "apparatus," "unit," and / or "module" as used herein are used generically to refer to different levels of aggregation of different components, elements, parts, sections, or assemblies. Other words, such as "include," "comprise," or "have," are used synonymously with "comprising" in the sense of including, but not limited to, as in the case of "comprising," the item listed after the "include," "comprise," or "have" is not the only item in a corresponding list of items.

[0119] As used in the specification and claims, unless otherwise specified, "a," "an," "the," and / or "at least one" are used generically and not in an exclusive sense, unless otherwise indicated. In general, the term "or" as used herein means "and / or," unless otherwise indicated.

[0120] The above detailed description has been given to the present application for the purpose of further explaining the purpose, technical solutions, and beneficial effects of the present application. It should be understood that the above description is only a specific embodiment of the present application and does not limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the protection scope of the present application.

[0121] It should be noted that the structures, proportions, sizes, etc. shown in the drawings attached to the present specification are only used to illustrate the content disclosed in the present specification, to be understood and read by those skilled in the art, and do not limit the conditions for implementing the present application, and therefore do not have technical significance. Any modification, change of proportion relationship, or adjustment of size, without affecting the effects and purposes that can be achieved by the present application, should still fall within the scope of the technology disclosed by the present application. At the same time, the terms such as "up", "down", "left", "right", "middle", etc. cited in the present specification are only for the convenience of clear description, and are not used to limit the scope of the present application. The change or adjustment of the relative relationship without substantially changing the technical content is also considered as the scope of the present application.

Claims

1. A UAV flight control method, characterized in that: The following steps are involved: Collect terahertz signals and test signals of multiple frequencies reflected by each electronic device in the target space; Extract features from terahertz signals to obtain radio frequency feature information of electronic devices; Establishing an impedance spectrum characteristic matrix based on test signals of multiple frequencies; Identify civilian electronic devices based on RF characteristic information and impedance spectrum characteristic matrix; Obtaining real-time positioning of civilian electronic devices; Generate virtual potential field vectors of civilian electronic equipment based on real-time positioning; The flight attitude of the UAV is controlled according to the virtual potential field vector.

2. The UAV flight control method according to claim 1, characterized in that: The following steps are also included: generating an entangled photon pair consisting of a probe photon and a reference photon; Sending detection photons to the target space; Collect detection photons reflected by each electronic device in the target space; Obtaining the quantum state deviation between the reference photon and the reflected detection photon; Mark the noise quantum state whose quantum state deviation is greater than the deviation threshold in the reflected detection photons; Generate an entangled state signal that is orthogonal to the noise quantum state; Entangled state signals are used to cancel out noise signals in terahertz signals.

3. A UAV flight control method according to claim 1 or 2, characterized in that: Feature extraction of terahertz signals includes the following steps: Convert terahertz signals into two-dimensional spectrum images; The two-dimensional spectrum image is input into the ResNet model, and the RF feature information of the electronic device is output; the RF feature information includes: peaks, valleys and fluctuations of the spectrum.

4. A UAV flight control method according to claim 1 or 2, characterized in that: Establishing an impedance spectrum characteristic matrix according to a plurality of test signals includes the following steps: Obtain the impedance value of each test signal; All impedance values ​​are organized into a matrix form to obtain an impedance spectrum characteristic matrix; a row of the impedance spectrum characteristic matrix corresponds to a frequency point, and a column of the impedance spectrum characteristic matrix corresponds to an electronic device.

5. A UAV flight control method according to claim 1 or 2, characterized in that: Generating a virtual potential field vector of a civilian electronic device according to real-time positioning includes the following steps: Establish a three-dimensional coordinate system of the target space with the signal acquisition point as the origin; Based on real-time positioning, the coordinate points of each civilian electronic device are marked in the three-dimensional coordinate system; Get the vector direction and vector magnitude of each coordinate point; the vector direction is the direction from the coordinate point to the origin, and the vector magnitude is the reciprocal of the distance from the coordinate point to the origin.

6. The UAV flight control method according to claim 5, characterized in that: Controlling the flight attitude of the UAV according to the virtual potential field vector includes the following steps: Superimpose the vector magnitudes and vector directions of all coordinate points to obtain the target flight direction of the UAV; The flight attitude of the UAV is controlled according to the current flight direction and target flight direction of the UAV.

7. A UAV flight control system, characterized in that: include: A first signal acquisition module is used to collect the terahertz signal reflected by each electronic device in the target space; A second signal acquisition module is used to collect test signals of multiple frequencies reflected by each electron in the target space; A feature information extraction module is used to extract features from terahertz signals to obtain radio frequency feature information of electronic devices; A characteristic matrix establishment module is used to establish an impedance spectrum characteristic matrix according to test signals of multiple frequencies; A civilian device identification module is used to identify civilian electronic devices based on radio frequency characteristic information and impedance spectrum characteristic matrix; A real-time positioning acquisition module is used to obtain the real-time positioning of civilian electronic equipment; A potential field vector acquisition module is used to generate a virtual potential field vector of a civilian electronic device based on real-time positioning; The flight attitude control module is used to control the flight attitude of the UAV according to the virtual potential field vector.

8. The UAV flight control system according to claim 7, characterized in that: Also includes: a detection photon emission module, configured to generate entangled photon pairs comprising detection photons and reference photons, emit detection photons into the target space, and collect detection photons reflected by each electronic device in the target space; A quantum state correlation acquisition module is used to obtain the quantum state correlation between the reference photon and the reflected detection photon; A noise quantum state marking module is used to mark noise quantum states with a quantum state correlation less than a correlation threshold in reflected detection photons; An entangled state signal generation module, used to generate an entangled state signal orthogonal to the noise quantum state; The noise signal cancellation module is used to use the entangled state signal to cancel the noise signal in the terahertz signal.

9. A UAV flight control system according to claim 7 or 8, characterized in that: The feature information extraction module includes: A spectrum image conversion unit, used for converting the terahertz signal into a two-dimensional spectrum image; The feature information extraction unit is used to input the two-dimensional spectrum image into the ResNet model and output the radio frequency feature information of the electronic device; the radio frequency feature information includes: peaks, valleys and fluctuations of the spectrum.

10. A UAV flight control system according to claim 7 or 8, characterized in that: The feature matrix building module includes: An impedance value obtaining unit, used to obtain the impedance value of each test signal; The characteristic matrix generating unit is used to organize all impedance values ​​into a matrix form to obtain an impedance spectrum characteristic matrix; a row of the impedance spectrum characteristic matrix corresponds to a frequency point, and a column of the impedance spectrum characteristic matrix corresponds to an electronic device.

11. A UAV flight control system according to claim 7 or 8, characterized in that: The potential field vector acquisition module includes: A coordinate system establishment unit, used to establish a three-dimensional coordinate system of the target space with the signal acquisition point as the origin; A coordinate point marking unit, used to mark the coordinate point of each civilian electronic device in a three-dimensional coordinate system based on real-time positioning; The vector acquisition unit is used to obtain the vector direction and vector magnitude of each coordinate point; the vector direction is the direction from the coordinate point to the origin, and the vector magnitude is the reciprocal of the distance from the coordinate point to the origin.

12. The UAV flight control system according to claim 11, characterized in that: The flight attitude control module includes: The flight direction acquisition unit is used to superimpose the vector magnitudes and vector directions of all coordinate points to obtain the target flight direction of the UAV; The flight attitude control unit is used to control the flight attitude of the UAV according to the current flight direction and target flight direction of the UAV.

13. The UAV flight control system according to claim 8, characterized in that: The first signal acquisition module and the detection photon emission module are coaxially arranged at the bottom of the UAV.

Images