Multi-uav-oriented image transmission signal detection method and device, equipment and medium

Abstract

This application discloses a method, apparatus, device, and medium for detecting image transmission signals from multiple UAVs, relating to the field of wireless radio frequency signal detection technology. The method includes: estimating the power spectral density of the multi-UAV radio frequency signals using the Welch improved periodogram method to obtain a power spectral density sequence; performing symmetric moving average smoothing and morphological filtering on the sequence to obtain a preprocessed spectral sequence; estimating local and global noise based on a noise candidate set to obtain an initial noise threshold; adjusting the initial noise threshold based on a signal strength quantization factor to obtain an adaptive detection threshold; locating the energy abrupt change boundary of the preprocessed spectral sequence using spectral gradients and generating a signal mask; and finally, aggregating and filtering the frequency points marked by the signal mask using DBSCAN clustering to obtain the image transmission signal detection result. This application enables accurate extraction of UAV image transmission signals in complex electromagnetic environments where multiple UAVs coexist.

Description

Technical Field

[0001] This application relates to the field of wireless radio frequency signal detection technology, and in particular to image transmission signal detection methods, devices, equipment and media for multiple UAVs. Background Technology

[0002] Because drone image transmission signals carry video or image information, they are characterized by large bandwidth, long transmission duration, and frequency hopping, making them important features in drone identification. Especially in scenarios where multiple drones coexist, image transmission signals can not only provide richer identification information but also improve the accuracy of drone target classification and the efficiency of surveillance.

[0003] Currently, drone identification mainly relies on statistical feature extraction of drone radio frequency (RF) signals and machine learning methods. For example, by analyzing signal bandwidth, modulation characteristics, and spectral features, and combining deep learning models such as convolutional neural networks (CNN) and gated recurrent units (GRU), it is possible to identify single-unit drones even with low signal-to-noise ratios.

[0004] However, existing methods have significant limitations in scenarios with multiple UAVs coexisting. Traditional statistical feature extraction methods struggle to handle issues such as signal superposition from multiple UAVs, feature confusion, and dynamic signal-to-noise ratio changes, leading to a significant increase in both false negative and false positive rates. While image transmission signals possess strong advantages in terms of time-frequency characteristics, most studies focus only on single-UAV identification, lacking adaptive detection mechanisms for signal superposition, enhanced electromagnetic interference, and dynamic feature changes in multi-UAV coexistence scenarios. Therefore, accurately extracting UAV image transmission signals in the complex electromagnetic environment of multiple UAVs coexisting has become an urgent problem to be solved. Summary of the Invention

[0005] The purpose of this application is to provide a method, apparatus, device and medium for detecting image transmission signals of multiple UAVs, aiming to solve the technical problem of how to accurately extract UAV image transmission signals in a complex electromagnetic environment where multiple UAVs coexist.

[0006] To achieve the above objectives, this application proposes a method for detecting image transmission signals for multiple unmanned aerial vehicles (UAVs), the method comprising:

[0007] Power spectral density sequence of radio frequency signals from multiple UAVs was obtained by estimating the power spectral density using the Welch improved periodogram method.

[0008] The power spectral density sequence is subjected to symmetric moving average smoothing and morphological filtering to obtain a preprocessed spectral sequence;

[0009] Local and global noise estimation is performed based on the noise candidate set in the preprocessed spectrum sequence to obtain the initial noise threshold;

[0010] The initial noise threshold is dynamically offset and adjusted according to the signal strength quantization factor to obtain an adaptive detection threshold.

[0011] The energy mutation boundary of the preprocessed spectral sequence is located using the spectral gradient algorithm, and a signal mask is generated based on the energy mutation boundary and the adaptive detection threshold.

[0012] The DBSCAN density clustering algorithm is used to perform connectivity aggregation and frequency band filtering on the frequency point set to be clustered marked by the signal mask, so as to obtain the image transmission signal detection result.

[0013] In one embodiment, the step of estimating the power spectral density of multiple UAV radio frequency signals using the Welch improved periodogram method to obtain a power spectral density sequence includes:

[0014] The radio frequency signals of multiple drones are divided into multiple overlapping signal segments;

[0015] Each signal segment is weighted using the Hanning window function;

[0016] Perform a Fourier transform on each weighted signal segment with a preset number of points to obtain a segmented spectrum sequence;

[0017] Calculate the initial segmented power spectrum corresponding to each signal segment based on the segmented spectrum sequence;

[0018] Calculate the energy normalization factor of the Hanning window function, and correct the amplitude of each initial segmented power spectrum according to the energy normalization factor to obtain the corrected segmented power spectrum;

[0019] The average value of all the corrected segmented power spectra is calculated to obtain the power spectral density sequence.

[0020] In one embodiment, the step of performing local and global noise estimation based on the noise candidate set in the preprocessed spectral sequence to obtain an initial noise threshold includes:

[0021] The frequency ranges below a preset power threshold in the preprocessed spectrum sequence are used as noise candidate sets.

[0022] The local noise estimate is obtained by calculating the local mean and local standard deviation of the noise candidate set through a preset sliding window.

[0023] When the noise candidate set is not empty, the global noise estimate is calculated by statistically analyzing the energy distribution of the noise candidate set.

[0024] When the noise candidate set is empty, the global noise estimate is calculated based on the preset low percentile coefficient.

[0025] The smaller of the local noise estimate and the global noise estimate is used as the initial noise threshold.

[0026] In one embodiment, the step of dynamically offsetting the initial noise threshold according to the signal strength quantization factor to obtain an adaptive detection threshold includes:

[0027] Effective components are selected from the preprocessed spectrum sequence according to the three-standard-deviation criterion, and the peak-to-peak value of the effective components is used as the signal strength quantization factor.

[0028] The real-time signal-to-noise ratio is determined based on the real-time power distribution of the preprocessed spectral sequence, and the scaling factor is determined based on the real-time signal-to-noise ratio.

[0029] Multiply the scaling factor by the signal strength quantization factor to obtain the base offset;

[0030] The local signal-to-noise ratio of the preprocessed spectral sequence is divided into multiple energy ranges by statistical quantiles, and an offset scaling factor is configured for each energy range.

[0031] The base offset is weighted and corrected according to the offset scaling factor to obtain the dynamic offset;

[0032] The initial noise threshold is adjusted by superimposing the dynamic offset to obtain a preliminary adjustment threshold.

[0033] The initial adjustment threshold is constrained by a lower limit based on the minimum offset of the preset threshold to obtain the adaptive detection threshold.

[0034] In one embodiment, the energy mutation boundary includes strong edge points and weak edge points;

[0035] The step of locating the energy abrupt change boundary of the preprocessed spectral sequence using a spectral gradient algorithm, and generating a signal mask based on the energy abrupt change boundary and the adaptive detection threshold includes:

[0036] The power decibel difference between adjacent frequency points in the preprocessed spectrum sequence is calculated using the spectrum gradient algorithm to obtain the spectrum gradient sequence.

[0037] The difference between the 75th percentile and the 25th percentile in the spectral gradient sequence is calculated using the interquartile range algorithm to obtain the dispersion value;

[0038] Strong edge points and weak edge points are determined based on the dispersion value and the mean gradient of the spectral gradient sequence, and an edge mask is generated based on the strong edge points and the weak edge points.

[0039] According to the statistical quantile interval of the local signal-to-noise ratio of each frequency point in the preprocessed spectrum sequence, a corresponding preset minimum signal-to-noise ratio threshold is matched. The local signal-to-noise ratio is the difference between the power decibel value of each frequency point and the local noise estimate obtained by synchronous calculation using a preset sliding window when calculating the local noise estimate.

[0040] Pseudo-signal frequency points with a signal-to-noise ratio lower than the preset minimum signal-to-noise ratio threshold are removed from the preprocessed spectrum sequence to obtain an effective frequency point set. The intersection of the edge mask and the effective frequency point set is then used to obtain a signal mask.

[0041] In one embodiment, the step of determining strong edge points and weak edge points based on the discrete value and the gradient mean of the spectral gradient sequence, and generating an edge mask based on the strong edge points and the weak edge points, includes:

[0042] When the mean gradient of the spectral gradient sequence is greater than the dispersion value, the gradient value corresponding to the preset first quantile in the spectral gradient sequence is selected as the high edge threshold.

[0043] When the mean gradient is less than or equal to the dispersion value, the gradient value corresponding to the preset second quantile in the spectral gradient sequence is selected as the high edge threshold;

[0044] Multiply the high edge threshold by a preset ratio to obtain the low edge threshold;

[0045] Frequency points in the spectral gradient sequence that are greater than or equal to the high edge threshold are identified as strong edge points, and frequency points that are greater than or equal to the low edge threshold and less than the high edge threshold are identified as weak edge points.

[0046] The weak edge points located within a preset neighborhood of the strong edge point are marked as the strong edge points, and an edge mask is generated based on the strong edge points and the weak edge points.

[0047] In one embodiment, the step of performing connectivity aggregation and frequency band filtering on the set of frequency points to be clustered marked by the signal mask using the DBSCAN density clustering algorithm to obtain the image transmission signal detection result includes:

[0048] Using the DBSCAN density clustering algorithm, core frequency points are marked in the set of frequency points to be clustered marked by the signal mask. The core frequency points refer to frequency points whose number of sample frequency points within a preset neighborhood radius is greater than or equal to the preset minimum number of clustering points.

[0049] Based on the density connectivity between the core frequency points, the corresponding effective frequency points are aggregated into multiple candidate signal clusters;

[0050] The signal bandwidth, duration, and power margin of each candidate signal cluster are compared with the preset image transmission bandwidth threshold, the preset minimum duration threshold, and the preset minimum power margin threshold, respectively.

[0051] When any one of the signal bandwidth, the duration, and the power margin is less than a preset value, the corresponding candidate signal cluster is marked as the target signal cluster;

[0052] The target signal cluster is removed from all the candidate signal clusters to obtain the image transmission signal frequency band, and the image transmission signal frequency band is used as the image transmission signal detection result.

[0053] Furthermore, to achieve the above objectives, this application also proposes an image transmission signal detection device for multiple unmanned aerial vehicles (UAVs), the device comprising:

[0054] The power spectral density estimation module is used to estimate the power spectral density of radio frequency signals from multiple UAVs using the Welch improved periodogram method, and obtain a power spectral density sequence.

[0055] The preprocessing module is used to perform symmetric moving average smoothing and morphological filtering on the power spectral density sequence to obtain a preprocessed spectral sequence.

[0056] The noise estimation module is used to perform local and global noise estimation based on the noise candidate set in the preprocessed spectrum sequence to obtain an initial noise threshold.

[0057] The threshold adjustment module is used to dynamically offset and adjust the initial noise threshold according to the signal strength quantization factor to obtain an adaptive detection threshold.

[0058] The mask generation module is used to locate the energy abrupt boundary of the preprocessed spectral sequence using a spectral gradient algorithm, and generate a signal mask based on the energy abrupt boundary and the adaptive detection threshold.

[0059] The frequency band filtering module is used to perform connectivity aggregation and frequency band filtering on the set of frequency points to be clustered marked by the signal mask using the DBSCAN density clustering algorithm, so as to obtain the image transmission signal detection result.

[0060] In addition, to achieve the above objectives, this application also proposes an image transmission signal detection device for multiple UAVs, the device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the image transmission signal detection method for multiple UAVs as described above.

[0061] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the image transmission signal detection method for multiple UAVs as described above.

[0062] In addition, to achieve the above objectives, this application also provides a computer program product, which includes a computer program that, when executed by a processor, implements the steps of the image transmission signal detection method for multiple UAVs described above.

[0063] One or more technical solutions proposed in this application have at least the following technical effects:

[0064] First, the power spectral density (PSD) of multi-UAV radio frequency signals is estimated using the Welch-improved periodogram method. The continuous signal is segmented, windowed, and the spectral average is calculated to obtain a PSD sequence. This step balances spectral resolution and variance, improving the accuracy of signal frequency characteristic analysis. Second, the PSD sequence undergoes symmetric moving average smoothing and morphological filtering to obtain a preprocessed spectral sequence. Smoothing reduces random noise fluctuations, and erosion and dilation operations suppress spike interference, making the spectrum more regular and providing a reliable foundation for subsequent noise separation. Then, local and global noise estimation is performed based on the noise candidate set in the preprocessed spectral sequence to obtain an initial noise threshold. This step accurately delineates signal and noise regions, reducing the risk of false positives and false negatives. Next, the initial noise threshold is dynamically offset and adjusted based on the signal strength quantization factor to obtain an adaptive detection threshold. By matching real-time signal strength changes, the threshold adapts to different signal-to-noise ratio scenarios, improving the effective signal recognition rate. Finally, the energy abrupt boundary of the preprocessed spectral sequence is located using a spectral gradient algorithm, and a signal mask is generated in conjunction with the adaptive detection threshold to mark potential effective signal regions, thus providing clear boundaries for signal screening. Finally, the DBSCAN density clustering algorithm is used to perform connectivity aggregation and frequency band filtering on the frequency point set to be clustered marked by the signal mask, eliminating discontinuous or low-quality signals to obtain the image transmission signal detection results. This enables accurate extraction of UAV image transmission signals in complex electromagnetic environments where multiple UAVs coexist. Attached Figure Description

[0065] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0066] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0067] Figure 1 This is a flowchart illustrating a first embodiment of the image transmission signal detection method for multiple UAVs in this application.

[0068] Figure 2 This is a schematic diagram of the clustering effect provided in Embodiment 1 of the image transmission signal detection method for multiple UAVs in this application;

[0069] Figure 3 This is a schematic diagram of the filtering effect provided in Embodiment 1 of the image transmission signal detection method for multiple UAVs in this application;

[0070] Figure 4 This is a schematic diagram of the signal detection effect provided in Embodiment 1 of the image transmission signal detection method for multiple UAVs in this application;

[0071] Figure 5 This is a flowchart illustrating Embodiment 2 of the image transmission signal detection method for multiple UAVs in this application.

[0072] Figure 6 This is a simplified flowchart of the image transmission signal detection method for multiple UAVs provided in Embodiment 2 of this application;

[0073] Figure 7 This is a schematic diagram of the module structure of the image transmission signal detection device for multiple UAVs according to an embodiment of this application;

[0074] Figure 8 This is a schematic diagram of the device structure of the hardware operating environment involved in the image transmission signal detection method for multiple UAVs in the embodiments of this application.

[0075] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0076] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0077] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0078] It should be noted that the executing entity of this application embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone, or an electronic device or signal detection system capable of performing the above functions. The following description uses a signal detection system as an example to illustrate this embodiment and the subsequent embodiments.

[0079] The user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0080] Based on this, embodiments of this application provide a method for detecting image transmission signals for multiple UAVs, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the image transmission signal detection method for multiple unmanned aerial vehicles (UAVs) according to this application.

[0081] In this embodiment, the image transmission signal detection method for multiple UAVs includes steps S10 to S60:

[0082] Step S10: The power spectral density of the radio frequency signals of multiple UAVs is estimated by using the Welch improved periodogram method to obtain the power spectral density sequence;

[0083] Step S20: Perform symmetric moving average smoothing and morphological filtering on the power spectral density sequence to obtain a preprocessed spectral sequence;

[0084] Step S30: Perform local and global noise estimation based on the noise candidate set in the preprocessed spectrum sequence to obtain the initial noise threshold;

[0085] Step S40: Dynamically offset and adjust the initial noise threshold according to the signal strength quantization factor to obtain an adaptive detection threshold;

[0086] Step S50: Locate the energy mutation boundary of the preprocessed spectral sequence using the spectral gradient algorithm, and generate a signal mask based on the energy mutation boundary and the adaptive detection threshold;

[0087] Step S60: The DBSCAN density clustering algorithm is used to perform connectivity aggregation and frequency band filtering on the frequency point set to be clustered marked by the signal mask to obtain the image transmission signal detection result.

[0088] It should be noted that the Welch Improved Periodogram Method is a classic power spectral density estimation method. By dividing the signal into overlapping segments, windowing each segment, calculating the periodogram, and averaging the results, it effectively reduces spectral estimation variance and spectral leakage, and is used to analyze the frequency characteristics of random signals. Multi-UAV RF signals refer to radio frequency signals emitted by multiple UAVs, their remote controllers, and ground stations within the same time and frequency band. These signals may overlap and be affected by electromagnetic interference. A power spectral density sequence is a signal frequency distribution sequence obtained through power spectral density estimation methods, reflecting the power intensity variations of the signal at various frequencies. A preprocessed spectral sequence is a spectral sequence obtained by suppressing noise and spike interference through smoothing and morphological filtering based on the power spectral density sequence, used for subsequent signal analysis. A noise candidate set is a set of frequency intervals in the preprocessed spectral sequence that may belong to background noise, used for subsequent noise level estimation. An initial noise threshold is a preliminary power threshold calculated based on local and global noise estimations to distinguish between signals and noise. The signal strength quantization factor is a coefficient used to measure the change in signal power relative to the noise level. It is used to dynamically adjust the noise threshold to adapt to different signal-to-noise ratio scenarios.

[0089] Adaptive detection threshold refers to a threshold dynamically adjusted by the signal strength quantization factor, which can change with the signal-to-noise ratio to achieve real-time identification and filtering of valid signals. Spectral gradient algorithm refers to a method that identifies locations of signal energy abrupt changes by calculating the rate of change of spectral power at adjacent frequency points. Energy abrupt change boundaries refer to locations where signal power changes rapidly in spectral gradient analysis; these locations typically correspond to the frequency boundaries of valid signals. Signal mask refers to an array in a spectral sequence that uses binary labels to represent valid signal regions, where 1 indicates that the frequency point belongs to a potentially valid signal, and 0 indicates a non-signal region. DBSCAN density clustering algorithm (Density-Based Spatial Clustering of Applications with Noise) is a density-based unsupervised clustering algorithm that divides frequency points into clusters based on neighborhood and minimum sample size, capable of identifying signal frequency bands of arbitrary shapes and filtering noise points. The set of frequency points to be clustered refers to the set of frequency points marked as valid by the signal mask, used as input to the DBSCAN algorithm to form continuous valid signal clusters. The image transmission signal detection result refers to the frequency band and feature information of the UAV image transmission signal that is finally determined after signal preprocessing, adaptive threshold filtering and density clustering, which is used for subsequent UAV identification and analysis.

[0090] Understandably, the signal detection system first estimates the power spectral density of received multi-UAV RF signals using the Welch improved periodogram method. The signal is divided into multiple overlapping segments, each windowed and subjected to a Fourier transform. The power spectrum of each segment is then amplitude-corrected and averaged to obtain a power spectral density sequence reflecting the power distribution at each frequency point. This quantifies the signal strength at different frequencies and reduces the impact of single-segment spectral fluctuations on the overall estimate. Second, the system smooths the power spectral density sequence using a symmetric moving average. A sliding window is used to calculate the neighborhood mean, reducing random noise fluctuations. Morphological filtering (dilation and erosion operations) is applied to eliminate small-scale spike interference, resulting in a more continuous and regular preprocessed spectral sequence for more accurate signal-to-noise differentiation. Then, the system constructs a noise candidate set based on low-power regions in the preprocessed spectral sequence. Local means and standard deviations are calculated within the sliding window, and the global frequency distribution is analyzed to obtain a global noise estimate. The smaller of the local and global estimates is then used as the initial noise threshold, providing a reliable signal-to-noise boundary and reducing the probability of false detections and false negatives.

[0091] Next, the signal detection system dynamically adjusts the initial noise threshold using a signal strength quantization factor. Combining real-time power distribution and signal-to-noise ratio changes, it weights and corrects the threshold to obtain an adaptive detection threshold. This allows the threshold to automatically adjust with signal strength fluctuations, ensuring that valid signals are not misidentified as noise. Subsequently, the system analyzes the power changes of adjacent frequency points in the preprocessed spectrum sequence using a spectral gradient algorithm, identifying energy abrupt change boundaries (including locations of significant and weak changes). It then generates a signal mask based on the adaptive detection threshold, marking potentially valid signal regions and providing initial localization for subsequent signal filtering. Finally, the system performs DBSCAN density clustering on the frequency points marked by the signal mask, aggregating connectivity among neighboring frequency points. Based on the clustering results, it filters out signal bands that meet persistence and frequency band requirements, generating the final image transmission signal detection result. This enables accurate extraction of image transmission signal frequency bands and feature information from the coexisting radio frequency signals of multiple UAVs.

[0092] As an example, the step of estimating the power spectral density of multi-UAV radio frequency signals using the Welch improved periodogram method to obtain a power spectral density sequence includes: dividing the multi-UAV radio frequency signals into multiple overlapping signal segments; weighting each signal segment using a Hanning window function; performing a Fourier transform on each weighted signal segment with a preset number of points to obtain a segmented spectrum sequence; calculating the initial segmented power spectrum corresponding to each signal segment based on the segmented spectrum sequence; calculating the energy normalization factor of the Hanning window function, and correcting the amplitude of each initial segmented power spectrum based on the energy normalization factor to obtain a corrected segmented power spectrum; and calculating the average value of all corrected segmented power spectra to obtain a power spectral density sequence.

[0093] It should be noted that signal segmentation refers to dividing the received multi-UAV RF signals into several continuous and partially overlapping small segments in chronological order. Each segment is used independently for power spectrum analysis to reduce the impact of single-segment signal instability on spectrum estimation. The Hanning window function is a weighting function used to weight the sampling points in the signal segments to suppress spectral leakage, making the Fourier transform spectrum smoother and reducing edge effects. The preset number of sampling points refers to the number of sampling points selected when performing a Fourier transform on each signal segment, used to determine the frequency resolution and the calculation accuracy of the Fourier transform. The Fourier transform is a method of converting a time-domain signal into a frequency-domain representation, generating the spectral information of the signal segment by calculating the amplitude and phase of each frequency component. The segmented spectrum sequence refers to the frequency component amplitude sequence obtained after Fourier transforming each signal segment, used to describe the power distribution of the signal segment at different frequencies. The initial segmented power spectrum is the power spectrum value sequence calculated from the segmented spectrum sequence, representing the energy magnitude at each frequency point, without amplitude correction. The energy normalization factor is a correction coefficient calculated based on the energy characteristics of the Hanning window function. It is used to correct the amplitude of the initial segmented power spectrum, making the power spectra of different segments comparable in energy and ensuring the accuracy of the averaging process.

[0094] Understandably, the signal detection system first divides the received multi-UAV RF signals into multiple overlapping signal segments in chronological order. The length of each segment is pre-set according to the analysis needs, and adjacent segments partially overlap to ensure a smooth transition during segmentation and reduce the impact of boundary effects on spectrum calculation. Second, the system applies a Hanning window function to each signal segment for weighting, multiplying the amplitude of each sampling point in the segment by the corresponding window function coefficient, causing the amplitude at both ends of the segment to gradually decay, thereby suppressing spectral leakage and enhancing frequency component resolution. Then, the system performs a preset number of discrete Fourier transforms on each weighted signal segment. By accumulating complex exponential multiplications of the sampling points in the segment, the system calculates the amplitude and phase at each frequency point, obtaining a segmented spectrum sequence to represent the energy distribution of that segment at different frequencies.

[0095] Subsequently, the system calculates the initial segmented power spectrum for each signal segment based on the segmented spectral sequence. This is achieved by squaring the amplitude at each frequency point and dividing by the number of sampling points to obtain the power distribution value. Next, the system calculates the energy normalization factor of the Hanning window function. By summing the squares of the window function coefficients and taking the square root, an energy scaling factor is obtained to compensate for the energy scaling introduced by windowing. This factor is then applied to the initial segmented power spectrum for amplitude correction, ensuring that the corrected segmented power spectra maintain energy consistency and comparability. Finally, the system averages all corrected segmented power spectra at their corresponding frequency points to obtain the overall power spectral density sequence. This provides a stable, continuous spectral basis that can be used for subsequent signal preprocessing and feature extraction, thereby improving the accuracy and robustness of the entire multi-UAV RF signal detection.

[0096] The expression for power spectral density estimation is:

[0097]

[0098] in, Power spectral density; Number of signal segments For signal Total length, For the length of the window function, This represents the number of overlapping samples in adjacent segments. This is a floor function; The window function energy normalization factor; Sampling rate; The number of points in the Fourier transform; The Hanning window was selected to suppress spectral leakage and improve frequency resolution.

[0099] As an example, the step of performing local and global noise estimation based on the noise candidate set in the preprocessed spectrum sequence to obtain an initial noise threshold includes: taking the frequency range below a preset power threshold in the preprocessed spectrum sequence as the noise candidate set; calculating the local mean and local standard deviation of the noise candidate set through a preset sliding window to obtain a local noise estimate; calculating a global noise estimate by statistically analyzing the energy distribution of the noise candidate set when the noise candidate set is not empty; calculating a global noise estimate based on a preset low percentile coefficient when the noise candidate set is empty; and taking the smaller value between the local noise estimate and the global noise estimate as the initial noise threshold.

[0100] It should be noted that the preset power threshold is a power value set during the signal preprocessing stage to distinguish between frequency ranges that may belong to background noise and potentially effective signal ranges. Frequency ranges below this threshold are identified as noise candidate sets. The preset sliding window refers to the number of consecutive frequency points used when calculating the local statistical characteristics of the noise candidate set. It forms a sliding window on the spectrum to calculate the local mean and local standard deviation, thereby reflecting changes in local noise levels. The local mean is the arithmetic average of the power of the frequency points in the noise candidate set within the sliding window, used to estimate the average noise power within that window range. The local standard deviation is the standard deviation calculated from the power of the frequency points in the noise candidate set within the sliding window, used to reflect the degree of fluctuation in local noise power.

[0101] Local noise estimation refers to the local noise level calculated by combining the local mean and local standard deviation, used to describe the noise intensity characteristics within a small range of the spectrum. Energy distribution refers to the distribution of power values ​​at various frequency points across the entire frequency range of the noise candidate set, used to statistically analyze global noise characteristics. Global noise estimation refers to the average noise level across the entire spectrum range calculated based on the energy distribution of the noise candidate set or a preset low percentile coefficient, used to provide a reference for the initial noise threshold. The preset low percentile coefficient is used to select the lowest percentile (e.g., 10%) of the power value at a frequency point as the benchmark for global noise estimation when the noise candidate set is empty, in order to exclude extreme signal interference. The initial noise threshold is the smaller value between the local noise estimation and the global noise estimation, used as a power benchmark to distinguish between valid signals and noise, and is used for subsequent adaptive threshold adjustment and signal filtering.

[0102] Understandably, firstly, the signal detection system filters out continuous frequency ranges below a preset power threshold from the preprocessed spectrum sequence, using these as a noise candidate set to centrally analyze potential background noise areas and eliminate interference from potentially valid signals. Secondly, the system uses a preset sliding window to perform sliding calculations within the noise candidate set. For the power of each frequency point within the window, the arithmetic mean and standard deviation are calculated to obtain a local noise estimate. This accurately reflects the noise intensity and fluctuations of each frequency band, avoiding the impact of a single outlier on the overall noise assessment. The formula for calculating the local noise estimate is as follows:

[0103]

[0104] in, It is a local mean; Local standard deviation; This is the result of local noise estimation; The weighting coefficients are dynamically adjusted in a positive direction with the local signal-to-noise ratio (SNR) to reduce noise estimation bias.

[0105] Then, the system determines whether the noise candidate set is empty. If the noise candidate set is not empty, the system calculates a global noise estimate by statistically analyzing the power distribution of the entire candidate set, which reflects the average noise level across the entire frequency spectrum. The formula for calculating the global noise estimate is as follows:

[0106]

[0107] in, This refers to the noise candidate set; This is the result of global noise estimation; These are weighting coefficients, which vary with... The sample dispersion is increased to improve the universality of global noise estimation.

[0108] If the noise candidate set is empty, the system selects the power value of a low-power frequency point from the preprocessed spectrum sequence based on a preset low percentile coefficient as the global noise estimate, in order to eliminate the interference of high-power signals on noise calculation. Finally, the system compares the local noise estimate with the global noise estimate and takes the smaller value as the initial noise threshold, which is used for the calculation of the subsequent adaptive detection threshold, ensuring that the threshold can accurately distinguish between noise and valid signals, and can adapt to the noise fluctuation characteristics of different frequency bands.

[0109] As an example, the step of dynamically offsetting the initial noise threshold according to the signal strength quantization factor to obtain an adaptive detection threshold includes: selecting effective components from the preprocessed spectrum sequence according to the three-standard-deviation criterion, and using the peak-to-peak value of the effective components as the signal strength quantization factor; determining the real-time signal-to-noise ratio (SNR) based on the real-time power distribution of the preprocessed spectrum sequence, and determining a scaling factor based on the real-time SNR; multiplying the scaling factor by the signal strength quantization factor to obtain a base offset; dividing the local SNR of the preprocessed spectrum sequence into multiple energy intervals using statistical quantiles, and configuring an offset scaling factor for each energy interval; weighting and correcting the base offset according to the offset scaling factor to obtain a dynamic offset; superimposing and adjusting the initial noise threshold according to the dynamic offset to obtain a preliminary adjustment threshold; and constraining the preliminary adjustment threshold with a preset minimum offset to obtain an adaptive detection threshold.

[0110] It should be noted that the three-standard-deviation criterion means that when calculating the effective components, the mean and standard deviation of the preprocessed spectrum sequence are first calculated, and then frequency points within the range of the mean plus or minus three standard deviations are selected to exclude abnormally large or small signal values. The effective components refer to the power values ​​of the frequency points selected from the preprocessed spectrum sequence according to the three-standard-deviation criterion; these frequency points are considered to primarily represent the normal fluctuation range of the signal. Peak-to-peak value refers to the difference between the maximum positive value and the minimum negative value of the signal within one period in the effective components, used to quantify the range of signal amplitude variation. The signal strength quantization factor is a value calculated from the peak-to-peak value of the effective components, used to represent the signal strength level and as a benchmark for subsequent noise threshold adjustment. Real-time power distribution refers to the power distribution of each frequency point in the preprocessed spectrum sequence at the current sampling time, used to reflect the change in signal strength over time. Real-time signal-to-noise ratio (SNR) is the instantaneous value of the ratio of signal power to noise power within the current frequency point or band, used to measure the signal strength relative to noise. The scaling factor is an adjustment factor calculated based on the real-time SNR; its magnitude decreases as the SNR increases, used to control the amplitude of the base offset. The base offset is an initial adjustment value obtained by multiplying the signal strength quantization factor by the scaling factor. It is used to offset the signal from the initial noise threshold to enhance signal detection capability.

[0111] Statistical quantiles refer to the percentage division of local signal-to-noise ratio (SNR), such as high, medium, and low intervals, used to group frequency points at different energy levels. Local SNR refers to the ratio of signal power to noise power within a local sliding window or local frequency range of a preprocessed spectral sequence, used to describe the signal quality of each small segment. Energy intervals refer to different levels of local SNR divided according to statistical quantiles, such as high-energy, medium-energy, and low-energy intervals, used for threshold adjustment for different signal intensities. Offset scaling factor refers to a weighting factor set for each energy interval, used to correct the base offset, making the offset more closely match the characteristics of different signal intensity intervals. Dynamic offset refers to the adjustment value obtained by weighting and correcting the base offset using the offset scaling factor, used for dynamic adjustment based on the initial noise threshold. Preliminary adjustment threshold is the threshold obtained by superimposing the initial noise threshold and the dynamic offset, used as a temporary reference to distinguish noise from signal. Preset minimum offset is a set value used to limit the lower limit of the preliminary adjustment threshold to avoid false noise detection due to an excessively low threshold, thus ensuring the reliability of the adaptive detection threshold.

[0112] Understandably, the signal detection system first calculates the mean and standard deviation of the preprocessed spectral sequence. Based on the three-standard-deviation criterion, it selects frequency points within the range of the mean plus or minus three standard deviations as effective components. This eliminates abnormally high or low noise values, ensuring the selected frequency points accurately reflect the main fluctuation characteristics of the signal. Then, the system calculates the peak-to-peak value of these effective components over the entire period and uses it as a signal strength quantization factor to quantify the maximum amplitude change of the signal in the current spectrum, providing a benchmark for subsequent threshold adjustment. Secondly, the system analyzes the real-time power distribution of the preprocessed spectral sequence, calculating the real-time signal-to-noise ratio (SNR) by comparing the power and noise levels at each frequency point. Based on the real-time SNR, it determines a scaling factor, assigning a smaller scaling factor to frequency bands with higher SNR to prevent excessive threshold offset and false noise detection, while ensuring more adequate adjustment in low SNR regions. Finally, the system multiplies the scaling factor by the signal strength quantization factor to obtain a base offset, which is used for initial adjustment of the initial noise threshold to enhance its adaptability to the effective signal.

[0113] Next, the system divides the local signal-to-noise ratio of the preprocessed spectral sequence into multiple energy ranges (such as high, medium, and low ranges) using statistical quantiles, and configures an offset scaling factor for each energy range to weightedly correct the base offset. This makes the threshold adjustment for different signal strength regions more precise and matches the actual signal distribution. Subsequently, the system uses the weighted corrected offset as a dynamic offset and adds it to the initial noise threshold to obtain the preliminary adjustment threshold. This allows for dynamic matching of signal strength changes while maintaining the overall noise baseline, improving the recognition rate of effective signals. Finally, the system imposes a lower limit constraint on the preliminary adjustment threshold based on a preset minimum offset to prevent noise from being misjudged as a signal due to an excessively low threshold, thus obtaining the final adaptive detection threshold for subsequent accurate extraction of effective image transmission signal frequency bands.

[0114] As an example, the step of performing connectivity aggregation and frequency band filtering on the set of frequency points to be clustered marked by the signal mask using the DBSCAN density clustering algorithm to obtain the image transmission signal detection result includes: using the DBSCAN density clustering algorithm, marking core frequency points in the set of frequency points to be clustered marked by the signal mask, wherein the core frequency points are frequency points whose number of sample frequency points within a preset neighborhood radius is greater than or equal to a preset minimum number of clustering points; aggregating the corresponding effective frequency point sets into multiple candidate signal clusters according to the density connectivity between the core frequency points; comparing the signal bandwidth, duration, and power margin of each candidate signal cluster with a preset image transmission bandwidth threshold, a preset minimum duration threshold, and a preset minimum power margin threshold, respectively; when any one of the signal bandwidth, duration, and power margin is less than a preset value, marking the corresponding candidate signal cluster as a target signal cluster; removing the target signal cluster from all the candidate signal clusters to obtain the image transmission signal frequency band, and using the image transmission signal frequency band as the image transmission signal detection result.

[0115] It should be noted that core frequency points refer to frequency points in the set of frequency points to be clustered whose neighborhood contains a number of sample frequency points greater than or equal to the preset minimum clustering number. These points are usually located in dense signal regions and are the basis for forming candidate signal clusters. The preset neighborhood radius refers to the frequency range defined in the DBSCAN algorithm, used to determine whether adjacent frequency points belong to the same density region. Frequency points with a distance less than this radius are considered to be adjacent. The number of sample frequency points refers to the number of frequency points to be clustered within the preset neighborhood radius, used to determine whether the point is a core frequency point. The preset minimum clustering number is the threshold set in the DBSCAN algorithm for determining core points, i.e., the minimum number of frequency points required in the neighborhood to mark a frequency point as a core frequency point.

[0116] Density connectivity refers to the criterion used in the DBSCAN algorithm to determine whether two core frequency points belong to the same cluster. If the frequency distance between two core frequency points is less than or equal to a preset neighborhood radius, they are considered directly connected. If there is a path connecting them via other core frequency points, such that the frequency distance between any adjacent points on the path does not exceed the preset neighborhood radius, they are considered indirectly connected. Through this direct or indirect connection, the system can determine which core frequency points can be aggregated together to form continuous and dense frequency clusters. A candidate signal cluster refers to a set of multiple frequency points obtained through density connectivity aggregation. Each set contains continuous effective frequency points that may belong to the same signal, used for further screening. Signal bandwidth refers to the width of the frequency range in the candidate signal cluster, i.e., the difference between the highest and lowest frequencies, used to determine whether the signal meets the bandwidth requirements of the image transmission signal. Duration refers to the shortest continuous effective time of a candidate signal cluster in the time domain, used to determine whether the signal is stable and identifiable.

[0117] Power margin refers to the difference between the minimum power of a candidate signal cluster and the noise baseline within a frequency range, used to measure whether the signal strength is sufficiently prominent. The preset transmission bandwidth threshold refers to the minimum and maximum allowed bandwidth range for the transmission signal (e.g., 2MHz to 20MHz), used to filter signal clusters that do not meet the bandwidth requirements. The preset minimum duration threshold refers to the shortest duration required for the signal to persist (e.g., 10 milliseconds), used to eliminate interference or spurious signals with excessively short durations. The preset minimum power margin threshold refers to the minimum power difference between the signal and noise (e.g., 3 dB), used to eliminate frequency clusters with insufficient power or that are difficult to identify. The target signal cluster refers to any candidate signal cluster whose characteristics (signal bandwidth, duration, or power margin) are below the preset threshold; these clusters are considered inconsistent with the characteristics of a transmission signal and need to be eliminated. The transmission signal frequency band refers to the set of frequency ranges remaining after filtering candidate signal clusters and eliminating target signal clusters, representing the finally detected valid UAV transmission signals.

[0118] Understandably, firstly, the signal detection system calculates the number of sample frequency points within a preset neighborhood radius for each frequency point in the cluster marked by the signal mask. Frequency points with a neighborhood number greater than or equal to the preset minimum clustering number are marked as core frequency points. This ensures that only regions with concentrated frequencies that may belong to valid signals are selected, providing a foundation for subsequent aggregation. Secondly, the system aggregates based on the density connectivity between core frequency points: specifically, the system connects each core frequency point to all core frequency points in its neighborhood. If two core frequency points have a direct neighborhood relationship or are indirectly connected through other core frequency points, they are considered to be in the same connected cluster, thus forming candidate signal clusters. Each cluster contains mutually adjacent and continuous frequency points, in order to group frequency points that may belong to the same signal together.

[0119] Then, the system calculates the signal bandwidth, duration, and power margin for each candidate signal cluster, and compares these indicators with preset transmission bandwidth thresholds, preset minimum duration thresholds, and preset minimum power margin thresholds. When any one of these thresholds is lower than the threshold, the candidate signal cluster is marked as a target signal cluster to eliminate low-quality or interference signals. Finally, the system removes the frequency points marked as target signal clusters from all candidate signal clusters, retains the remaining frequency points, and combines them into a transmission signal frequency band, outputting it as the transmission signal detection result. This ensures that the finally detected transmission signal has sufficient bandwidth, stable duration, and significant power margin, thereby achieving high-precision identification and reliable extraction of multi-UAV transmission signals.

[0120] Please refer to Figure 2 , Figure 2 This diagram illustrates the clustering effect of the image transmission signal detection method for multiple UAVs in this application. The horizontal axis represents frequency (in GHz), and the vertical axis represents power spectral density (in dB / Hz). The blue solid line represents the smoothed power spectral sequence after symmetrical moving average and smoothing filtering, visually demonstrating the energy fluctuations of multiple UAV signals in the frequency domain. The green dashed line represents the initial noise threshold (noise estimation) estimated through the noise candidate set, serving as a reference benchmark for environmental noise floor. The red solid line represents the adaptive detection threshold (signal threshold) obtained after dynamic offset adjustment based on the signal strength quantization factor. Located above the noise estimation line, it accurately distinguishes between signal and noise regions. The yellow dots scattered on the blue peaks represent the identified valid signal points. These points are frequency components extracted after signal masking constraints, satisfying the energy mutation boundary characteristics and meeting the signal-to-noise ratio requirements. They constitute the frequency point set to be clustered by the DBSCAN density clustering algorithm, providing a reliable data foundation for finally aggregating the complete image transmission signal frequency band and eliminating false signals.

[0121] As an example, the steps of performing symmetric moving average smoothing and morphological filtering on the power spectral density sequence to obtain a preprocessed spectral sequence include: converting the power spectral density sequence to a decibel scale to obtain a decibel power spectral sequence; calculating the local mean of non-boundary points in the decibel power spectral sequence within a preset smoothing window using a symmetric moving average algorithm to obtain a smoothed power spectral sequence; constructing a one-dimensional rectangular structuring element based on a preset interference bandwidth and a preset frequency resolution; performing a point-by-point erosion operation on the smoothed power spectral sequence based on the one-dimensional rectangular structuring element, taking the minimum value in the neighborhood to obtain an eroded sequence; and performing a point-by-point dilation operation on the eroded sequence based on the one-dimensional rectangular structuring element, taking the maximum value in the neighborhood to obtain a preprocessed spectral sequence.

[0122] It should be noted that the decibel power spectrum sequence refers to the conversion of the power spectral density sequence into a representation in decibels. This is used to enhance the discernibility of low-power signals and facilitate subsequent smoothing and filtering, as shown below:

[0123]

[0124] in, This is a decibel power spectrum sequence. This represents the power spectral density.

[0125] The symmetric moving average algorithm refers to a method that averages data points of equal length taken on both sides of the current frequency point within a preset window. It is used to smooth out noise with random fluctuations in a power spectrum sequence, as shown below:

[0126]

[0127] in, To smooth the power spectrum, The power spectrum is a sequence of decibels; the smoothing window is set to the current point. k Take a length of centered at the length of W The symmetric neighborhood is calculated by... The mean value achieves local smoothing.

[0128] The preset smoothing window refers to the number of frequency points selected in the symmetric moving average calculation, used to define the data range included when calculating the local mean. The local mean of non-boundary points refers to the average power value of the center frequency point calculated within the smoothing window, used to reduce the impact of random fluctuations at a single frequency point on the entire spectrum.

[0129] A smoothed power spectrum sequence refers to a power spectrum sequence processed by symmetric moving average, exhibiting a more regular energy distribution and effectively suppressing noise fluctuations. The preset interference bandwidth refers to the frequency range set in morphological filtering, used to determine the neighborhood size, thereby identifying and eliminating local spikes or interference signals. The preset frequency resolution refers to the actual frequency interval represented by each frequency point in the power spectrum sequence, used to calculate the actual frequency range covered by the filter structure element. A one-dimensional rectangular structure element refers to a fixed-length window used for erosion and dilation operations in morphological filtering, composed of consecutive 1s, used to define the neighborhood size and the range of morphological operations, represented as follows:

[0130]

[0131] in, The length of the structuring element; B For interference bandwidth; This is for rounding up to the nearest integer. Frequency resolution; Ensure that the length of the structuring element is not less than 3 to avoid invalid morphological operations.

[0132] The erosion sequence refers to the sequence obtained by applying a rectangular structuring element to a smooth power spectrum sequence point by point. The minimum value in the neighborhood of each point is taken to remove local spike interference.

[0133] Understandably, the signal detection system first converts the power spectral density sequence point by point to a decibel scale. The power value at each frequency point is converted to decibel form by multiplying by the logarithm of 10, resulting in a decibel power spectral sequence. This enhances the identifiability of low-power signals and facilitates subsequent smoothing and filtering operations. Second, the system applies a symmetric moving average algorithm to the decibel power spectral sequence. For each non-boundary frequency point, it takes the left and right frequency points within a preset smoothing window (e.g., the window length covers 5 to 11 frequency points) centered on that point, calculates the average of these frequency points as the new power value at the center point, forming a smoothed power spectral sequence. This suppresses random noise fluctuations and small spikes, making the overall spectrum more regular. Then, the system constructs a one-dimensional rectangular structuring element based on a preset interference bandwidth and preset frequency resolution to determine the actual coverage range of the filtering neighborhood. It then performs an erosion operation point by point on the smoothed power spectral sequence, taking the minimum power value within the neighborhood covered by the structuring element at each point to generate an eroded sequence. This eliminates local spike interference and highlights continuous signal characteristics. Next, the system uses the same one-dimensional rectangular structuring element to perform dilation operation on the erosion sequence point by point. The maximum value in the neighborhood of each point is taken to generate the final preprocessed spectrum sequence. This step not only compensates for the signal edge shrinkage caused by the erosion operation, but also maintains the continuity and amplitude characteristics of the effective signal, providing a stable and clean spectrum basis for subsequent signal threshold judgment and feature extraction.

[0134] Please refer to Figure 3 , Figure 3 This diagram illustrates the filtering effect of the image transmission signal detection method for multiple UAVs according to Embodiment 1 of this application. The horizontal axis represents the frequency range (GHz), and the vertical axis represents the power spectral density (dB / Hz). The diagram contains three curves: the original power spectrum curve shows the change of unprocessed signal power with frequency, reflecting the superposition of multiple UAV signals and the presence of noise; the smoothed power spectrum curve shows the power spectrum after symmetrical moving average processing, where noise fluctuations are suppressed and the overall trend is more stable; the power spectrum curve after morphological filtering further eliminates local spike interference on the basis of smoothing, while maintaining the main signal peaks, showing the separation effect of effective signal frequency bands and noise after erosion and dilation processing. The reference power level used for signal detection is marked with a threshold line in the diagram, intuitively demonstrating the enhancement effect of the filtering step on the extraction of radio frequency signal features of multiple UAVs.

[0135] Please refer to Figure 4 , Figure 4This is a schematic diagram of the signal detection effect provided in Embodiment 1 of the image transmission signal detection method for multiple UAVs in this application. The left side of the figure shows the original signal waveform, reflecting the amplitude change of the radio frequency signals of multiple UAVs over time. The middle part is the adaptive detection part, which includes signal preprocessing, adaptive threshold calculation, and signal filtering, representing the system's processing flow of noise suppression, dynamic threshold adjustment, and effective signal extraction of the original signal. The right side shows the filtering results, displaying the distribution of signals and interference signals of each UAV after adaptive detection. The frequency range and duration of different signals are marked. Signal 1, signal 2, and signal 3 represent the image transmission signal frequencies of different UAVs, respectively, intuitively demonstrating that the system can distinguish the effective signal frequency bands of multiple UAVs from the aliased radio frequency signals.

[0136] This embodiment provides a method for detecting image transmission signals from multiple UAVs. First, the power spectral density of the multi-UAV radio frequency signals is estimated using the Welch improved periodogram method. The continuous signal is segmented, windowed, and the spectral average is calculated to obtain a power spectral density sequence. This step balances spectral resolution and variance, improving the accuracy of signal frequency characteristic analysis. Second, the power spectral density sequence is smoothed using symmetric moving average and morphological filtering to obtain a preprocessed spectral sequence. Smoothing reduces random noise fluctuations, and erosion and dilation operations suppress spike interference, making the spectrum more regular and providing a reliable foundation for subsequent noise separation. Then, local and global noise estimation is performed based on the noise candidate set in the preprocessed spectral sequence to obtain an initial noise threshold. This step accurately distinguishes signal and noise regions, reducing the risk of false detection and missed detection. Next, the initial noise threshold is dynamically offset and adjusted according to the signal strength quantization factor to obtain an adaptive detection threshold. By matching signal strength changes in real time, the threshold can adapt to different signal-to-noise ratio scenarios, improving the effective signal recognition rate. Subsequently, the energy abrupt boundary of the preprocessed spectral sequence is located using the spectral gradient algorithm, and a signal mask is generated by combining it with an adaptive detection threshold to mark potentially effective signal regions, thus providing clear boundaries for signal screening. Finally, the DBSCAN density clustering algorithm is used to perform connectivity aggregation and frequency band screening on the frequency point set to be clustered marked by the signal mask, eliminating discontinuous or low-quality signals to obtain the image transmission signal detection results. This enables accurate extraction of UAV image transmission signals in complex electromagnetic environments where multiple UAVs coexist.

[0137] Based on the first embodiment of this application, in the second embodiment of this application, the content that is the same as or similar to that in Embodiment 1 above can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 5 , Figure 5This is a flowchart illustrating the second embodiment of the image transmission signal detection method for multiple UAVs according to this application. The energy abrupt change boundary includes strong edge points and weak edge points. Step S50 of the image transmission signal detection method for multiple UAVs includes steps S51 to S56:

[0138] Step S51: Calculate the power decibel difference between adjacent frequency points in the preprocessed spectrum sequence using the spectrum gradient algorithm to obtain the spectrum gradient sequence;

[0139] Step S52: Calculate the difference between the 75th percentile and the 25th percentile in the spectral gradient sequence using the interquartile range algorithm to obtain the dispersion value;

[0140] Step S53: Determine strong edge points and weak edge points based on the discreteness value and the mean gradient of the spectral gradient sequence, and generate an edge mask based on the strong edge points and the weak edge points;

[0141] Step S54: Match the corresponding preset minimum signal-to-noise ratio threshold according to the statistical quantile interval of the local signal-to-noise ratio of each frequency point in the preprocessed spectrum sequence. The local signal-to-noise ratio is the difference between the power decibel value of each frequency point and the local noise estimate obtained by synchronous calculation using a preset sliding window when calculating the local noise estimate.

[0142] Step S55: Remove pseudo signal frequency points in the preprocessed spectrum sequence whose signal-to-noise ratio is lower than the preset minimum signal-to-noise ratio threshold to obtain an effective frequency point set, and take the intersection of the edge mask and the effective frequency point set to obtain a signal mask.

[0143] It should be noted that the power decibel difference refers to the difference in power between adjacent frequency points in the preprocessed spectral sequence, expressed in decibels, and is used to measure the magnitude of energy change between frequency points. The spectral gradient sequence is a sequence obtained by arranging the power decibel differences in frequency order, used to characterize the characteristics of energy abrupt changes in the preprocessed spectral sequence. The interquartile range (IQR) algorithm is a method for calculating the difference between the 75th percentile and the 25th percentile in the sequence, used to measure the dispersion of the sequence. The 75th percentile is the value located at the 75th percentile after sorting the spectral gradient sequence from smallest to largest. The 25th percentile is the value located at the 25th percentile after sorting the spectral gradient sequence from smallest to largest. The dispersion value is the difference between the 75th percentile and the 25th percentile of the spectral gradient sequence, used to represent the dispersion of the gradient distribution. The gradient mean is the arithmetic mean of the spectral gradient sequence, used to reflect the overall gradient level.

[0144] Strong edge points are frequency points with significant changes in the gradient sequence, typically corresponding to prominent signal boundaries. Weak edge points are frequency points with smaller changes in the gradient sequence but may still be considered signal boundaries. An edge mask is a sequence of binary markers representing signal edge locations on the spectrum, where 1 indicates the point belongs to an edge and 0 indicates it does not. Power in decibels (dB) refers to the power value of each frequency point in the preprocessed spectral sequence, expressed in decibels. Statistical quantile intervals are intervals divided based on local signal-to-noise ratios (SNR), such as low, medium, and high intervals, used to match thresholds corresponding to different SNRs. The preset minimum SNR threshold is the lowest SNR threshold set based on the statistical quantile interval, used to eliminate low-quality signals. Pseudo-signal frequency points are frequency points with power or SNR below the minimum SNR threshold, typically representing noise or interference. The effective frequency point set is the set of high SNR frequency points that have been filtered and retained for subsequent signal feature extraction and analysis.

[0145] Understandably, the process begins with the following steps: First, the preprocessed spectral sequence is processed using a spectral gradient algorithm to calculate the power difference in decibels between adjacent frequency points. These differences are then arranged in frequency order to form a spectral gradient sequence, reflecting the amplitude of signal energy variation along the frequency axis (e.g., if one frequency point has a power of 20 dB and the next has 25 dB, the difference is 5 dB). Second, the interquartile range algorithm is used to calculate the difference between the 75th percentile and the 25th percentile of the spectral gradient sequence, yielding a dispersion value to measure the overall fluctuation range of the gradient sequence and provide a reference for edge point identification. Then, based on the dispersion value and the gradient mean of the spectral gradient sequence, a threshold is set to distinguish between strong and weak edge points, generating an edge mask. Identified edge positions are marked as 1, and others as 0, for subsequent signal filtering. Finally, a preset minimum signal-to-noise ratio threshold is matched to the statistical quantile interval of the local signal-to-noise ratio at each frequency point in the preprocessed spectral sequence to determine whether that point is a valid signal. Finally, frequency points with a signal-to-noise ratio lower than the preset minimum signal-to-noise ratio threshold are removed. The intersection of the remaining frequency points and the edge mask is taken to obtain the final signal mask, thereby preserving edge frequency points with high signal-to-noise ratios, which facilitates feature extraction and accurate positioning of subsequent image transmission signals.

[0146] As an example, the step of determining strong edge points and weak edge points based on the dispersion value and the gradient mean of the spectral gradient sequence, and generating an edge mask based on the strong edge points and the weak edge points includes: when the gradient mean of the spectral gradient sequence is greater than the dispersion value, selecting the gradient value corresponding to a preset first quantile in the spectral gradient sequence as a high edge threshold; when the gradient mean is less than or equal to the dispersion value, selecting the gradient value corresponding to a preset second quantile in the spectral gradient sequence as a high edge threshold; multiplying the high edge threshold by a preset ratio to obtain a low edge threshold; determining frequency points in the spectral gradient sequence that are greater than or equal to the high edge threshold as strong edge points, and determining frequency points that are greater than or equal to the low edge threshold and less than the high edge threshold as weak edge points; marking the weak edge points located within a preset neighborhood of the strong edge points as strong edge points, and generating an edge mask based on the strong edge points and the weak edge points.

[0147] It should be noted that the preset first quantile refers to a representative value (e.g., the median or mean within this range) located within the 75th to 80th percentile interval of the spectral gradient sequence, after arranging the gradient values ​​in ascending order. The high edge threshold is a gradient value selected based on the preset first or second quantile, used to distinguish strong edge points from other frequency points and mark obvious energy abrupt changes in the signal. The preset second quantile refers to a representative value (e.g., the median or mean within this range) located within the 60th to 70th percentile interval of the spectral gradient sequence, after arranging the gradient values ​​in ascending order. The preset ratio is the multiplier used when converting the high edge threshold to the low edge threshold, for example, 0.5, used to control the range for judging weak edge points. The low edge threshold is the value obtained by multiplying the high edge threshold by the preset ratio, used to distinguish weak edge points from non-edge points and mark possible signal boundaries. The preset neighborhood range refers to the process of checking whether a certain number of frequency points (e.g., 5 frequency points) around a strong edge point contain weak edge points when determining strong edge points, and promoting these weak edge points to strong edge points to supplement edge continuity.

[0148] Understandably, the system first calculates the mean gradient of the spectral gradient sequence and compares it with the dispersion value. When the mean gradient is greater than the dispersion value, the gradient value located in the preset first quantile interval is taken as the high edge threshold. When the mean gradient is less than or equal to the dispersion value, the gradient value located in the preset second quantile interval is taken as the high edge threshold. Next, the high edge threshold is multiplied by a preset ratio to obtain the low edge threshold, which is used to distinguish between strong and weak edge points. Then, the system marks frequency points in the spectral gradient sequence that are greater than or equal to the high edge threshold as strong edge points and frequency points that are greater than or equal to the low edge threshold but less than the high edge threshold as weak edge points. It also checks weak edge points within a preset neighborhood around each strong edge point, promoting these weak edge points to strong edge points to maintain edge continuity. Finally, an edge mask is generated based on the determined strong and weak edge points, and the edge positions are marked in binary form for subsequent signal filtering.

[0149] This embodiment first calculates the power decibel difference between adjacent frequency points in the preprocessed spectral sequence using a spectral gradient algorithm, obtaining a spectral gradient sequence to reflect the energy changes of the signal along the frequency axis, which helps identify abrupt change regions in the spectrum. Second, it uses a quartile range algorithm to calculate the difference between the 75th percentile and the 25th percentile of the spectral gradient sequence, obtaining a dispersion value to measure the overall fluctuation of the gradient sequence and provide a reference for determining edge points. Then, based on the dispersion value and the gradient mean of the spectral gradient sequence, strong and weak edge points are determined, and an edge mask is generated to mark possible boundary regions of the signal, thereby improving edge localization accuracy. Next, a preset minimum signal-to-noise ratio (SNR) threshold is matched to the statistical quantile interval of the local SNR of each frequency point in the preprocessed spectral sequence to distinguish between effective signals and noise. Finally, pseudo-signal frequency points with SNR below the minimum threshold are removed to obtain a set of effective frequency points, which is then intersected with the edge mask to obtain a signal mask, providing a reliable foundation for the accurate extraction of subsequent image transmission signals.

[0150] For example, to help understand the implementation flow of the image transmission signal detection method for multiple UAVs obtained by combining this embodiment with the above embodiment one, please refer to... Figure 6 , Figure 6 A simplified flowchart of a multi-UAV image transmission signal detection method is provided, specifically:

[0151] The top of the diagram shows the original UAV RF signal input, representing the acquired multi-UAV coexistence signal data. The first stage below is the signal preprocessing stage, which includes Welch power spectrum estimation to calculate the signal power spectral density; decibel conversion to convert the power spectrum to a decibel scale; symmetric moving average to smooth spectral fluctuations; and morphological filtering to remove local spike interference, resulting in a regular spectral sequence. The second stage is the adaptive threshold design stage, where an initial threshold is calculated through local and global noise estimation, combined with signal strength quantization for dynamic offsetting, to obtain the final adaptive noise threshold. The third stage is the signal filtering stage, including gradient edge localization to determine signal boundaries, signal mask generation, and DBSCAN density clustering multi-dimensional threshold filtering to extract effective signal frequency bands. The bottom of the flowchart shows the filtered image transmission signal feature output, providing reliable data for subsequent UAV identification and feature analysis.

[0152] This application also provides an image transmission signal detection device for multiple UAVs, please refer to... Figure 7 The image transmission signal detection device for multiple UAVs includes:

[0153] The power spectral density estimation module 10 is used to estimate the power spectral density of radio frequency signals from multiple UAVs using the Welch improved periodogram method, and obtain a power spectral density sequence.

[0154] Preprocessing module 20 is used to perform symmetric moving average smoothing and morphological filtering on the power spectral density sequence to obtain a preprocessed spectral sequence;

[0155] Noise estimation module 30 is used to perform local and global noise estimation based on the noise candidate set in the preprocessed spectrum sequence to obtain an initial noise threshold;

[0156] Threshold adjustment module 40 is used to dynamically offset and adjust the initial noise threshold according to the signal strength quantization factor to obtain an adaptive detection threshold;

[0157] The mask generation module 50 is used to locate the energy abrupt boundary of the preprocessed spectral sequence using a spectral gradient algorithm, and generate a signal mask based on the energy abrupt boundary and the adaptive detection threshold.

[0158] The frequency band filtering module 60 is used to perform connectivity aggregation and frequency band filtering on the set of frequency points to be clustered marked by the signal mask using the DBSCAN density clustering algorithm to obtain the image transmission signal detection result.

[0159] The image transmission signal detection device for multiple UAVs provided in this application employs the image transmission signal detection method for multiple UAVs described in the above embodiments, and can solve the technical problem of accurately extracting UAV image transmission signals in a complex electromagnetic environment where multiple UAVs coexist. Compared with the prior art, the beneficial effects of the image transmission signal detection device for multiple UAVs provided in this application are the same as those of the image transmission signal detection method for multiple UAVs provided in the above embodiments, and other technical features in the image transmission signal detection device for multiple UAVs are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0160] This application provides a multi-UAV image transmission signal detection device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the multi-UAV image transmission signal detection method in the first embodiment described above.

[0161] The following is for reference. Figure 8 This document illustrates a structural schematic diagram of a multi-UAV image transmission signal detection device suitable for implementing embodiments of this application. The multi-UAV image transmission signal detection device in this application embodiment may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Android Devices), PMPs (Portable Media Players), vehicle terminals (e.g., vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 8 The image transmission signal detection device for multiple UAVs shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0162] like Figure 8As shown, a multi-UAV image transmission signal detection device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in ROM (Read Only Memory) 1002 or a program loaded from storage device 1003 into RAM (Random Access Memory) 1004. RAM 1004 also stores various programs and data required for the operation of the multi-UAV image transmission signal detection device. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via bus 1005. Input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, LCDs (Liquid Crystal Displays), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the multi-UAV image transmission signal detection equipment to wirelessly or wiredly communicate with other devices to exchange data. Although a multi-UAV image transmission signal detection equipment with various systems is shown in the figure, it should be understood that it is not required to implement or possess all the systems shown. More or fewer systems can be implemented alternatively.

[0163] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0164] The image transmission signal detection device for multiple UAVs provided in this application, employing the image transmission signal detection method for multiple UAVs described in the above embodiments, can solve the technical problem of accurately extracting UAV image transmission signals in complex electromagnetic environments where multiple UAVs coexist. Compared with the prior art, the beneficial effects of the image transmission signal detection device for multiple UAVs provided in this application are the same as those of the image transmission signal detection method for multiple UAVs provided in the above embodiments, and other technical features of this image transmission signal detection device for multiple UAVs are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0165] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0166] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, the computer-readable program instructions being used to execute the image transmission signal detection method for multiple UAVs in the above embodiments.

[0167] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, RAM (Random Access Memory), ROM (Read Only Memory), EPROM (Erasable Programmable Read Only Memory or Flash Memory), optical fibers, CD-ROM (CD-Read Only Memory), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0168] The aforementioned computer-readable storage medium carries one or more programs. When these programs are executed by a multi-UAV image transmission signal detection device, the multi-UAV image transmission signal detection device performs the following actions: estimating the power spectral density of the multi-UAV radio frequency signals using the Welch improved periodogram method to obtain a power spectral density sequence; performing symmetric moving average smoothing and morphological filtering on the power spectral density sequence to obtain a preprocessed spectral sequence; estimating local and global noise based on a noise candidate set in the preprocessed spectral sequence to obtain an initial noise threshold; dynamically shifting and adjusting the initial noise threshold based on a signal strength quantization factor to obtain an adaptive detection threshold; locating the energy mutation boundary of the preprocessed spectral sequence using a spectral gradient algorithm, and generating a signal mask based on the energy mutation boundary and the adaptive detection threshold; and performing connectivity aggregation and frequency band filtering on the frequency point set to be clustered marked by the signal mask using the DBSCAN density clustering algorithm to obtain the image transmission signal detection result.

[0169] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including LAN (Local Area Network) or WAN (Wide Area Network)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0170] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0171] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0172] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the above-described image transmission signal detection method for multiple UAVs. This solves the technical problem of accurately extracting UAV image transmission signals in a complex electromagnetic environment where multiple UAVs coexist. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the image transmission signal detection method for multiple UAVs provided in the above embodiments, and will not be repeated here.

[0173] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the image transmission signal detection method for multiple UAVs described above.

[0174] The computer program product provided in this application can solve the technical problem of accurately extracting UAV image transmission signals in a complex electromagnetic environment where multiple UAVs coexist. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as those of the image transmission signal detection method for multiple UAVs provided in the above embodiments, and will not be repeated here.

[0175] The above description is only a part of the embodiments of this application and does not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A method for detecting image transmission signals for multiple unmanned aerial vehicles (UAVs), characterized in that, The method includes: Power spectral density sequence of radio frequency signals from multiple UAVs was obtained by estimating the power spectral density using the Welch improved periodogram method. The power spectral density sequence is subjected to symmetric moving average smoothing and morphological filtering to obtain a preprocessed spectral sequence; Local and global noise estimation is performed based on the noise candidate set in the preprocessed spectrum sequence to obtain the initial noise threshold; The initial noise threshold is dynamically offset and adjusted according to the signal strength quantization factor to obtain an adaptive detection threshold. The energy mutation boundary of the preprocessed spectral sequence is located using the spectral gradient algorithm, and a signal mask is generated based on the energy mutation boundary and the adaptive detection threshold. The DBSCAN density clustering algorithm is used to perform connectivity aggregation and frequency band filtering on the frequency point set to be clustered marked by the signal mask, so as to obtain the image transmission signal detection result.

2. The method as described in claim 1, characterized in that, The step of estimating the power spectral density of multiple UAV radio frequency signals using the Welch improved periodogram method to obtain a power spectral density sequence includes: The radio frequency signals of multiple drones are divided into multiple overlapping signal segments; Each signal segment is weighted using the Hanning window function; Perform a Fourier transform on each weighted signal segment with a preset number of points to obtain a segmented spectrum sequence; Calculate the initial segmented power spectrum corresponding to each signal segment based on the segmented spectrum sequence; Calculate the energy normalization factor of the Hanning window function, and correct the amplitude of each initial segmented power spectrum according to the energy normalization factor to obtain the corrected segmented power spectrum; The average value of all the corrected segmented power spectra is calculated to obtain the power spectral density sequence.

3. The method as described in claim 1, characterized in that, The step of performing local and global noise estimation based on the noise candidate set in the preprocessed spectrum sequence to obtain the initial noise threshold includes: The frequency ranges below a preset power threshold in the preprocessed spectrum sequence are used as noise candidate sets. The local noise estimate is obtained by calculating the local mean and local standard deviation of the noise candidate set through a preset sliding window. When the noise candidate set is not empty, the global noise estimate is calculated by statistically analyzing the energy distribution of the noise candidate set. When the noise candidate set is empty, the global noise estimate is calculated based on the preset low percentile coefficient. The smaller of the local noise estimate and the global noise estimate is used as the initial noise threshold.

4. The method as described in claim 1, characterized in that, The step of dynamically shifting and adjusting the initial noise threshold according to the signal strength quantization factor to obtain the adaptive detection threshold includes: Effective components are selected from the preprocessed spectrum sequence according to the three-standard-deviation criterion, and the peak-to-peak value of the effective components is used as the signal strength quantization factor. The real-time signal-to-noise ratio is determined based on the real-time power distribution of the preprocessed spectral sequence, and the scaling factor is determined based on the real-time signal-to-noise ratio. Multiply the scaling factor by the signal strength quantization factor to obtain the base offset; The local signal-to-noise ratio of the preprocessed spectral sequence is divided into multiple energy ranges by statistical quantiles, and an offset scaling factor is configured for each energy range. The base offset is weighted and corrected according to the offset scaling factor to obtain the dynamic offset; The initial noise threshold is adjusted by superimposing the dynamic offset to obtain a preliminary adjustment threshold. The initial adjustment threshold is constrained by a lower limit based on the minimum offset of the preset threshold to obtain the adaptive detection threshold.

5. The method as described in claim 1, characterized in that, The energy mutation boundary includes strong edge points and weak edge points; The step of locating the energy abrupt change boundary of the preprocessed spectral sequence using a spectral gradient algorithm, and generating a signal mask based on the energy abrupt change boundary and the adaptive detection threshold includes: The power decibel difference between adjacent frequency points in the preprocessed spectrum sequence is calculated using the spectrum gradient algorithm to obtain the spectrum gradient sequence. The difference between the 75th percentile and the 25th percentile in the spectral gradient sequence is calculated using the interquartile range algorithm to obtain the dispersion value; Strong edge points and weak edge points are determined based on the dispersion value and the mean gradient of the spectral gradient sequence, and an edge mask is generated based on the strong edge points and the weak edge points. According to the statistical quantile interval of the local signal-to-noise ratio of each frequency point in the preprocessed spectrum sequence, a corresponding preset minimum signal-to-noise ratio threshold is matched. The local signal-to-noise ratio is the difference between the power decibel value of each frequency point and the local noise estimate obtained by synchronous calculation using a preset sliding window when calculating the local noise estimate. Pseudo-signal frequency points with a signal-to-noise ratio lower than the preset minimum signal-to-noise ratio threshold are removed from the preprocessed spectrum sequence to obtain an effective frequency point set. The intersection of the edge mask and the effective frequency point set is then used to obtain a signal mask.

6. The method as described in claim 5, characterized in that, The step of determining strong edge points and weak edge points based on the discreteness value and the mean gradient of the spectral gradient sequence, and generating an edge mask based on the strong edge points and the weak edge points, includes: When the mean gradient of the spectral gradient sequence is greater than the dispersion value, the gradient value corresponding to the preset first quantile in the spectral gradient sequence is selected as the high edge threshold. When the mean gradient is less than or equal to the dispersion value, the gradient value corresponding to the preset second quantile in the spectral gradient sequence is selected as the high edge threshold; Multiply the high edge threshold by a preset ratio to obtain the low edge threshold; Frequency points in the spectral gradient sequence that are greater than or equal to the high edge threshold are identified as strong edge points, and frequency points that are greater than or equal to the low edge threshold and less than the high edge threshold are identified as weak edge points. The weak edge points located within a preset neighborhood of the strong edge point are marked as the strong edge points, and an edge mask is generated based on the strong edge points and the weak edge points.

7. The method as described in claim 1, characterized in that, The step of performing connectivity aggregation and frequency band filtering on the frequency point set to be clustered marked by the signal mask using the DBSCAN density clustering algorithm to obtain the image transmission signal detection result includes: Using the DBSCAN density clustering algorithm, core frequency points are marked in the set of frequency points to be clustered marked by the signal mask. The core frequency points refer to frequency points whose number of sample frequency points within a preset neighborhood radius is greater than or equal to the preset minimum number of clustering points. Based on the density connectivity between the core frequency points, the corresponding effective frequency points are aggregated into multiple candidate signal clusters; The signal bandwidth, duration, and power margin of each candidate signal cluster are compared with the preset image transmission bandwidth threshold, the preset minimum duration threshold, and the preset minimum power margin threshold, respectively. When any one of the signal bandwidth, the duration, and the power margin is less than a preset value, the corresponding candidate signal cluster is marked as the target signal cluster; The target signal cluster is removed from all the candidate signal clusters to obtain the image transmission signal frequency band, and the image transmission signal frequency band is used as the image transmission signal detection result.

8. A signal detection device for multiple unmanned aerial vehicles (UAVs), characterized in that, The device includes: The power spectral density estimation module is used to estimate the power spectral density of radio frequency signals from multiple UAVs using the Welch improved periodogram method, and obtain a power spectral density sequence. The preprocessing module is used to perform symmetric moving average smoothing and morphological filtering on the power spectral density sequence to obtain a preprocessed spectral sequence. The noise estimation module is used to perform local and global noise estimation based on the noise candidate set in the preprocessed spectrum sequence to obtain an initial noise threshold. The threshold adjustment module is used to dynamically offset and adjust the initial noise threshold according to the signal strength quantization factor to obtain an adaptive detection threshold. The mask generation module is used to locate the energy abrupt boundary of the preprocessed spectral sequence using a spectral gradient algorithm, and generate a signal mask based on the energy abrupt boundary and the adaptive detection threshold. The frequency band filtering module is used to perform connectivity aggregation and frequency band filtering on the set of frequency points to be clustered marked by the signal mask using the DBSCAN density clustering algorithm, so as to obtain the image transmission signal detection result.

9. A signal detection device for multiple unmanned aerial vehicles (UAVs), characterized in that, The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the image transmission signal detection method for multiple unmanned aerial vehicles as claimed in any one of claims 1 to 7.

10. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the image transmission signal detection method for multiple UAVs as described in any one of claims 1 to 7.

Images