A method and system for flight control of a drone

Abstract

The application belongs to the technical field of unmanned aerial vehicle flight control, and relates to an unmanned aerial vehicle flight control method and system. Terahertz signals reflected by each electronic device in a target space and test signals of multiple frequencies reflected by each electronic device are collected; feature extraction is performed on the terahertz signals to obtain radio frequency characteristic information of the electronic device; an impedance spectrum characteristic matrix is established according to the test signals of the multiple frequencies; the civil electronic device is identified according to the radio frequency characteristic information and the impedance spectrum characteristic matrix; real-time positioning of the civil electronic device is acquired; a virtual potential field vector of the civil electronic device is generated according to the real-time positioning; and the flight attitude of the unmanned aerial vehicle is controlled according to the virtual potential field vector, so that the unmanned aerial vehicle can respond to changes in the device position in real time to optimize the path and avoid obstacles.

Description

Technical Field

[0001] This invention belongs to the field of unmanned aerial vehicle (UAV) flight control technology, specifically relating to a UAV flight control method and system. Background Technology

[0002] With the widespread application of drone technology in civilian, industrial, and public service fields, its autonomous flight control and obstacle avoidance capabilities in complex environments have become core technological challenges. Especially in flight missions involving densely populated areas, accurately identifying the range of human activity and achieving safe obstacle avoidance directly impacts flight safety and mission compliance.

[0003] Traditional drone obstacle avoidance technology mainly relies on single perception methods such as visual sensors, lidar, or ultrasonic sensors to achieve path planning by constructing a 3D environmental model. However, these methods have significant limitations in complex scenarios: visual sensors are susceptible to changes in lighting, weather interference (such as fog, haze, and dust), and obstructions, easily creating blind spots in areas shaded by buildings, in dark environments at night, or in densely built-up areas, leading to a significant decrease in the accuracy of perception of areas with human activity; while lidar has a certain penetration capability, its resolution for recognizing non-metallic objects is low, and it is prone to misjudgment in multi-target mixed scenarios, making it difficult to accurately distinguish between human and non-human targets.

[0004] In terms of personnel identification, existing technologies are mostly based on image feature analysis, extracting visual features such as human contours and behavioral postures to achieve personnel identification. However, such methods have two major problems: first, privacy protection risks, as high-resolution image acquisition is prone to the leakage of personal privacy information; second, insufficient robustness of recognition, as human features are easily interfered with by factors such as clothing obscuring them, group gatherings, and changes in posture, resulting in an accuracy rate of less than 60% in complex environments, making it difficult to meet the requirements of high-precision obstacle avoidance. At the same time, traditional methods often rely on static maps or preset no-fly zones for path constraints, which cannot respond to dynamic personnel movement in real time, and are prone to obstacle avoidance lag problems in scenarios where personnel activity areas are dynamically changing (such as gatherings and community activities). Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for controlling unmanned aerial vehicles (UAVs) flight, which breaks through the environmental limitations and blind spots of traditional sensing technologies, constructs a personnel identification mechanism that takes into account recognition accuracy, real-time performance and anti-interference, and deeply integrates it with UAV flight control to form a dynamic response automated control link.

[0006] This invention is achieved through the following technical solution:

[0007] Firstly, a method for controlling the flight of an unmanned aerial vehicle (UAV) is proposed, comprising the following steps: collecting terahertz signals reflected by each electronic device in the target space and test signals of multiple frequencies; extracting features from the terahertz signals to obtain the radio frequency (RF) characteristic information of the electronic devices; establishing an impedance spectrum feature matrix based on the test signals of multiple frequencies; identifying civilian electronic devices based on the RF characteristic information and the impedance spectrum feature matrix; obtaining the real-time positioning of the civilian electronic devices; generating a virtual potential field vector of the civilian electronic devices based on the real-time positioning; and controlling the flight attitude of the UAV based on the virtual potential field vector.

[0008] Furthermore, the method also includes the following steps: generating an entangled photon pair containing a probe photon and a reference photon; emitting a probe photon into the target space; collecting probe photons reflected by each electronic device in the target space; obtaining the quantum state correlation degree between the reference photon and the reflected probe photon; marking noisy quantum states with a quantum state correlation degree < correlation degree threshold in the reflected probe photons; generating an entangled state signal orthogonal to the noisy quantum state; and using the entangled state signal to cancel the noise signal in the terahertz signal.

[0009] Secondly, a UAV flight control system is proposed, comprising: a first signal acquisition module for acquiring terahertz signals reflected by each electronic device in the 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 from the terahertz signals to obtain radio frequency feature information of the electronic devices; a feature matrix establishment module for establishing an impedance spectrum feature matrix based on the test signals of multiple frequencies; a civilian device identification module for identifying civilian electronic devices based on the radio frequency feature information and the impedance spectrum feature matrix; a real-time positioning acquisition module for acquiring the real-time positioning of the civilian electronic devices; a potential field vector acquisition module for generating a virtual potential field vector of the civilian electronic devices based on the real-time positioning; and a flight attitude control module for controlling the flight attitude of the UAV based on the virtual potential field vector.

[0010] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0011] 1. Based on the widespread use of civilian electronic devices, this invention indirectly achieves personnel identification by using the feature recognition method of electromagnetic signals reflected by civilian electronic devices, thereby avoiding the limitations of existing technologies that rely on image feature recognition. Specifically, by simultaneously acquiring reflection data of terahertz signals and multi-frequency test signals, the radio frequency (RF) characteristics of the equipment are extracted using the sensitivity of terahertz signals to hardware radiation characteristics. Combined with an impedance spectrum feature matrix constructed from multi-frequency reflections, the physical properties of the equipment are cross-verified, forming a dual identification mechanism of "RF characteristics + impedance spectrum characteristics." This mechanism not only enhances anti-interference capabilities in complex environments due to the strong penetrating power of terahertz signals and the multi-frequency noise reduction capabilities of impedance spectra, but also more accurately distinguishes between civilian and non-civilian electronic devices. Furthermore, by real-time positioning of civilian electronic devices and strongly correlating their location with the areas of human activity, it overcomes visual blind spots and achieves dynamic perception of the core activity areas of personnel. Then, based on the positioning data, a virtual potential field vector is generated, transforming the area associated with personnel into a dynamic "repulsive field." This allows the UAV to achieve millisecond-level smooth adjustments to its flight attitude, forming an automated closed-loop link of "signal perception - feature recognition - positioning analysis - potential field decision - attitude control." This process reduces the misjudgment rate of single feature recognition and can respond in real-time to changes in equipment position to optimize path obstacle avoidance.

[0012] 2. Utilizing the quantum state correlation of entangled photon pairs, a noise identification mechanism based on "detection-reference-correlation analysis" was constructed. By emitting probe photons into the target space and collecting reflected photons, and combining this with reference photons to calculate the quantum state correlation degree, noise quantum states with correlation degrees below a threshold can be accurately identified. This process relies on the inherent correlation of quantum states, overcoming the dependence of traditional noise filtering on signal strength and frequency characteristics. It can distinguish effective signals (correlated photons reflected by civilian electronic devices) from environmental noise (such as electromagnetic interference and randomly scattered photons) at the quantum level, achieving precise noise localization and labeling. Based on this, entangled state signals orthogonal to the noise quantum states are generated for noise cancellation. Leveraging the physical property of quantum state orthogonality, the labeled noise can be specifically eliminated without damaging the effective terahertz signal. This quantum domain noise cancellation method is more efficient than traditional filtering techniques, maximizing the preservation of device radio frequency characteristic information contained in the terahertz signal. It solves the problem of effective signals being submerged by noise in complex electromagnetic environments (such as areas with mixed signals from multiple devices and strong interference), significantly improving the clarity and completeness of subsequent radio frequency feature extraction. Attached Figure Description

[0013] The accompanying drawings, which are included to provide a further understanding of embodiments of the invention and form part of this application, do not constitute a limitation thereof. In the drawings:

[0014] Figure 1 This is a schematic diagram of a drone flight control process provided in Embodiment 1 of the present invention.

[0015] Figure 2 This is a schematic diagram of the visualization of the virtual potential field vector provided in Embodiment 1 of the present invention.

[0016] Figure 3 This is a visual schematic diagram illustrating the vector overlay generation of the target flight direction of the UAV provided in Embodiment 1 of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. The illustrative embodiments and descriptions of this invention are for illustrative purposes only and are not intended to limit the invention. The embodiments described below are some, but not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0018] In the following description, numerous specific details are set forth to provide a thorough understanding of the invention. However, it will be apparent to those skilled in the art that these specific details are not necessary to practice the invention. In other embodiments, well-known structures, materials, or methods are not specifically described to avoid obscuring the invention. Unless otherwise specified, the materials, instruments, and reagents used in the following embodiments are commercially available. Unless otherwise specified, the techniques used in the embodiments are conventional methods well known to those skilled in the art.

[0019] 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 technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0020] Example 1: A method for controlling the flight of an unmanned aerial vehicle (UAV) is provided, including... Figure 1 The following steps are shown:

[0021] Step 1: Collect terahertz signals reflected by each electronic device in the target space and test signals of multiple frequencies.

[0022] With the acceleration of urbanization and the widespread use of consumer electronic devices, smartphones, tablets, smartwatches, and other consumer electronic devices have become necessities in the daily lives of modern residents, and their distribution is strongly correlated with the areas where people are active. On the one hand, when the internal circuits of electronic devices are working, their chips, antennas, capacitors, and other components naturally generate electromagnetic signals inherent to the device due to changes in current. Terahertz radar utilizes the penetrating power of the 0.3-3THz frequency band to directly penetrate the non-metallic casing (such as plastic and glass) of electronic devices, accurately collecting these native electromagnetic signals generated inside the devices. The terahertz signals reflected by the electronic devices carry the signal characteristics of the devices' internal components and are collected by the terahertz radar, creating conditions for obtaining the true signal characteristics inside the devices.

[0023] On the other hand, the impedance of the wireless transmission module of electronic devices (such as the RF antenna of a mobile phone and the communication module of a smartwatch) varies uniquely with factors such as operating frequency, transmission power, and component characteristics during operation. Impedance is the ratio of voltage to current in a circuit, reflecting the circuit's resistance to current and energy conversion characteristics; it is an inherent physical property of the device hardware. This method uses a vector network analyzer (parameter specifications: 100kHz-20GHz frequency band, impedance measurement accuracy ±0.5%) to collect the impedance characteristics of the wireless transmission module of electronic devices. This vector network analyzer transmits test signals of different frequencies to the electronic device and calculates the impedance values ​​(including resistance and reactance components) at different frequencies based on the reflection and transmission parameters of the measured signals within the module, ultimately generating an impedance variation curve of the device over a wide frequency band—i.e., an "impedance spectrum."

[0024] Therefore, prior to this step, terahertz signals and test signals of multiple frequencies need to be continuously transmitted from multiple directions within the target space to detect electronic equipment within the target space.

[0025] Step 2: Eliminate environmental electromagnetic noise interference in terahertz signals by detecting quantum entangled state signals.

[0026] The reflected terahertz signals from electronic devices are typically weak (≤-110dBm) and easily drowned out by ambient noise in complex electromagnetic environments. To eliminate noise interference in these weak signals, this embodiment employs the following steps:

[0027] Step 2.1: Generate an entangled photon pair containing a probe photon and a reference photon, and emit the probe photon into the target space.

[0028] Quantum entanglement refers to the "non-local correlation" of quantum states 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 instantaneously determined, unaffected by distance or environmental interference. This embodiment generates entangled photon pairs using a quantum radar module. One photon is emitted as the "detector photon" into the target environment, while the other is retained as the "reference photon." Electromagnetic noise in the environment (such as interference signals from industrial equipment or 5G base stations) is a classical electromagnetic signal. Its mechanism of action is to interfere with the detector 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 detector photon, but cannot change the original entanglement relationship between the reference photon and the detector photon. Therefore, even if the detector photon is contaminated by noise during propagation, the quantum state of the reference photon remains "pure," providing an absolute benchmark for distinguishing between "real signals" and "noise interference."

[0029] Step 2.2: Collect the probe photons reflected by 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 in the reflected probe photon where the quantum state deviation is greater than the deviation threshold.

[0030] After the probe photon interacts with the target device (or environment), it returns carrying the target signal characteristics (such as phase shifts and polarization changes caused by terahertz frequency signals radiated from the circuit), while simultaneously being superimposed with environmental noise. Based on the correlation of quantum entanglement, the quantum states of the reference photon and the probe photon should be strictly matched in the absence of interference. However, the quantum state deviation caused by noise has random and irregular characteristics (classical noise cannot form a stable correlation), while the real signal of the electronic device affects the probe photon through electromagnetic interaction, and the resulting deviation exhibits a regularity related to the device hardware (such as specific phase patterns radiated by the chip). Therefore, by calculating the quantum state deviation between the probe photon and the reference photon (such as phase difference fluctuations or polarization angles), we can obtain the quantum state deviation caused by the real signal of the target device and the quantum state deviation caused by noise interference.

[0031] Furthermore, a quantum state analyzer is used to jointly measure the reference photon and the returned probe photon, calculating the quantum state deviation between them. A quantum state with a quantum state deviation greater than the deviation threshold is considered a noise quantum state. For example, the phase difference bands of the reference photon and the returned probe photon are calculated separately. If the phase difference band of a certain segment of the signal in the returned probe photon is greater than 2°, then that segment of the signal can be determined to be affected by environmental noise interference.

[0032] 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.

[0033] Electromagnetic noise in the environment can be considered as a "noise quantum state" with specific quantum state characteristics at the quantum level. This step uses a quantum entanglement generator to generate an entangled signal orthogonal to this noise quantum state. In quantum mechanics, "orthogonality" means that there are no overlapping components between the two quantum states; they are independent 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 signal will be vertical (orthogonal relationship); if the noise quantum state has a specific phase distribution, the entangled signal will form a phase mode orthogonal to it through phase modulation. This orthogonality ensures that the newly generated entangled signal is not "contaminated" by noise, providing a "clean" reference for subsequent signal purification.

[0034] The acquired terahertz signal and the generated orthogonal entangled state signal are subjected to quantum interference measurement. By utilizing the principle of orthogonality, the noise quantum state and the entangled state signal cancel each other out, while the quantum state of the real signal is preserved because it is not orthogonal to the entangled state.

[0035] Step 3: Extract features from the terahertz signal to obtain the radio frequency characteristic information of the electronic device.

[0036] Utilizing the signal processing and visualization capabilities of terahertz radar, the terahertz signals acquired by the radar are converted into a two-dimensional spectrum image (horizontal axis represents frequency, and vertical axis represents signal strength). This two-dimensional spectrum image is then input into a trained ResNet model. The ResNet model extracts features layer by layer from the spectrum image through multiple residual convolutional layers. Shallow layers capture basic morphological features such as peaks and valleys in the spectrum, while deeper layers uncover advanced features such as subtle fluctuations in the spectrum. These features correspond to hardware physical attributes such as chip model, antenna material and structure, and circuit board wiring.

[0037] By processing and identifying terahertz frequency signals reflected from the internal circuitry of electronic devices, unique and stable characteristic information can be extracted. This collection of characteristic information constitutes the device's "radio frequency fingerprint." Just as a person's fingerprint is unique and can be used for identification, the "radio frequency fingerprint" of an electronic device is also unique and stable. The "radio frequency fingerprint" generated based on terahertz waves can accurately distinguish between different electronic devices. Even devices of the same type can have different "radio frequency fingerprints" due to factors such as manufacturing processes and wear and tear. In drone-based personnel identification scenarios, drones can determine whether an electronic device is a civilian device by identifying its "radio frequency fingerprint," thus providing a reliable identification basis for subsequent positioning and obstacle avoidance decisions.

[0038] Before this, the ResNet model needs to be trained, which is crucial for feature extraction from terahertz signals. By training the ResNet model, the advantages of residual learning can be effectively utilized to achieve high-precision recognition in tasks such as terahertz spectrum image classification, while also possessing strong generalization ability to adapt to device recognition needs in complex environments. The method for training the ResNet model is as follows:

[0039] 1. Constructing the dataset

[0040] (1) Data source: Collect samples (such as terahertz spectrum images), which need to cover all categories involved in the task (such as mobile phones, watches, and headphones). The sample size of each category can be ≥3000 images to ensure category balance (avoiding that the proportion of a certain category of samples exceeds 70%).

[0041] (2) Data labeling: Label the samples (e.g., “Mobile phone - Model A”, “Watch - Model B”). Manual labeling can be combined with automated tools (e.g., Label Studio) to improve efficiency.

[0042] (3) Data partitioning: Divide the data into training set (model learning), validation set (parameter tuning), and test set (final evaluation) in a ratio of 7:2:1. Keep the category distribution consistent during partitioning (e.g., if mobile phones account for 60% in the training set, the validation set and test set should also be close to 60%).

[0043] 2. Construction and Configuration of ResNet Model

[0044] Choose the appropriate ResNet architecture (such as ResNet-18 / 34 / 50) based on the task requirements and make targeted adjustments.

[0045] (1) Model selection

[0046] Given that this method needs to identify fine-grained spectral features of electronic devices from terahertz signals, the ResNet-50 / 101 architecture is chosen to extract subtle features using a deeper network.

[0047] (2) Network structure adaptation

[0048] Input layer: The original ResNet's 3-channel input is changed to 1 channel to process single-channel 2D spectral images. The parameters of the first convolutional layer are adjusted (in_channels=1).

[0049] Output layer: Modify the output dimension of the last fully connected layer according to the number of electronic device categories (e.g., set it to 3 for 3 categories of devices), and output the category probability in conjunction with the Softmax activation function.

[0050] Residual block optimization: The key residual blocks adopt the structure of "1×1 convolution for dimensionality reduction → 3×3 convolution for feature extraction → 1×1 convolution for dimensionality increase", and the gradient vanishing is alleviated by shortcut connections (skip connections) to ensure that the deep network is trainable.

[0051] (3) Initialize parameters

[0052] The weights of the convolutional layers are initialized using the Kaiming normal distribution (adapted to the ReLU activation function), and the weights of the fully connected layers are initialized using the Xavier distribution. The γ of the Batch Normalization layer is set to 1 and the β is set to 0 to ensure that the initial distribution is stable.

[0053] 3. Training parameters and strategy settings

[0054] (1) Core parameter configuration

[0055] Optimizer: Adam (adaptive learning rate) is preferred, with an initial learning rate of 0.001 and weight decay (L2 regularization) of 1e-5 (to suppress overfitting).

[0056] Learning rate scheduling: Use cosine annealing or piecewise decay (e.g., halve the learning rate every 30 epochs) to avoid oscillations caused by an excessively high learning rate or slow convergence caused by an excessively low learning rate.

[0057] Batch Size: Based on GPU memory settings (e.g., 32 for 12GB of VRAM), too small a size will cause large gradient fluctuations, while too large a size will consume too many resources.

[0058] Training epochs: Initially set to 100-200, combined with an early stopping mechanism (stop if the validation set accuracy does not improve for 10 consecutive epochs).

[0059] (2) Loss function

[0060] Similarly, based on fine-grained recognition, this embodiment selects the Center Loss function to narrow the distance between similar features.

[0061] 4. Training the model

[0062] (1) Training process

[0063] Forward propagation: The enhanced sample is input, features are extracted through convolutional layers and residual blocks, and finally the prediction result is output through a fully connected layer.

[0064] Backpropagation: Calculate the loss between the predicted value and the true label, update the network parameters through gradient descent (such as Adam), and update the mean and variance of the Batch Normalization layer.

[0065] Iterative optimization: After each round of training, evaluate the performance on the validation set and adjust the learning rate or stop early based on the validation results.

[0066] (2) Monitoring of key indicators

[0067] Loss: The training set loss should gradually decrease and stabilize. If the fluctuation is too large, the batch size or learning rate should be checked.

[0068] Accuracy: The accuracy of the training set and the validation set should be improved in tandem. If the accuracy of the validation set decreases, overfitting may occur, and regularization needs to be strengthened.

[0069] Confusion matrix: Analyze misclassified samples (such as misclassifying headphones as watches) and supplement samples or adjust the model accordingly.

[0070] 5. Model Evaluation and Optimization

[0071] (1) Test set evaluation

[0072] After training, the model performance is evaluated on an independent test set. Key metrics include:

[0073] Accuracy: Overall classification accuracy, ≥95% (can be appropriately reduced for complex tasks);

[0074] Precision and Recall: Ensure the reliability of identification for each type of device (e.g., headphone recall rate ≥90%).

[0075] Robustness test: Verify the model's stability under noise, occlusion and other interference scenarios (performance degradation should be ≤10%).

[0076] (2) Model optimization

[0077] For categories with high error rates, supplementary samples are used for fine-tuning (the learning rate is reduced to 1 / 10 of the initial value).

[0078] Transfer learning is employed: weights are initialized based on pre-trained ResNet (such as models trained on ImageNet), and only the output layer and some residual blocks are adjusted to accelerate convergence and improve performance.

[0079] Step 4: Establish the impedance spectrum characteristic matrix based on test signals at multiple frequencies.

[0080] Step 4.1: Obtain the impedance value of each test signal.

[0081] Based on the explanation in step 1, a vector network analyzer is used to collect the impedance characteristics of the wireless transmission module of the electronic device. A vector network analyzer is an instrument capable of accurately measuring the characteristics of radio frequency and microwave circuits over a wide frequency range. In this embodiment, its operating frequency band is 100kHz-20GHz, and its impedance measurement accuracy reaches ±0.5%. During operation, it sends test signals of different frequencies to the wireless transmission module of the electronic device. By measuring the reflected and transmitted signals, it calculates the impedance values ​​of the electronic device at various frequency points. These impedance values ​​change with frequency in the form of curves, i.e., impedance variation curves. Due to differences in internal circuit design and component characteristics, different devices exhibit different impedance variation curves. For example, mobile phones and smartwatches, due to differences in function and circuit complexity, will show significant differences in impedance variation trends and values ​​within the same frequency range.

[0082] Step 4.2: Arrange all impedance values ​​into matrix form to obtain the impedance spectrum characteristic matrix.

[0083] The impedance values ​​collected at different frequencies are organized into a matrix to construct an impedance spectrum feature matrix. The matrix can be organized as follows: each row represents a frequency point, with elements in each row corresponding to the impedance value (including resistance, reactance, etc.) at that frequency point; each column represents an electronic device, with elements in each column corresponding to the impedance values ​​of that electronic device at different frequencies. This impedance spectrum feature matrix comprehensively records the impedance characteristics of each electronic device within a specific frequency range, serving as an "electrical fingerprint" for the device. For example, measuring the impedance data of multiple mobile phones of the same model and organizing them into an impedance spectrum feature matrix clearly reveals the commonalities in the impedance characteristics of that model of mobile phones, while also identifying subtle differences between individual devices.

[0084] Step 5: Identify civilian electronic devices based on radio frequency characteristic information and impedance spectrum characteristic matrix.

[0085] This embodiment utilizes Support Vector Machine (SVM) to identify consumer electronic devices from radio frequency (RF) feature information and impedance spectrum feature matrices. Different devices (such as mobile phones, routers, microwave ovens, and Bluetooth headsets) exhibit significant differences in their RF signal characteristics and impedance spectrum characteristics; these differences constitute the "discriminative features" for classification. By fusing RF and impedance spectrum features, SVM can learn the joint distribution pattern of the two types of features, thereby distinguishing different device types. The specific method is as follows:

[0086] Similarly, before using the SVM model to classify and identify civilian electronic devices, the SVM model needs to be trained. The methods for training and optimizing the SVM model are as follows:

[0087] 1. Data Acquisition and Preprocessing

[0088] (1) Radio frequency signal acquisition

[0089] Use a spectrum analyzer and software-defined radio (such as USRP or HackRF) to collect the radio frequency signals during device operation, and record time-domain or frequency-domain data (the sampling rate must cover the device's operating frequency band, such as 2.4GHz, 5GHz, and other civilian frequency bands). Remove environmental noise by filtering (such as low-pass filtering), and extract the effective signal during the device's operating period by signal interception (avoiding interference from idle state).

[0090] (2) Impedance spectroscopy data acquisition

[0091] The impedance spectrum of the device is measured using an impedance analyzer within a specific frequency range (e.g., 1kHz to 100MHz), yielding a matrix showing the impedance magnitude and phase variation with frequency. Preprocessing involves removing measurement noise through smoothing (e.g., moving average) and normalizing the impedance spectrum (eliminating individual device variations or measurement environment influences).

[0092] 2. Feature Extraction and Fusion

[0093] High-discriminative features are extracted from the preprocessed raw data and multimodal fusion is performed.

[0094] (1) Radio frequency feature extraction

[0095] The characteristics of radio frequency signals need to reflect the "communication / operational characteristics" of the device, and key characteristics include:

[0096] Frequency domain characteristics: center frequency, bandwidth, peak frequency, power spectral density (PSD), spectral entropy (measures the uniformity of spectral distribution), harmonic components, etc. (extracted through Fourier transform and short-time Fourier transform (STFT).

[0097] Modulation domain characteristics: modulation mode (such as ASK, FSK, PSK, QAM), symbol rate, modulation depth, etc. (through constellation diagram analysis and cyclic spectral density extraction).

[0098] Time-domain characteristics: signal duration, pulse width, frequency hopping period (for frequency hopping devices), envelope peak value, etc.

[0099] (2) Impedance spectral feature extraction

[0100] Impedance spectrum reflects the "circuit characteristics" of a device, and key features include:

[0101] Characteristic frequency parameters: resonant frequency (frequency point with minimum / maximum impedance magnitude), cutoff frequency, impedance magnitude and phase at the characteristic frequency.

[0102] Trend characteristics: the slope of the impedance modulus as a function of frequency (such as the difference in slope between low and high frequency bands), the rate of phase change, and the curvature extrema of the impedance spectrum curve.

[0103] Statistical characteristics: the mean, variance, number of peaks, and energy concentration regions of the impedance spectrum (quantified by integration or entropy).

[0104] (3) Feature fusion

[0105] The radio frequency features and impedance spectrum features are concatenated into a joint feature vector, which is then used as the input to the SVM.

[0106] Assuming the RF feature dimension is D1 and the impedance spectrum feature dimension is D2, the combined feature dimension is D1+D2.

[0107] Feature selection: Use ANOVA, mutual information (MI), or recursive feature elimination (RFE) to screen features that are strongly correlated with the device type and remove redundant features (reduce dimensionality and avoid overfitting).

[0108] 3. Feature Preprocessing

[0109] To improve the classification performance of SVM, the fused features need to be standardized.

[0110] Normalization: Maps eigenvalues ​​to the [0,1] interval, which is suitable for features with a defined range, such as impedance spectra.

[0111] Standardization: Transforms features into a distribution with a mean of 0 and a variance of 1, which is suitable for features with a large dynamic range, such as radio frequency power.

[0112] 4. SVM Model Training and Optimization

[0113] (1) Dataset partitioning

[0114] The samples were divided into a training set (70%), a validation set (15%), and a test set (15%). The training set was used for learning SVM model parameters; the validation set was used for hyperparameter tuning (such as kernel function selection and regularization coefficient); and the test set was used to evaluate the model's final generalization ability.

[0115] (2) Selection of SVM core parameters

[0116] SVM performance is highly dependent on parameters, and the following parameters need to be optimized:

[0117] Kernel function: The RBF kernel (radial basis function) is selected, which is suitable for nonlinear feature scenarios (RF and impedance spectrum features are usually nonlinearly correlated) and is the preferred kernel function for civilian device identification.

[0118] Regularization coefficient (C): controls model complexity and fault tolerance. The larger C is, the stronger the model fits the training set (prone to overfitting); the smaller C is, the stronger the fault tolerance (prone to underfitting).

[0119] Kernel parameter (γ): For RBF kernel, it controls the range of influence of the sample. The larger the γ is, the smaller the range of influence of a single sample (the more complex the model).

[0120] The model is trained on the training set based on the optimal parameters, and the model stability is evaluated through cross-validation (such as 5-fold cross-validation).

[0121] By accurately acquiring radio frequency and impedance spectrum data, extracting discriminative features with physical significance, and combining the advantages of SVM kernel functions and parameter optimization, efficient classification of different civilian electronic devices can be achieved.

[0122] Step 6: Obtain the real-time location of civilian electronic devices.

[0123] This embodiment uses triangulation to obtain the real-time location of civilian electronic devices. The specific method is as follows:

[0124] 1. Obtain signal acquisition points

[0125] The coordinates of three signal acquisition points are extracted from the GPS positioning data and set as follows: A ( x A , y A ), B ( x B , y B )and C ( x C , y C ).

[0126] 2. Obtain the distance between the target electronic device and the signal acquisition point.

[0127] Based on the time when the terahertz imaging radar emits terahertz waves t 1. The time between the acquisition of the terahertz wave reflected by the electronic device t 2. Calculate the distance from the target electronic device to each data collection point. d A , d B and d C The distance calculation formula is: d = c ×( t 2- t 1) / 2, c It is the speed of light.

[0128] 3. Solve for the coordinates of the target electronic device.

[0129] The coordinates of the three data collection points are known.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 data collection point. d A , d B and d C Establish a system of equations to solve for the coordinates of the target electronic device. x , y ):

[0130] (1) Establish the distance equation

[0131] The expression for the distance equation is: .

[0132] (2) Solving by linearization

[0133] Squaring both sides of the system of equations and eliminating the quadratic terms, we transform it into a linear system of equations:

[0134] ,

[0135] The coordinates of the target electronic device are obtained by solving a system of linear equations using matrix operations. x , y ).

[0136] Step 7: Generate a virtual potential field vector for civilian electronic devices based on real-time positioning.

[0137] Drawing inspiration from the concept of "interaction of forces" in physics, the locations of human-carried electronic devices are abstracted as "repulsive field sources." A virtual "virtual potential field vector" is constructed around the drone—the location of each identified civilian electronic device is considered a field source, and the "repulsive force" of the source varies with 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 of electric or magnetic fields). In the vector diagram, the vector direction of each point represents the direction of the resultant force acting on the drone at that location, and the vector magnitude represents the intensity of the resultant force. This visualization method transforms complex obstacle avoidance decisions into calculations of "force balance and direction," intuitively reflecting the areas the drone should avoid and the optimal movement path.

[0138] Based on the concept of force interaction, the method for generating the virtual potential field vector of civilian electronic devices is as follows:

[0139] Step 7.1: Establish a three-dimensional coordinate system of the target space with the signal acquisition point as the origin.

[0140] Step 7.2: Based on the real-time positioning, mark the coordinates of each civilian electronic device in the three-dimensional coordinate system.

[0141] Step 7.3: Obtain the vector direction and magnitude of each coordinate point.

[0142] Wherein, 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. After visualization, the established virtual potential field vector is as follows: Figure 2 As shown. Figure 2 In the diagram, points M, N, and P represent the real-time positions of civilian electronic devices in the target space, with the acquisition point located at the origin O of the coordinate system; the vector directions of points M, N, and P are represented by dashed arrows; 1 / d M 1 / d represents the vector magnitude of point M, i.e., the magnitude of the repulsive force exerted by point M on point O; N 1 / d represents the vector magnitude at point N, i.e., the magnitude of the repulsive force exerted by point N on point O; P Let P be the vector magnitude, which represents the magnitude of the repulsive force exerted by point P on point O.

[0143] Step 8: Control the flight attitude of the UAV based on the virtual potential field vector.

[0144] Step 8.1: Superimpose the vector magnitudes and vector directions of all coordinate points to obtain the target flight direction of the UAV.

[0145] When multiple targets are present, the total repulsive force experienced by the drone is the vector superposition of the repulsive forces from all sources. The safest obstacle avoidance path is determined by calculating the direction of the resultant force. (See reference...) Figure 3 First, the vector at point M is superimposed on the vector at point N. In this embodiment, point N and its vector arrow are translated to obtain point N' and its vector arrow. After translation, this vector is superimposed on the vector at point M (the direction and magnitude of the resultant force are drawn). The superimposed vector is... Figure 3 The dotted-line arrow is used to represent the target direction. Then, the dotted-line arrow is translated, and the translated arrow is superimposed on the vector at point P to obtain the target flight direction of the UAV. Figure 3 The solid arrow in the middle indicates...

[0146] Step 8.2: Control the flight attitude of the UAV based on its current flight direction and the target flight direction.

[0147] The core of drone flight attitude control is to dynamically adjust the drone's attitude (pitch angle, roll angle, yaw angle) 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 the context of obstacle avoidance scenarios for civilian electronic devices, the specific steps are as follows:

[0148] 1. Flight direction data acquisition and analysis

[0149] (1) Definition of core parameters

[0150] The real-time movement direction of the drone is analyzed into a three-dimensional direction vector using sensor data. Where: the horizontal direction (xy plane) reflects the flight direction of the UAV projected onto the ground (e.g., 30° east of north); the vertical direction (z-axis) reflects the ascent or descent trend (positive for ascent, negative for descent). The target flight direction of the UAV is defined as a three-dimensional vector. .

[0151] (2) Measure the fuselage angular rate and acceleration in real time using accelerometers and gyroscopes, and analyze the current attitude angles (pitch angle θ, roll angle φ, yaw angle ψ).

[0152] 2. Calculation of direction angle deviation

[0153] (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 the control of direction selection (such as direct steering for small angle deviations, and step-by-step adjustment for large angle deviations).

[0154] (2) Calculate the pitch deviation (vertical deviation). z Axial deviation 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 the pitch angle needs to be increased for descent).

[0155] (3) Set the deviation threshold

[0156] Set the maximum allowable deviation threshold (e.g., horizontal deviation ≤ 5°, vertical deviation ≤ 3°). When the deviation is less than the threshold, the UAV is determined to be flying stably in the target direction without significant adjustments. When the deviation exceeds the threshold, the attitude control algorithm is triggered to dynamically adjust the control quantity according to the magnitude of the deviation (the larger the deviation, the greater the adjustment).

[0157] 3. Flight attitude control

[0158] The required attitude adjustment is calculated based on the deviation and then converted into motor control signals by the flight control system.

[0159] (1) Proportional-integral-derivative (PID) control algorithm.

[0160] The PID controller is a classic closed-loop control algorithm framework in industrial control. It stands for Proportional-Integral-Derivative control. It calculates the control input by combining the current system error, historical cumulative error, and error trend to adjust the controlled object, stabilizing the system output at the target value.

[0161] Taking horizontal yaw control as an example, the core formula is: .in, Kp As the proportional coefficient, it directly outputs the control quantity based on the current deviation (e.g., outputting 5% of the motor differential speed when the deviation is 5°), quickly reducing the deviation; Ki The coefficients of the integral term accumulate long-term small deviations (such as a continuous 2° deviation) to eliminate static errors and ensure that the deviation approaches 0 in steady state. Kd The coefficient is the differential term coefficient, which is used to suppress overshoot based on the rate of change of deviation (e.g., to slow down in advance when the deviation increases rapidly) and avoid violent shaking. Kp, Ki, Kd It needs to be determined by debugging based on the system characteristics.

[0162] Vertical pitch control uses a PID structure, with the pitch angle PID parameter adjusted individually (usually...). Kp (Smaller than yaw control to avoid vertical oscillation).

[0163] (2) Attitude angle and motor output mapping

[0164] The control quantity output by the PID controller is converted into a specific attitude angle adjustment command.

[0165] Yaw angle adjustment: Achieved through differential speed adjustment of the left and right motors (e.g., the left motor speeds up, the right motor decelerates, and the drone turns clockwise). The magnitude of the differential speed is related to... Proportional to the maximum turning angular velocity, the maximum turning speed is limited to 30° / s (to avoid fuselage sway).

[0166] Pitch angle adjustment: The pitch of the fuselage is changed by the differential speed of the front and rear motors (such as the front motor decelerating and the rear motor accelerating, the fuselage pitches up to increase the pitch angle and increase lift to achieve climbing). The pitch angle adjustment range is limited to [-10°, 15°] (exceeding the range is easy to lose control).

[0167] Roll angle assist: Small horizontal angular deviations (<5°) can be smoothly corrected by fine-tuning the roll angle (±3°), reducing energy consumption and oscillation during yaw adjustment.

[0168] (3) Adaptive dynamic parameters

[0169] When flying at low speeds (<5m / s), reduce the proportional gain of the PID controller. Kp Reduce adjustment sensitivity to ensure stability.

[0170] During high-speed flight (>10m / s), increase Kp and Kd This improves response speed and avoids the accumulation of deviations.

[0171] In emergency obstacle avoidance scenarios, the response priority of temporarily increasing the deviation threshold allows for greater attitude adjustments (such as increasing the maximum yaw rate to 50° / s) to quickly avoid targets.

[0172] Example 2: Corresponding to Example 1, this example provides a drone flight control system, including:

[0173] The first signal acquisition module is used to acquire the terahertz signals reflected by each electronic device in the target space;

[0174] The second signal acquisition module is used to acquire test signals of multiple frequencies reflected by each electron in the target space;

[0175] The feature information extraction module is used to extract features from terahertz signals to obtain the radio frequency feature information of electronic devices;

[0176] The feature matrix establishment module is used to establish an impedance spectrum feature matrix based on test signals of multiple frequencies.

[0177] The civilian equipment identification module is used to identify civilian electronic devices based on radio frequency characteristic information and impedance spectrum characteristic matrix;

[0178] A real-time location acquisition module is used to acquire the real-time location of civilian electronic devices;

[0179] The potential field vector acquisition module is used to generate virtual potential field vectors for civilian electronic devices based on real-time positioning.

[0180] The flight attitude control module is used to control the flight attitude of the UAV based on the virtual potential field vector.

[0181] Furthermore, the feature information extraction module includes:

[0182] The spectrum image conversion unit is used to convert terahertz signals into two-dimensional spectrum images;

[0183] The feature information extraction unit is used to input a 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: peak value, valley value and fluctuation of the spectrum.

[0184] Furthermore, the feature matrix construction module includes:

[0185] Impedance value acquisition unit, used to acquire the impedance value of each test signal;

[0186] The characteristic matrix generation unit is used to organize all impedance values ​​into matrix form to obtain the impedance spectrum characteristic matrix; one row of the impedance spectrum characteristic matrix corresponds to one frequency point, and one column of the impedance spectrum characteristic matrix corresponds to one electronic device.

[0187] Furthermore, the potential field vector acquisition module includes:

[0188] The coordinate system establishment unit is used to establish a three-dimensional coordinate system of the target space with the signal acquisition point as the origin;

[0189] The coordinate point marking unit is used to mark the coordinate points of each civilian electronic device in a three-dimensional coordinate system based on real-time positioning.

[0190] The vector acquisition unit is used to acquire the vector direction and 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.

[0191] Furthermore, the flight attitude control module includes:

[0192] The flight direction acquisition unit is used to superimpose the vector magnitude and vector direction of all coordinate points to obtain the target flight direction of the UAV;

[0193] The flight attitude control unit is used to control the flight attitude of the UAV based on its current flight direction and the target flight direction.

[0194] The system also includes:

[0195] The probe photon emission module is used to generate entangled photon pairs containing probe photons and reference photons, emit probe photons into the target space, and collect probe photons reflected by each electronic device in the target space;

[0196] The quantum state correlation acquisition module is used to acquire the quantum state correlation between the reference photon and the reflected probe photon;

[0197] A noise quantum state labeling module is used to label noise quantum states in reflected probe photons whose quantum state correlation is less than the correlation threshold;

[0198] Entangled state signal generation module, used to generate entangled state signals orthogonal to the noisy quantum state;

[0199] The noise cancellation module is used to cancel noise signals in terahertz signals using entangled state signals.

[0200] Furthermore, the first signal acquisition module and the photon emission detection module are coaxially mounted on the bottom of the UAV.

[0201] It should be understood that the terms "system," "device," "unit," and / or "module" as used in this specification are a method of distinguishing different components, elements, parts, sections, or assemblies at different levels. However, if other terms can achieve the same purpose, they may be replaced by other expressions.

[0202] As indicated in this specification and claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" do not specifically refer to the singular and may also include the plural. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of expressly identified steps and elements, which do not constitute an exclusive list, and the method or apparatus may also include other steps or elements.

[0203] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0204] It should be noted that the structures, proportions, sizes, etc., illustrated in the accompanying drawings are merely for illustrative purposes to aid those skilled in the art and are not intended to limit the scope of the invention. Therefore, they have no substantial technical significance. Any modifications to the structure, changes in proportions, or adjustments to size, without affecting the effectiveness and purpose of the invention, should still fall within the scope of the disclosed technical content. Furthermore, terms such as "upper," "lower," "left," "right," and "middle" used in this specification are merely for clarity and not intended to limit the scope of the invention. Changes or adjustments to their relative relationships, without substantially altering the technical content, should also be considered within the scope of the invention.

Claims

1. A method for controlling the flight of an unmanned aerial vehicle (UAV), characterized in that, Includes the following steps: Collect terahertz signals reflected by each electronic device in the target space and test signals of multiple frequencies; Feature extraction is performed on terahertz signals to obtain radio frequency characteristic information of electronic devices; An impedance spectrum feature matrix is ​​established based on test signals at multiple frequencies; Civilian electronic devices can be identified based on radio frequency characteristic information and impedance spectrum characteristic matrix. To obtain the real-time location of civilian electronic devices; A virtual potential field vector for civilian electronic devices is generated based on real-time positioning. The flight attitude of the UAV is controlled based on the virtual potential field vector; Establishing an impedance spectrum characteristic matrix based on multiple test signals includes the following steps: Obtain the impedance value of each test signal; All impedance values ​​are arranged into a matrix form to obtain the impedance spectrum characteristic matrix; each row of the impedance spectrum characteristic matrix corresponds to a frequency point, and each column of the impedance spectrum characteristic matrix corresponds to an electronic device.

2. The UAV flight control method according to claim 1, characterized in that, It also includes the following steps: Generate entangled photon pairs containing a probe photon and a reference photon; Firing probe photons into the target space; Collect detector photons reflected by each electronic device in the target space; To obtain the quantum state deviation between the reference photon and the reflected probe photon; In the reflected probe photons, noisy quantum states with quantum state deviation > deviation threshold are marked; Generate entangled state signals orthogonal to the noisy quantum states; Entangled state signals are used to cancel noise signals in terahertz signals.

3. A method for controlling the flight of an unmanned aerial vehicle (UAV) according to claim 1 or 2, characterized in that, Feature extraction of terahertz signals includes the following steps: Convert terahertz signals into two-dimensional spectral images; The two-dimensional spectrum image is input into the ResNet model, which outputs the radio frequency (RF) characteristics of the electronic device. The RF characteristics include the peak, valley, and fluctuation of the spectrum.

4. A method for controlling the flight of an unmanned aerial vehicle (UAV) according to claim 1 or 2, characterized in that, The process of generating a virtual potential field vector for civilian electronic devices based on real-time positioning includes the following steps: A three-dimensional coordinate system for the target space is established with the signal acquisition point as the origin; Based on real-time positioning, the coordinates of each civilian electronic device are marked in a three-dimensional coordinate system; Obtain the vector direction and 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.

5. The UAV flight control method according to claim 4, characterized in that, Controlling the flight attitude of the UAV based on the virtual potential field vector includes the following steps: The target flight direction of the UAV is obtained by superimposing the vector magnitudes and directions of all coordinate points; The drone's flight attitude is controlled based on its current flight direction and the target flight direction.

6. A flight control system for unmanned aerial vehicles (UAVs), characterized in that, include: The first signal acquisition module is used to acquire the terahertz signals reflected by each electronic device in the target space; The second signal acquisition module is used to acquire test signals of multiple frequencies reflected by each electron in the target space; The feature information extraction module is used to extract features from terahertz signals to obtain the radio frequency feature information of electronic devices; The feature matrix establishment module is used to establish an impedance spectrum feature matrix based on test signals of multiple frequencies. The civilian equipment identification module is used to identify civilian electronic devices based on radio frequency characteristic information and impedance spectrum characteristic matrix; A real-time location acquisition module is used to acquire the real-time location of civilian electronic devices; The potential field vector acquisition module is used to generate virtual potential field vectors for civilian electronic devices based on real-time positioning. The flight attitude control module is used to control the flight attitude of the UAV based on the virtual potential field vector. The feature matrix construction module includes: Impedance value acquisition unit, used to acquire the impedance value of each test signal; The characteristic matrix generation unit is used to organize all impedance values ​​into matrix form to obtain the impedance spectrum characteristic matrix; one row of the impedance spectrum characteristic matrix corresponds to one frequency point, and one column of the impedance spectrum characteristic matrix corresponds to one electronic device.

7. A UAV flight control system according to claim 6, characterized in that, Also includes: The probe photon emission module is used to generate entangled photon pairs containing probe photons and reference photons, emit probe photons into the target space, and collect probe photons reflected by each electronic device in the target space; The quantum state correlation acquisition module is used to acquire the quantum state correlation between the reference photon and the reflected probe photon; A noise quantum state labeling module is used to label noise quantum states in reflected probe photons whose quantum state correlation is less than the correlation threshold; Entangled state signal generation module, used to generate entangled state signals orthogonal to the noisy quantum state; The noise cancellation module is used to cancel noise signals in terahertz signals using entangled state signals.

8. A UAV flight control system according to claim 6 or 7, characterized in that, The feature information extraction module includes: The spectrum image conversion unit is used to convert terahertz signals into two-dimensional spectrum images; The feature information extraction unit is used to input a 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: peak value, valley value and fluctuation of the spectrum.

9. A UAV flight control system according to claim 6 or 7, characterized in that, The potential field vector acquisition module includes: The coordinate system establishment unit is used 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 used to mark the coordinate points of each civilian electronic device in a three-dimensional coordinate system based on real-time positioning. The vector acquisition unit is used to acquire the vector direction and 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.

10. A UAV flight control system according to claim 9, characterized in that, The flight attitude control module includes: The flight direction acquisition unit is used to superimpose the vector magnitude and vector direction 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 based on its current flight direction and the target flight direction.

11. A UAV flight control system according to claim 7, characterized in that, The first signal acquisition module and the photon emission detection module are coaxially mounted on the bottom of the UAV.

Images