Unmanned aerial vehicle radio detection system and apparatus based on frequency hopping sequence sorting

Abstract

The application relates to the technical field of unmanned aerial vehicle radio detection, and particularly discloses an unmanned aerial vehicle radio detection system and device based on frequency hopping sequence sorting. The system comprises an environment sensing and feature extraction module, an adaptive threshold decision module, a frequency hopping sequence reconstruction module and a multi-dimensional feature fusion recognition module. Through real-time extraction of electromagnetic environment features, dynamic adjustment of a detection threshold, reconstruction of a frequency hopping sequence and fusion of multi-dimensional features, high-reliability detection and recognition of unmanned aerial vehicle frequency hopping signals are realized, and the detection sensitivity and anti-interference capability in a complex environment are effectively improved.

Description

Technical Field

[0001] This invention belongs to the field of unmanned aerial vehicle (UAV) radio detection technology, specifically relating to a UAV radio detection system and device based on frequency hopping sequence sorting. Background Technology

[0002] In the field of radio detection and target identification, the detection and sorting of non-cooperative communication signals is one of the core technologies, playing a crucial role, especially in security applications such as drone detection. Among these, signal sorting technology for drone targets using frequency-hopping communication has become a key research focus, aiming to accurately separate target signal sequences from complex electromagnetic environments.

[0003] In existing technologies, UAV detection systems based on frequency hopping sequence sorting typically employ signal detection and sorting algorithms with preset fixed thresholds. These methods face significant challenges in scenarios with dense electromagnetic interference: fixed thresholds cannot adapt to dynamic fluctuations in the channel environment, leading to an inherent contradiction between signal detection sensitivity and anti-interference performance; under strong interference, traditional algorithms are prone to generating numerous false alarms, such as misidentifying frequency hopping signals from civilian Wi-Fi devices as UAV signals; simultaneously, existing sorting methods lack adaptability to the time-varying characteristics of frequency hopping patterns, making it difficult to maintain stable sorting accuracy in complex electromagnetic environments.

[0004] With the increasing variety of drone types and communication modes, the parameter range of frequency-hopping signals is expanding. Traditional sorting algorithms based on fixed thresholds can no longer meet the requirements for detection reliability and environmental adaptability in practical applications. Especially in application scenarios with complex electromagnetic environments such as cities, where signal density is high and interference sources are diverse, the performance bottlenecks of existing technologies are more prominent, necessitating the development of novel frequency-hopping sequence sorting methods that can adapt to environmental changes. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a radio detection system and device for unmanned aerial vehicles based on frequency hopping sequence sorting, so as to overcome the defects of the fixed threshold detection method in the prior art, such as poor adaptability in complex electromagnetic environment, high false alarm rate and lack of adaptive ability to the time-varying characteristics of frequency hopping pattern.

[0006] The technical solution of this invention is as follows: The system includes an environmental perception and feature extraction module, an adaptive threshold decision module, a frequency hopping sequence reconstruction module, and a multi-dimensional feature fusion and recognition module. The environmental perception and feature extraction module is used to collect frequency hopping signals in the electromagnetic environment in real time and extract the time-domain features, frequency-domain features, and statistical features of the signals. This module includes a broadband receiver unit and a digital signal processing unit. The broadband receiver unit operates in a frequency band covering 400 MHz to 6 GHz. The digital signal processing unit uses a 128-point fast Fourier transform algorithm to perform spectral analysis on the received signals and simultaneously calculates three types of characteristic parameters: short-time energy, spectral centroid, and zero-crossing rate.

[0007] The adaptive threshold decision module connects to the output of the environmental perception and feature extraction module, and is used to dynamically adjust the decision threshold for signal detection and sorting based on the real-time extracted electromagnetic environment features. This module includes an environmental interference assessment submodule and a threshold calculation submodule. The environmental interference assessment submodule calculates the environmental interference index based on the current signal characteristics. The calculation of the interference index integrates three dimensions: signal bandwidth occupancy, background noise level, and the number of co-channel interferences. The threshold calculation submodule uses a nonlinear mapping function to convert the interference index into the corresponding detection threshold. This mapping function is trained through a large amount of measured data to ensure that a higher sensitivity threshold is used in low-interference environments and automatically switches to a higher anti-interference threshold in high-interference environments.

[0008] The frequency hopping sequence reconstruction module is connected to the output of the adaptive threshold decision module and is used to reconstruct the sequence and identify patterns of the frequency hopping signal after threshold decision. This module includes a time-series alignment submodule and a frequency hopping pattern analysis submodule. The time-series alignment submodule uses a sliding window mechanism to synchronize the discrete frequency hopping signals, with a window length of 50 milliseconds and a step size of 10 milliseconds. The frequency hopping pattern analysis submodule, based on the aligned time-series data, uses a hidden Markov model to model the state transition probabilities of the frequency hopping sequence, thereby identifying potential frequency hopping pattern regularities.

[0009] The multi-dimensional feature fusion and recognition module is connected to the output of the frequency hopping sequence reconstruction module, and is used to achieve accurate identification of UAV targets by integrating multi-dimensional features. This module includes a feature weighting submodule and a classification decision submodule. The feature weighting submodule dynamically assigns weights based on the classification contribution of each feature in historical data, where the weight coefficient for the frequency hopping period feature is 0.35, the weight coefficient for the dwell time feature is 0.25, the weight coefficient for the frequency set size feature is 0.2, and the weight coefficient for the signal modulation type feature is 0.2. The classification decision submodule uses the support vector machine algorithm to classify the weighted feature vectors and outputs the final target recognition result.

[0010] As one embodiment of the present invention, the digital signal processing unit in the environmental perception and feature extraction module also includes a signal preprocessing stage. This stage first performs a 4x downsampling process on the original signal, then performs bandpass filtering using an 8th-order Butterworth filter, and finally performs amplitude normalization to ensure the stability of subsequent feature extraction.

[0011] Furthermore, the nonlinear mapping function in the adaptive threshold decision module is specifically manifested as a piecewise linear function. When the environmental interference index is below 0.3, the detection threshold is set to 6 dB above the noise floor. When the environmental interference index is between 0.3 and 0.7, the detection threshold increases linearly with the interference index to 12 dB above the noise floor. When the environmental interference index is above 0.7, the detection threshold is fixed at 15 dB above the noise floor.

[0012] Furthermore, the hidden Markov model in the frequency hopping sequence reconstruction module contains 5 hidden states, which correspond to 5 typical frequency hopping modes in frequency hopping communication. The model parameters are obtained by training 2000 sets of known UAV frequency hopping sequence samples, and the update period of the state transition probability matrix is ​​30 seconds.

[0013] Furthermore, the support vector machine classifier in the multidimensional feature fusion recognition module uses the radial basis function as the kernel function. The width parameter of the kernel function is determined by optimization through the grid search method. The training sample set of the classifier includes frequency hopping signal data of 15 common UAVs and sample data of 10 typical interference signals.

[0014] In another embodiment of the present invention, the system further includes a feedback learning and optimization module, which is connected to the output of the multi-dimensional feature fusion recognition module and is used to continuously optimize the parameters of each module based on the system's performance in a real environment. The feedback learning and optimization module includes a performance evaluation submodule and a parameter adjustment submodule. The performance evaluation submodule statistically analyzes three indicators of the system in real time: detection rate, false alarm rate, and recognition accuracy. The parameter adjustment submodule, based on the changing trends of these indicators, uses a gradient descent algorithm to fine-tune the mapping function parameters in the adaptive threshold decision module and the feature weights in the multi-dimensional feature fusion recognition module, with an adjustment cycle of 24 hours.

[0015] Furthermore, the gradient descent algorithm in the feedback learning and optimization module adopts a learning rate of 0.01, and the magnitude of each parameter update does not exceed 5% of the original parameter value, ensuring the smooth evolution of system performance.

[0016] This invention also provides a UAV radio detection device based on frequency hopping sequence sorting. This device includes all the functional modules of the above-mentioned system and is integrated into a portable hardware platform. The hardware platform uses a multi-core digital signal processor as the main control chip, is equipped with two radio frequency receiving channels, each with a sampling rate of not less than 100 megahertz per second, a dynamic range of not less than 90 dB, and an overall power consumption controlled within 25 watts.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0018] 1. The electromagnetic environment features are acquired in real time through the environmental perception and feature extraction module, providing a data foundation for adaptive threshold decision-making; the adaptive threshold decision-making module can dynamically adjust the detection threshold according to the degree of environmental interference, effectively suppressing false alarms while ensuring detection sensitivity;

[0019] 2. The frequency hopping sequence reconstruction module models the frequency hopping pattern using a hidden Markov model, which significantly improves the tracking capability of time-varying frequency hopping sequences.

[0020] 3. The multi-dimensional feature fusion recognition module integrates multiple features for weighted decision-making, which greatly improves the accuracy and reliability of UAV recognition; the feedback learning and optimization module enables the system to continuously evolve and adapt to the ever-changing electromagnetic environment and new UAV signal characteristics.

[0021] 4. The entire system forms a complete technology chain from signal acquisition, feature extraction, threshold decision-making, sequence reconstruction to target recognition, and achieves highly reliable detection and recognition of UAV frequency hopping signals in complex electromagnetic environments. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the overall technical architecture of the UAV radio detection system based on frequency hopping sequence sorting proposed in this invention;

[0023] Figure 2 This is a schematic diagram of the core principle framework of the adaptive threshold decision module in this invention;

[0024] Figure 3 This is a logical flow diagram of the frequency hopping sequence reconstruction module in this invention;

[0025] Figure 4 This is a schematic diagram of the feature weighting and classification decision-making framework of the multi-dimensional feature fusion recognition module in this invention;

[0026] Figure 5 This is a schematic diagram illustrating the adaptive parameter adjustment principle of the feedback learning and optimization module in this invention. Detailed Implementation

[0027] Please refer to the attached document. Figures 1 to 5This embodiment details the specific hardware configuration and software implementation of a UAV radio detection system based on frequency-hopping sequence sorting. The core of this system lies in its ability to reliably detect and identify UAV frequency-hopping signals in complex electromagnetic environments through a series of precisely coordinated modules. The system hardware platform uses a multi-core digital signal processor as the central control unit, which integrates two independent RF receiving channels. Each RF receiving channel is equipped with a high-speed analog-to-digital converter, with a sampling rate precisely set to 100 megahertz per second, ensuring complete capture of the transient characteristics of the frequency-hopping signal. The dynamic range of the entire hardware platform is strictly maintained at 90 dB, enabling simultaneous processing of weak signals and strong interference signals. The total system power consumption is controlled to within 25 watts through a refined power management strategy, making it suitable for portable field deployment scenarios.

[0028] The environmental perception and feature extraction module serves as the system's front-end data entry point. This module comprises a broadband receiver unit and a digital signal processing unit. The broadband receiver unit operates in the frequency band from 400 MHz to 6 GHz and internally employs multi-stage low-noise amplifiers and adjustable filter banks to suppress out-of-band interference. The digital signal processing unit receives the raw intermediate frequency signal from the broadband receiver unit.

[0029] Before feature extraction, the digital signal processing unit (DSP) performs a rigorous signal preprocessing procedure. The preprocessing first downsamples the original signal by a factor of 4 to reduce subsequent computational complexity. The downsampled signal is then bandpass filtered by an 8th-order Butterworth filter. The passband range of this filter is dynamically configured according to the current scanning frequency to maximize the retention of target signal components. After filtering, the signal undergoes amplitude normalization to unify its peak amplitude to a preset reference level, ensuring that the extracted feature parameters are not affected by fluctuations in the absolute value of the received signal strength. After preprocessing, the DSP initiates the core feature extraction task.

[0030] It employs a 128-point Fast Fourier Transform algorithm to perform real-time spectral analysis on the input signal block, generating a frequency domain representation of the signal. Based on this frequency domain representation, the module synchronously calculates three key characteristic parameters: short-time energy, spectral centroid, and zero-crossing rate. Short-time energy reflects the total power of the signal within a specific time window; the spectral centroid describes the concentrated location of the signal energy in the frequency domain; and the zero-crossing rate characterizes the frequency with which the signal waveform crosses zero level in the time domain. These three characteristic parameters together constitute a preliminary characterization of the current electromagnetic environment and signal characteristics.

[0031] Please refer to the attached document. Figure 2The adaptive threshold decision module receives three types of feature parameters and other environmental data from the environmental perception and feature extraction module. Its core function is to dynamically calculate and output the optimal signal detection threshold based on real-time electromagnetic environment conditions. The module contains two cooperating sub-modules: an environmental interference assessment sub-module and a threshold calculation sub-module. The environmental interference assessment sub-module is responsible for quantifying the current level of interference in the environment.

[0032] It calculates a unified environmental interference index by integrating information from three dimensions: signal bandwidth occupancy (the proportion of bandwidth occupied by significant energy within the current frequency band); background noise level (derived through statistical analysis of signal sampling during quiet periods); and the number of co-channel interference signals (estimated by detecting multiple signal peaks occurring simultaneously at similar frequencies). The environmental interference assessment submodule uses a weighted summation model to fuse these three dimensions into an environmental interference index ranging from 0 to 1. The threshold calculation submodule is responsible for mapping the environmental interference index to specific detection thresholds.

[0033] This mapping is achieved through a pre-defined nonlinear function, specifically a piecewise linear function. When the calculated environmental interference index is below 0.3, it is considered a low-interference environment, and the detection threshold is set at 6 dB above the noise floor measured by the system to ensure high detection sensitivity. When the environmental interference index is between 0.3 and 0.7, the system is in a medium-interference environment, and the detection threshold increases linearly with the interference index, gradually increasing from 6 dB above the noise floor to 12 dB above the noise floor. When the environmental interference index is above 0.7, it indicates a high-interference environment, and the detection threshold is fixed at 15 dB above the noise floor to minimize false alarms. The parameters of this piecewise linear mapping function were determined through regression analysis of thousands of sets of measured data collected under different electromagnetic environments, ensuring the environmental adaptability of the threshold decision.

[0034] Please refer to the attached document. Figure 3The frequency hopping sequence reconstruction module receives frequency hopping signal pulses deemed valid by the adaptive threshold decision module. This module's task is to reorganize these discrete signal pulses, which hop in both time and frequency, into a continuous frequency hopping sequence and identify its inherent frequency hopping pattern. The module consists of a timing alignment submodule and a frequency hopping pattern analysis submodule. The timing alignment submodule uses a sliding window mechanism to process the input signal pulse stream. The sliding window length is set to 50 milliseconds, and the window slides forward in 10-millisecond increments. At each window position, the submodule performs time synchronization processing on all frequency hopping signal pulses captured within the window. Through precise measurement and compensation of the pulse arrival time difference, these pulses are aligned to a unified time reference, laying the foundation for subsequent sequence analysis. The frequency hopping pattern analysis submodule analyzes the time-aligned data, its core being the use of a Hidden Markov Model (HMM) to model and identify the patterns in the frequency hopping sequence. In this embodiment, the HMM is designed to contain five hidden states, each corresponding to one of five typical UAV frequency hopping communication modes summarized through extensive data analysis.

[0035] The model's parameters, including the initial state probability distribution, state transition probability matrix, and observation symbol probability distribution, were obtained through supervised training using 2000 known, labeled UAV frequency hopping sequence samples. The training process employed the Baum-Welch algorithm to maximize the probability of the model generating the observed sequence. During system operation, the frequency hopping pattern analysis submodule inputs the time-aligned frequency hopping sequence as observations into the trained Hidden Markov Model (HMM), and decodes the most probable state sequence using the Viterbi algorithm, thereby identifying the pattern followed by the current frequency hopping signal. To adapt to potential environmental changes, the state transition probability matrix of the HMM is not fixed but has an update mechanism with an update period set to 30 seconds. The system fine-tunes the model parameters based on the characteristics of recently received signals.

[0036] Please refer to the attached document. Figure 4 The multi-dimensional feature fusion and recognition module is the final target determination stage of the system. This module receives the frequency hopping pattern and other relevant feature information identified by the frequency hopping sequence reconstruction module. Internally, the module contains a feature weighting submodule and a classification decision submodule. The feature weighting submodule is responsible for evaluating the importance of the input multi-dimensional features and assigning corresponding weights.

[0037] These features mainly include: frequency hopping periodicity, which refers to the periodicity of the signal's dwell time at each frequency point; dwell time, which refers to the duration of the signal at a single frequency point; frequency set size, which represents the total number of different carrier frequencies used in the frequency hopping sequence; and signal modulation type, which is the modulation method identified through signal waveform analysis, such as frequency shift keying (FPS) or phase shift keying (PPS). Based on the contribution of these features to the classification results in historical data, the feature weighting submodule assigns them fixed weight coefficients: 0.35 for frequency hopping periodicity, 0.25 for dwell time, 0.2 for frequency set size, and 0.2 for signal modulation type.

[0038] The weighted feature combinations form a comprehensive feature vector. The classification decision submodule uses a support vector machine (SVM) algorithm as the classifier. This SVM uses radial basis functions (RBFs) as its kernel function to map the original feature space to a higher-dimensional space, thereby solving nonlinear classification problems. The width parameter of the RBF is optimized during the pre-training phase using a grid search method. The grid search evaluates the classification accuracy under different parameter combinations on the validation set to select the optimal value.

[0039] The classifier's training sample set is extremely rich, containing frequency-hopping signal data samples from 15 common drone models on the market, as well as 10 typical non-drone interference signal samples, ensuring the classifier has good discriminative ability. The trained support vector machine model takes the weighted feature vector as input, calculates the relationship between its position in the feature space and the decision hyperplane, and outputs the final classification result, determining whether the current signal originates from a drone target and further identifying its possible type.

[0040] Please refer to the attached document. Figure 5 The feedback learning and optimization module provides the entire system with the ability to continuously improve itself. This module connects to the output of the multi-dimensional feature fusion and recognition module and can access key parameters within the system. The module consists of a performance evaluation submodule and a parameter tuning submodule. The performance evaluation submodule continuously monitors the system's performance in a real-world operating environment. It provides real-time statistics on three core performance indicators: detection rate (the percentage of correctly identified drone signals); false alarm rate (the percentage of incorrectly identifying non-drone signals as drone signals); and recognition accuracy (the overall percentage of correctly classified signals).

[0041] These metrics are calculated by comparing the system output with real ground data obtained through other reliable means. The parameter adjustment submodule, based on the metric change trends provided by the performance evaluation submodule, uses a gradient descent algorithm to fine-tune the parameters of two key parts of the system: the parameters of the piecewise linear mapping function in the adaptive threshold decision module; and the feature weight coefficients of the feature weighting submodule in the multidimensional feature fusion and recognition module.

[0042] The learning rate of the gradient descent algorithm is set to 0.01 to ensure the stability of parameter updates. To prevent drastic fluctuations in system performance caused by sudden parameter changes, the adjustment range is limited to no more than 5% of the original parameter value during each update. The parameter adjustment trigger cycle is set to 24 hours. During relatively quiet periods each day, the system performs a parameter optimization process once based on performance data accumulated over the past 24 hours. This allows the system to slowly but continuously adapt to long-term changes in the electromagnetic environment and the emergence of new UAV signal characteristics.

[0043] This embodiment provides another specific implementation of a UAV radio detection system based on frequency hopping sequence sorting, focusing on enhanced processing strategies and resource scheduling mechanisms in extremely high-dynamic electromagnetic environments. Based on the above embodiment, the system hardware platform adds a dedicated hardware acceleration unit to its multi-core digital signal processor. This unit is specifically designed to perform Fast Fourier Transform (FFT) and matrix operations, reducing the computation latency of the 128-point FFT to 60% of its original value, thereby enabling it to handle UAV signals with higher hopping speeds.

[0044] The environmental perception and feature extraction module incorporates an adaptive filtering algorithm in the signal preprocessing stage. When a transient strong interference is detected, the module automatically switches to a 12th-order Chebyshev filter with a steeper cutoff characteristic to provide stronger out-of-band interference suppression. Simultaneously, the amplitude normalization process no longer uses a fixed reference level but instead employs a moving average based on the historical signal amplitude as a dynamic reference, further enhancing robustness against slow signal fading.

[0045] The environmental interference assessment submodule within the adaptive threshold decision module introduces an additional dimension of impulse interference density when calculating the interference index. Impulse interference density is assessed by statistically analyzing the number of narrow impulses exceeding the noise floor by 10 dB per unit time. This makes the interference index calculation more comprehensive, especially in environments with a large number of sudden impulse noises. The piecewise linear mapping function of the threshold calculation submodule is also correspondingly expanded, adding a range for extremely high interference. When the environmental interference index exceeds 0.9, the detection threshold is further raised to 18 dB above the noise floor, activating auxiliary detection logic based on signal sparsity analysis to ensure that high-confidence target signals can still be captured under extreme interference conditions.

[0046] The timing alignment submodule of the frequency hopping sequence reconstruction module employs a Kalman prediction-based compensation mechanism when facing severe signal loss. This mechanism utilizes an established Hidden Markov Model (HMM) to predict the frequency points of the lost signal time slots and inserts the predicted values ​​into the sequence with lower confidence to maintain sequence continuity for subsequent analysis. The HMM in the frequency hopping pattern analysis submodule can reduce the update period of its state transition probability matrix to 10 seconds in high-dynamic environments, accelerating the model's tracking speed of frequency hopping pattern changes.

[0047] In this embodiment, the feature weighting submodule of the multi-dimensional feature fusion and recognition module does not use fixed weights, but instead introduces a simple adaptive weighting mechanism. This mechanism fine-tunes the weights of various features based on their stability within the current time window (measured by the inverse of their variance), with features exhibiting higher stability receiving slightly higher temporary weights. The support vector machine classifier in the classification decision submodule, in addition to outputting the classification result, also outputs a score representing the classification confidence level. When the confidence level falls below a preset threshold, the sample is marked and stored in a review buffer, which can then be utilized by the feedback learning module.

[0048] The parameter tuning submodule of the feedback learning and optimization module can temporarily activate an emergency optimization mode when its gradient descent algorithm detects a rapid deterioration in performance metrics (such as the false alarm rate). In this mode, the learning rate can be briefly increased to 0.05, the parameter update limit is relaxed to 10%, and iterations are performed at shorter time intervals (e.g., 1 hour) to quickly converge to a new, better parameter set and cope with drastic environmental changes. The performance evaluation submodule strengthens the monitoring of new interference patterns. Once a persistent interference pattern that cannot be well distinguished by existing features is detected, an alert from the system administrator will be triggered, indicating that model retraining or feature engineering updates may be necessary.

Claims

1. A UAV radio detection system based on frequency hopping sequence partitioning, characterized in that, include: The environmental perception and feature extraction module is used to collect frequency hopping signals in the electromagnetic environment in real time and extract the time domain features, frequency domain features and statistical features of the signals. The environmental perception and feature extraction module includes a broadband receiver unit and a digital signal processing unit; An adaptive threshold decision module, connected to the output of the environmental perception and feature extraction module, is used to dynamically adjust the decision threshold for signal detection and sorting based on the real-time extracted electromagnetic environment features. The adaptive threshold decision module includes an environmental interference assessment submodule and a threshold calculation submodule. The environmental interference assessment submodule calculates an environmental interference index based on the current signal characteristics. The interference index calculation integrates three dimensions: signal bandwidth occupancy, background noise level, and the number of co-channel interferences. The threshold calculation submodule uses a nonlinear mapping function to convert the interference index into a corresponding detection threshold. The frequency hopping sequence reconstruction module is connected to the output of the adaptive threshold decision module and is used to perform sequence reconstruction and pattern recognition on the frequency hopping signal after threshold decision; the frequency hopping sequence reconstruction module includes a timing alignment submodule and a frequency hopping pattern analysis submodule. The timing alignment submodule uses a sliding window mechanism to synchronize the time of discrete frequency hopping signals; The frequency hopping pattern analysis submodule uses a hidden Markov model to model the state transition probability of the frequency hopping sequence based on the aligned time series data. A multi-dimensional feature fusion and recognition module is connected to the output of the frequency hopping sequence reconstruction module and is used to achieve accurate identification of UAV targets by integrating multi-dimensional features. The multi-dimensional feature fusion and recognition module includes a feature weighting submodule and a classification decision submodule. The feature weighting submodule dynamically allocates weights according to the classification contribution of each feature in historical data. The classification decision submodule uses the support vector machine algorithm to classify the weighted feature vectors; Multidimensional features include: Frequency hopping periodicity refers to the periodic pattern exhibited by the dwell time of a signal at each frequency point. Dwell time characteristic refers to the duration of a signal at a single frequency point; The frequency set size characteristic represents the total number of different carrier frequencies used by the frequency hopping sequence; Signal modulation type characteristics, which are the modulation methods identified through signal waveform analysis.

2. The drone radio detection system based on frequency hopping sequence partitioning of claim 1, wherein, The digital signal processing unit in the environmental perception and feature extraction module also includes a signal preprocessing stage; the signal preprocessing stage first performs a 4x downsampling process on the original signal, then performs bandpass filtering using an 8th-order Butterworth filter, and finally performs amplitude normalization.

3. The drone radio detection system based on frequency hopping sequence partitioning of claim 1, wherein, The nonlinear mapping function in the adaptive threshold decision module is specifically manifested as a piecewise linear function. When the environmental interference index is below 0.3, the detection threshold is set to 6 dB above the noise floor; when the environmental interference index is between 0.3 and 0.7, the detection threshold increases linearly with the interference index to 12 dB above the noise floor. When the environmental interference index is higher than 0.7, the detection threshold is fixed at 15 dB above the noise floor.

4. The drone radio detection system based on frequency hopping sequence partitioning of claim 1, wherein, The timing alignment submodule in the frequency hopping sequence reconstruction module uses a sliding window mechanism with a window length of 50 milliseconds and a step size of 10 milliseconds.

5. The drone radio detection system based on frequency hopping sequence partitioning of claim 1, wherein, The hidden Markov model in the frequency hopping sequence reconstruction module contains 5 hidden states, corresponding to 5 typical frequency hopping patterns in frequency hopping communication respectively; The parameters of the hidden Markov model are obtained by training 2000 groups of known UAV frequency hopping sequence samples; The update period of the state transition probability matrix of the hidden Markov model is 30 seconds.

6. The drone radio detection system based on frequency hopping sequence partitioning of claim 1, wherein, The feature weighting submodule in the multi-dimensional feature fusion recognition module assigns a weight coefficient of 0.35 to the frequency hopping period feature, a weight coefficient of 0.25 to the residence time feature, a weight coefficient of 0.2 to the frequency set size feature, and a weight coefficient of 0.2 to the signal modulation type feature.

7. The drone radio detection system based on frequency hopping sequence partitioning of claim 1, wherein, The support vector machine classifier in the multi-dimensional feature fusion recognition module uses a radial basis function as a kernel function; the width parameter of the kernel function is determined by grid search optimization; and the training sample set of the classifier contains frequency hopping signal data of 15 common UAVs and sample data of 10 typical interference signals.

8. The drone radio detection system based on frequency hopping sequence partitioning of claim 1, wherein, It also includes a feedback learning and optimization module; The feedback learning and optimization module is connected to the output end of the multi-dimensional feature fusion recognition module, and is used to continuously optimize the parameters of each module according to the performance of the system in the actual environment; The feedback learning and optimization module contains a performance evaluation submodule and a parameter adjustment submodule; The performance evaluation submodule real-time statistics the detection rate, false alarm rate and recognition accuracy of the system; The parameter adjustment submodule adjusts the mapping function parameters in the adaptive threshold decision module and the feature weights in the multi-dimensional feature fusion recognition module based on the change trend of these indicators using the gradient descent algorithm.

9. The drone radio detection system based on frequency hopping sequence partitioning of claim 8, wherein, The learning rate of the gradient descent algorithm in the feedback learning and optimization module is 0.01; the amplitude of each parameter update does not exceed 5% of the original parameter value; and the parameter adjustment period is 24 hours.

10. A drone radio detection apparatus based on frequency hopping sequence sorting, characterized by, All functional modules of the system in any one of claims 1 to 9 are included and integrated in a portable hardware platform; the hardware platform uses a multi-core digital signal processor as a main control chip, is equipped with 2 radio frequency receiving channels; the sampling rate of each channel is not less than 100 million times per second, the dynamic range is not less than 90 decibels, and the overall power consumption is controlled within 25 watts. All functional modules of the system in any one of claims 1 to 9 are included and integrated in a portable hardware platform; the hardware platform uses a multi-core digital signal processor as a main control chip, is equipped with 2 radio frequency receiving channels; the sampling rate of each channel is not less than 100 million times per second, the dynamic range is not less than 90 decibels, and the overall power consumption is controlled within 25 watts.

Images