Low-altitude unmanned aerial vehicle detection method

Abstract

The application discloses a low-altitude unmanned aerial vehicle detection method and relates to the technical field of low-altitude unmanned aerial vehicle monitoring, which comprises the following steps: acquiring a sequence of optical image frames of a monitored airspace, generating a distance-speed two-dimensional mapping atlas through background difference operation and fast wide-beam scanning operation, estimating an ambient noise power level through incoherent accumulation processing and setting a detection threshold, obtaining target rough estimation information by extracting peak points, controlling a full-array antenna to complete accurate narrow-beam pointing detection according to a rough estimation angle, positioning a corresponding distance-speed coordinate interval in echo data, extracting micro-Doppler spectrum characteristics, filtering random noise to generate a micro-Doppler heat map, inputting the heat map into a pre-trained pattern recognition model to obtain aircraft type determination results, and integrating parameters to generate a target state report. The method can realize rapid rough measurement and accurate directional detection of a target, reduce noise interference, and improve the accuracy of unmanned aerial vehicle detection and type identification.

Description

Technical Field

[0001] This invention belongs to the field of low-altitude unmanned aerial vehicle (UAV) monitoring technology, specifically a method for detecting low-altitude UAVs. Background Technology

[0002] Existing low-altitude UAV detection technologies mostly employ independent working methods of optical image monitoring or radar beam detection. Radar detection often uses a fixed beam full-domain scanning mode, and the target coarse detection stage often uses fixed numerical values ​​to set the detection threshold. Target identification mostly relies on basic radar echo parameters to complete the discrimination. There is no collaborative working mechanism between optical image processing and radar beam scanning, and no hierarchical detection process for wide beams and narrow beams has been formed.

[0003] Independent detection methods are susceptible to external environmental factors. Fixed-beam scanning cannot dynamically adjust its direction based on the target's real-time azimuth, and the fixed detection threshold does not match the actual environmental noise power, directly leading to missed or false detections of target peaks. Traditional micro-Doppler feature extraction does not incorporate target coarse estimation information for directional screening, and random noise components in the echo data cannot be effectively removed. Identification methods based on single echo features cannot accurately determine the aircraft type, and the target's distance, velocity, and angle parameters cannot be integrated with type identification information for coordinated output.

[0004] This invention aims to address the problems of lack of coordination between optical image processing and radar beam scanning, lack of implementation of wide-beam and narrow-beam graded detection, and inability to adaptively set the detection threshold based on the ambient noise power. It also aims to solve the problems of inability to extract micro-Doppler spectral features in a directional manner and the lack of application of micro-Doppler heatmap combined with pattern recognition models to UAV type determination, thereby achieving accurate detection and type identification of low-altitude UAV targets. Summary of the Invention

[0005] This invention aims to solve at least one of the technical problems existing in the prior art; Therefore, this invention proposes a low-altitude unmanned aerial vehicle (UAV) detection method, comprising: Acquire an optical image frame sequence of the monitored airspace, perform background difference operation on the optical image frame sequence, and then perform fast wide-beam phase scanning operation to generate a two-dimensional range-velocity mapping map; Incoherent accumulation processing is performed on the distance-velocity two-dimensional mapping map to estimate the ambient noise power level. A detection threshold is set based on the ambient noise power level, and peak points exceeding the detection threshold are extracted to obtain target rough estimation information including coarse distance, coarse velocity, and coarse angle. Based on the coarse angle in the target coarse estimation information, control the full array antenna to perform a precise narrow beam pointing detection operation, and re-align the detection beam with the target direction; In the echo data of the precise narrow beam velocity pointing detection operation, the distance-velocity coordinate interval corresponding to the target coarse estimation information is located, and the micro-Doppler spectral features within the distance-velocity coordinate interval are extracted; Analyze the micro-Doppler spectral characteristics, filter out random noise components, and generate the target micro-Doppler heatmap; Input the target microDoppler heat map into the pre-trained pattern recognition model and output the aircraft type determination result; By integrating the estimated distance, estimated speed, and estimated angle from the target coarse estimation information, as well as the aircraft type determination result, a final target status report is generated.

[0006] Further, the step of acquiring the optical image frame sequence of the monitored airspace, performing background subtraction on the optical image frame sequence, and then performing a fast wide-beam phase scan operation to generate a two-dimensional range-velocity mapping map includes: The optical image frame sequence of the monitored airspace is captured by an image sensing device, and background subtraction operation is performed on the optical image frame sequence to extract the moving foreground pixel clusters; Based on the spatial distribution density of the moving foreground pixel clusters, the monitoring airspace is divided into multiple sector-shaped detection sectors; The radio frequency transmitting array is driven to perform a fast wide beam phase scan operation. The radio frequency transmitting array is decomposed into multiple transmitting sub-array combinations. Each transmitting sub-array combination controls multiple array antennas. Each transmitting sub-array combination is aligned with a different sector detection area and transmits a frequency-modulated continuous wave signal. The echo reflection signal of the frequency-modulated continuous wave signal is received, and a distance-dimensional fast Fourier transform and a velocity-dimensional fast Fourier transform are performed on the echo reflection signal to generate a two-dimensional distance-velocity mapping spectrum. The step of capturing an optical image frame sequence of the monitored spatial domain using an image sensing device, performing background subtraction on the optical image frame sequence, and extracting moving foreground pixel clusters includes: The image sensing device is controlled to continuously acquire visible light images of the monitored airspace at a fixed frame rate; Construct a background model, which stores static scene images when there are no moving targets; The currently acquired visible light image is subtracted pixel by pixel from the background model to obtain the difference image; The difference image is binarized, and regions with pixel gray values ​​greater than a threshold are marked as motion regions; Connectivity analysis is performed on the motion region to remove tiny noise points with an area smaller than a preset minimum threshold, thus obtaining the motion foreground pixel cluster.

[0007] Furthermore, the driving radio frequency transmitting array performs a fast wide-beam phase scan operation, decomposing the radio frequency transmitting array into multiple transmitting subarray combinations. Each transmitting subarray combination controls multiple array antennas, and each transmitting subarray combination is aligned with a different sector detection area and transmits a frequency-modulated continuous wave signal, including: Determine the physical aperture size and the total number of antenna elements of the radio frequency transmitting array; Calculate the number of sector detection areas to be divided according to the horizontal coverage of the monitored airspace; The antenna elements of the radio frequency transmitting array are equally grouped to form multiple transmitting subarray combinations, each transmitting subarray combination containing the same number of antenna elements; Each transmitter subarray combination is configured with an independent beamforming coefficient, so that the main lobe of the beam generated by each transmitter subarray combination points to a different sector of the detection field. Each transmitter subarray is assigned a non-overlapping transmission time slot or a non-interfering modulation frequency, and the frequency-modulated continuous wave signal is transmitted in a time-division multiplexing or frequency-division multiplexing manner.

[0008] Further, the echo reflection signal of the frequency-modulated continuous wave signal is received, and a range-dimensional fast Fourier transform and a velocity-dimensional fast Fourier transform are performed on the echo reflection signal to generate a two-dimensional range-velocity mapping spectrum, including: The echo reflection signal is de-modulated to obtain an intermediate frequency analog signal; The intermediate frequency analog signal is converted into a digital sampling sequence; Perform a distance-dimensional Fast Fourier Transform along the fast time dimension on the digital sampling sequence to obtain distance-oriented spectral data; Perform a velocity-dimensional fast Fourier transform along the slow time dimension on the distance-oriented spectral data to obtain velocity-oriented spectral data; The velocity spectrum data is rearranged according to distance and velocity units to construct a two-dimensional matrix structure, forming the distance-velocity two-dimensional mapping spectrum.

[0009] Further, incoherent accumulation processing is performed on the distance-velocity two-dimensional mapping map to estimate the ambient noise power level. A detection threshold is set based on the ambient noise power level, and peak points exceeding the detection threshold are extracted to obtain coarse target estimation information including coarse distance, coarse velocity, and coarse angle, including: Select a blank area in the two-dimensional distance-velocity mapping that does not contain the target echo, and statistically analyze the power value distribution of all units within the blank area; Calculate the statistical mean and standard deviation of the power value distribution, and use them as characterization parameters of the environmental background noise power level; Based on the ambient noise floor power level and the preset false alarm probability, a dynamic detection threshold is calculated using a constant false alarm rate detection algorithm; Traverse the entire distance-velocity two-dimensional mapping spectrum and mark the positions of all cells whose power values ​​are higher than the dynamic detection threshold; For each marked unit position, the corresponding coarse estimated distance, coarse estimated velocity, and coarse estimated angle are calculated by combining the beam pointing information of the transmitting subarray combination to which it belongs, and compiled into the target coarse estimation information.

[0010] Furthermore, based on the coarse angle in the target coarse estimation information, the full array antenna is controlled to perform a precise narrow beam pointing detection operation, and the detection beam is re-aligned with the target direction to transmit the detection beam, including: Analyze the target coarse estimation information and read the coarse estimation angle value; Discontinue the fast wide-beam phase scanning operation mode and switch to the full array cooperative working mode; Calculate the phase delay of each antenna element in the full array antenna. The phase delay is used to make the electromagnetic waves radiated by all antenna elements superimpose in phase in the coarsely estimated angular direction. The calculated phase delay is applied to the radio frequency channel of the full array antenna; Transmit a frequency-modulated continuous wave signal with a narrow beamwidth, aligning the main lobe of the narrow beamwidth with the spatial direction indicated by the estimated angle.

[0011] Furthermore, in the echo data of the precise narrow-beam velocity pointing detection operation, the range-velocity coordinate interval corresponding to the target coarse estimation information is located, and the micro-Doppler spectral features within the range-velocity coordinate interval are extracted, including: Receive the baseband signal data returned by the precise narrow beam velocity pointing detection operation; Perform range-dimensional fast Fourier transform and velocity-dimensional fast Fourier transform on the baseband signal data to generate a high-resolution, fine-grained two-dimensional range-velocity mapping spectrum. Obtain the rough distance range and rough speed range from the target rough estimation information; In the fine distance-velocity two-dimensional mapping map, a local rectangular region defined by the coarse distance range and the coarse velocity range is extracted; A short-time Fourier transform is performed on the signal data within the local rectangular region along the time axis to obtain a time-frequency distribution matrix, which is then used as the micro-Doppler spectral feature.

[0012] Further, the micro-Doppler spectral characteristics are analyzed, random noise components are filtered out, and a target micro-Doppler heatmap is generated, including: Calculate the energy intensity at each time-frequency point in the micro-Doppler spectral characteristics; A morphological filtering algorithm is used to remove isolated noise points and sudden interference pulses from the time-frequency distribution matrix; The filtered time-frequency distribution matrix is ​​subjected to energy normalization to eliminate amplitude differences caused by distance attenuation; The normalized time-frequency distribution matrix is ​​converted into a pseudo-color image representation, where the color intensity represents the energy level. The pseudo-color images are arranged according to time and frequency sequences to form a two-dimensional thermal image, namely the target micro-Doppler thermal image.

[0013] Further, the target micro-Doppler heatmap is input into a pre-trained pattern recognition model, and the aircraft type determination result is output, including: The target microDoppler heatmap is size-normalized to conform to the input layer specifications of the pre-trained pattern recognition model; Extract the texture features, edge features, and periodic features of the target microDoppler thermogram; The texture features, edge features, and periodic features are concatenated into a feature vector; The feature vector is input into a pre-trained convolutional neural network classifier; The activation values ​​of the output layer of the convolutional neural network classifier are read, and the aircraft type determination result is determined based on the category label corresponding to the highest probability.

[0014] Furthermore, by integrating the coarse distance, coarse speed, and coarse angle from the target coarse estimation information, as well as the aircraft type determination result, a final target status report is generated, including: Create the target data structure, defining the distance field, velocity field, angle field, and type field; Read the estimated distance value from the target coarse estimation information and fill it into the distance field of the target data structure; Read the rough speed value from the target rough estimate information and fill it into the speed field of the target data structure; Read the rough estimate angle value from the target rough estimate information and fill it into the angle field of the target data structure; The aircraft type determination result is filled into the type field of the target data structure; The target data structure is serialized into a standard data message by appending the current system timestamp and sensor number information, which serves as the final target status report.

[0015] Compared with the prior art, the beneficial effects of the present invention are: After performing background subtraction on the optical image frame sequence of the monitored airspace, a fast wide-beam phase scan is performed to generate a two-dimensional range-velocity mapping spectrum. Incoherent accumulation processing is then performed on this spectrum to estimate the ambient noise power level. By setting a detection threshold based on the ambient noise power level, peak points exceeding the threshold can be accurately extracted, yielding coarse target information including estimated range, velocity, and angle. Optical image background subtraction eliminates static background interference within the monitored airspace. Fast wide-beam phase scan enables rapid full-coverage scanning of the monitored airspace. Incoherent accumulation processing stably obtains the actual power value of the ambient noise. Matching the detection threshold to the actual noise power improves the accuracy of peak point extraction. Controlling the full-array antenna based on the coarse angle allows for precise narrow-beam pointing detection, ensuring the detection beam is accurately aligned with the target direction and reducing the beam's ineffective radiation range.

[0016] By locating the range and velocity coordinate interval corresponding to the coarsely estimated target information in the echo data of precise narrow-beam pointing detection, micro-Doppler spectral features within this interval can be extracted. After filtering out random noise components, a target micro-Doppler heatmap can be generated. Inputting the target micro-Doppler heatmap into a pre-trained pattern recognition model can directly output the aircraft type determination result. Targeted extraction of micro-Doppler spectral features within the corresponding range and velocity coordinate interval avoids redundant data interference caused by global feature extraction. Filtering out random noise components purifies the spectral feature data. The micro-Doppler heatmap can intuitively present the distribution of features. The pattern recognition model's analysis of the heatmap can directly determine the aircraft type. Integrating the coarsely estimated distance, coarsely estimated velocity, coarsely estimated angle, and aircraft type determination results can form complete target state information. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the steps of the low-altitude unmanned aerial vehicle (UAV) detection method described in this invention. Figure 2 A flowchart illustrating how to drive a radio frequency transmitter array to perform a fast wide-beam phase scan operation; Figure 3 An image showing the power distribution of ambient noise floor and the dynamic detection threshold. Figure 4 Micro-Doppler spectral feature extraction and time-frequency distribution map; Figure 5 This is a micro-Doppler thermogram with standardized dimensions. Detailed Implementation

[0018] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] See Figure 1 This invention provides a method for detecting low-altitude unmanned aerial vehicles (UAVs). The method includes: completing target detection and identification through a collaborative processing flow that integrates photoelectric sensing and radio frequency detection. A monitoring system acquires an optical image frame sequence of the monitored airspace, performs background subtraction on the optical image frame sequence to initially sense moving targets, and conducts a fast wide-beam phase scan to generate a two-dimensional range-velocity mapping spectrum covering the airspace. Incoherent accumulation processing is performed on the generated range-velocity mapping spectrum to estimate the ambient noise floor power level. Based on this ambient noise floor power level, a dynamic detection threshold is set, and peak points exceeding the detection threshold are extracted to obtain coarse target information including the estimated target range, estimated velocity, and estimated angle. Based on the coarse angle in the target coarse information, the control system controls the full-array antenna to perform a precise narrow-beam pointing detection operation, re-aligning the detection beam with the target direction to obtain high-quality echoes. In the echo data obtained from the precise narrow-beam pointing detection operation, the range-velocity coordinate interval corresponding to the target coarse information is located, and the micro-Doppler spectral features within this interval are extracted. The extracted micro-Doppler spectral features are analyzed to filter out random noise components, generating a clear target micro-Doppler heatmap. This target micro-Doppler heatmap is then input into a pre-trained pattern recognition model, which outputs a determination of the aircraft type. By integrating the coarsely estimated distance, velocity, and angle from the target estimation information with the aircraft type determination from the pattern recognition model, a final target status report containing the target's position, motion state, and type attributes is generated.

[0020] In one embodiment of the present invention, the monitoring system captures visible light images of the monitored airspace through an image sensing device. The image sensing device continuously acquires images at a fixed frame rate, forming an optical image frame sequence. The system constructs and maintains a background model that stores static scene images when there are no moving targets. The currently acquired visible light image is subtracted pixel by pixel from the background model to obtain a difference image. The difference image is binarized, and regions with pixel grayscale values ​​greater than a preset threshold are marked as moving regions. Subsequently, connected component analysis is performed on the marked moving regions to remove tiny noise points with areas smaller than a preset minimum threshold, resulting in the final moving foreground pixel clusters. Based on the spatial distribution density of the moving foreground pixel clusters, the monitored airspace is divided into multiple sector-shaped detection sectors in the horizontal direction. The system drives the radio frequency transmitting array to perform a fast wide-beam phase scan operation. During operation, the radio frequency transmitting array is decomposed into multiple transmitting subarray combinations. Each transmitting subarray combination controls multiple array antennas, and each transmitting subarray combination is aligned with a different sector-shaped detection sector. Each transmitting subarray combination transmits a frequency-modulated continuous wave signal. The system receives the echo reflection signal of the frequency-modulated continuous wave signal reflected back from the target, performs range-dimensional fast Fourier transform and velocity-dimensional fast Fourier transform on the echo reflection signal, and generates a two-dimensional range-velocity mapping spectrum.

[0021] In practical implementation, the low-altitude UAV detection system is deployed in the perimeter monitoring airspace of an urban park. The image sensing device uses a high-definition visible light camera to continuously acquire visible light images of the monitoring airspace at a fixed frame rate of 30 frames per second, forming an optical image frame sequence. The system constructs a background model, which is generated by continuously acquiring multiple frames of static scene images without moving targets and calculating the average pixel grayscale value. The background model is stored as a matrix with the same resolution as the current image. In practical implementation, the currently acquired visible light image is subtracted pixel by pixel from the background model to obtain a difference image. The grayscale value of each pixel in the difference image is calculated by the formula: in: Represents the difference image in pixel coordinates grayscale value at that location This indicates the pixel coordinates of the currently acquired visible light image. grayscale value at that location Indicates the background model in pixel coordinates In some embodiments, the grayscale value at a given location is binarized by setting a fixed threshold. A value of 20 is set to the threshold for pixels with grayscale values ​​greater than 20. The regions are labeled as motion regions, generating a binary image where the pixel value of the motion region is 1 and the pixel value of the background region is 0. In specific implementation, connected component analysis is performed on the motion regions in the binary image to identify and label each independent connected region. A preset minimum area threshold of 50 pixels is set, and tiny noise points with an area less than 50 pixels are removed. Connected regions with an area greater than or equal to 50 pixels are retained as motion foreground pixel clusters. The motion foreground pixel clusters represent potential moving target regions detected in the image sequence. It can be understood that fan-shaped detection sectors are divided based on the spatial distribution density of the motion foreground pixel clusters. The system calculates the position of all motion foreground pixel clusters in the image. The system calculates the centroid coordinates on a plane and statistically analyzes their horizontal distribution. When the centroids are concentrated on the left side of the image, the system divides the monitored airspace into three sector-shaped detection areas horizontally. The left sector has a wider angular range. In some embodiments, the system drives the RF transmitting array to perform a fast wide-beam phase scan. The RF transmitting array contains 256 antenna elements. The system decomposes the RF transmitting array into four transmitting subarray combinations, each containing 64 antenna elements. Each transmitting subarray combination is aligned with the four divided sector-shaped detection areas. Optionally, each transmitting subarray combination is configured with an independent beamforming coefficient. This ensures that the main lobe of the beam generated by each transmitting subarray combination points towards the center of its corresponding sector detection area. In practice, each transmitting subarray combination transmits frequency-modulated continuous wave signals sequentially in a time-division multiplexing manner. The transmission time slice for each transmitting subarray combination is 1 millisecond. The system receives the echo reflection signal of the frequency-modulated continuous wave signal. The echo reflection signal undergoes low-noise amplification and mixing processing. It can be understood that a range-dimensional fast Fourier transform is performed on the echo reflection signal to convert the time-domain sampling sequence into a range-oriented spectrum. Then, a velocity-dimensional fast Fourier transform is performed along the slow time dimension on the range-oriented spectrum of multiple pulse repetition periods to obtain the velocity-oriented spectrum. The spectrum data is rearranged according to distance and velocity units to construct a two-dimensional matrix. The row index of the matrix corresponds to the distance unit, the column index corresponds to the velocity unit, and the matrix element value represents the signal power, forming a distance-velocity two-dimensional mapping spectrum. In specific implementation, the distance-velocity two-dimensional mapping spectrum is used for subsequent processing. For example, in a park monitoring scenario, when a drone flies into the left sector, the echo signal of the corresponding transmitting subarray combination will show a clear peak point in the distance-velocity two-dimensional mapping spectrum. Optionally, by comparing the distance-velocity two-dimensional mapping spectra generated by different transmitting subarray combinations, the system can preliminarily determine the sector where the target is located.

[0022] In one embodiment of the present invention, when driving the radio frequency transmitting array to perform a fast wide-beam phase scan operation, see [reference needed]. Figure 2The system determines the physical aperture size and total number of antenna elements of the RF transmitting array. Based on the horizontal coverage of the monitored airspace, the required number of sector detection zones is calculated. The antenna elements of the RF transmitting array are equally grouped to form multiple transmitting subarray combinations with the same number of sector detection zones, each containing the same number of antenna elements. Independent beamforming coefficients are configured for each transmitting subarray combination, ensuring that the main lobe of the beam generated by each subarray combination points to its corresponding sector detection zone. Non-overlapping transmission time slots or non-interfering modulation frequencies are assigned to each transmitting subarray combination, and the system controls the transmission of frequency-modulated continuous wave signals by each subarray combination using time-division multiplexing or frequency-division multiplexing. When receiving echo signals, the reflected echo signals are de-modulated to obtain intermediate frequency (IF) analog signals, which are then converted into digital sampling sequences. A range-dimensional Fast Fourier Transform (FFT) is performed on the digital sampling sequences along the fast time dimension to obtain range-oriented spectral data. The obtained range-oriented spectral data is then subjected to a velocity-dimensional FFT along the slow time dimension to obtain velocity-oriented spectral data. The velocity spectrum data is rearranged according to distance and velocity units to construct a two-dimensional matrix structure, forming the final two-dimensional distance-velocity mapping spectrum.

[0023] In a specific implementation, the low-altitude UAV detection system is deployed in the perimeter monitoring airspace of a city park. The radio frequency (RF) transmitting array contains 256 antenna elements, with a physical aperture size of 1.2 meters. The horizontal coverage of the monitoring airspace is 120 degrees. The system calculates the required number of sector-shaped detection zones based on the horizontal coverage of the monitoring airspace, setting the beamwidth of each sector to 30 degrees. Therefore, four sector-shaped detection zones are required. In practice, the system equally groups the 256 antenna elements of the RF transmitting array into four transmitting subarray combinations. Each transmitting subarray combination contains 64 antenna elements. In some embodiments, each transmitting subarray combination is configured with an independent beamforming coefficient, which is calculated using the formula: in: Indicates the first In the nth emitter subarray combination, the nth The complex weighting coefficients of each antenna element, Indicates the wavelength of the transmitted signal. Indicates the first The position coordinates of each antenna element in the array Indicates the first The beam pointing angle of each transmitting subarray combination can be understood as follows: by applying different beamforming coefficients, the main lobe of the beam generated by each transmitting subarray combination points to a different preset sector detection area. For example, the first transmitting subarray combination points to the sector with an azimuth angle of -45 degrees to -15 degrees, and the second transmitting subarray combination points to the sector with an azimuth angle of -15 degrees to 15 degrees. In specific implementation, the system allocates non-overlapping transmission time slots to each transmitting subarray combination and transmits frequency-modulated continuous wave signals sequentially using time-division multiplexing. The transmission time slot length of each transmitting subarray combination is 1 millisecond. Within one complete scan cycle, the four transmitting subarrays... The arrays operate sequentially with a total period of 4 milliseconds. Optionally, in another configuration, the system can assign non-interfering modulation frequencies to each transmitter subarray combination, simultaneously transmitting frequency-modulated continuous wave signals with different center frequencies using frequency division multiplexing. In some embodiments, the receiver receives echo reflection signals from the monitored airspace, performs frequency-modulated processing on the echo reflection signals, and mixes the received signal with a copy of the transmitted signal to obtain an intermediate frequency (IF) analog signal containing target distance information. The frequency of the IF analog signal is proportional to the target distance. In specific implementations, the IF analog signal is converted into a digital sampling sequence using an analog-to-digital converter, with a sampling rate of... At 50 MHz, a range-dimensional Fast Fourier Transform (FFT) is performed along the fast time dimension on the digital sampling sequence, transforming the sampling sequence within each time slice to the range frequency domain, yielding range-oriented spectral data. This range-oriented spectral data is a complex sequence, and the frequency units corresponding to its amplitude peaks can be converted into target distances. Alternatively, a velocity-dimensional FFT is performed along the slow time dimension on the range-oriented spectral data obtained from multiple consecutive time slices. The slow time dimension corresponds to a pulse repetition period sequence. The velocity-dimensional FFT can extract the Doppler frequency shift caused by the radial motion of the target, yielding velocity-oriented spectral data. In practical implementation, the velocity-oriented... The spectral data is rearranged according to distance and velocity units to construct a two-dimensional matrix structure. The row index of the matrix corresponds to the distance unit, and the column index corresponds to the velocity unit. The value of each element in the matrix represents the signal power amplitude at a specific distance and velocity unit. This two-dimensional matrix structure forms the final distance-velocity two-dimensional mapping spectrum. The distance-velocity two-dimensional mapping spectrum intuitively shows whether there is a reflecting target at different distances and velocities. Optionally, for a system with 512 distance units and 128 velocity units, the generated distance-velocity two-dimensional mapping spectrum is a two-dimensional power matrix with 512 rows and 128 columns.

[0024] In one embodiment of the present invention, incoherent accumulation processing is performed on the generated range-velocity two-dimensional mapping spectrum to estimate the ambient noise floor power level. Specifically, a blank area in the range-velocity two-dimensional mapping spectrum that does not contain the target echo is selected, and the power value distribution of all cells within this blank area is statistically analyzed. The statistical mean and standard deviation of this power value distribution are calculated and used as characterization parameters of the ambient noise floor power level. Based on the estimated ambient noise floor power level and the system's preset false alarm probability, a dynamic detection threshold is calculated using a constant false alarm rate detection algorithm. The system traverses the entire range-velocity two-dimensional mapping spectrum and marks the positions of all cells with power values ​​higher than the dynamic detection threshold. For each marked cell position, combined with the beam pointing information of its corresponding transmitting subarray combination, the corresponding coarse estimated range, coarse estimated velocity, and coarse estimated angle are calculated, and this information is compiled into target coarse estimation information. Based on the coarse angle in the target coarse estimation information, the system controls the execution of a precise narrow beam velocity pointing detection operation. During the operation, the system parses the target coarse estimation information and reads the coarse angle value. The system disables the fast wide beam phase scan operation mode and switches to the full array cooperative working mode. The phase delay of each antenna element in the full array antenna is calculated. This phase delay is used to ensure that the electromagnetic waves radiated by all antenna elements are superimposed in phase along the coarsely estimated angular direction. The calculated phase delay is then applied to each RF channel of the full array antenna. Finally, the system transmits a narrow-beamwidth frequency-modulated continuous wave signal, aligning the main lobe of this narrow beam with the spatial direction indicated by the coarsely estimated angle.

[0025] In practical implementation, after generating a two-dimensional range-velocity mapping map, the low-altitude UAV detection system performs incoherent accumulation processing on the map. The range-velocity mapping map is a 512-row, 128-column two-dimensional matrix. A blank region within the map that does not contain target echoes is selected; this blank region is a rectangular area with row indices 300 to 400 and column indices 80 to 100. The power value distribution of all cells within the blank region is statistically analyzed, and the statistical mean of the power value distribution is calculated. and standard deviation This is used as a characterization parameter for the ambient noise floor power level. In some embodiments, it is determined based on the ambient noise floor power level and a preset false alarm probability. The dynamic detection threshold is calculated using a constant false alarm rate (CFAR) detection algorithm. Dynamic detection threshold The result is calculated using the formula: in: Indicates the dynamic detection threshold. The statistical mean of the power value distribution in the blank area. This represents the standard deviation of the power value distribution in the blank region. Indicates the preset false alarm probability The determined scaling factor can be understood as the system traversing the entire distance-velocity two-dimensional mapping spectrum and marking all power values ​​higher than the dynamic detection threshold. The cell locations are shown in Table 1, which displays some of the marked cell locations and their information during a single scan.

[0026] Table 1: Location of Marked Units and Corresponding Coarse Estimation Information In specific implementation, for each marker unit location, combined with the beam pointing information of its corresponding transmitting subarray combination, the corresponding coarse estimated distance, coarse estimated velocity, and coarse estimated angle are calculated. For example, for a marker unit with row index 150 and column index 40, its corresponding distance is 150 multiplied by the distance resolution of 1.5 meters, resulting in a coarse estimated distance of 225.0 meters. The corresponding velocity is (40-64) multiplied by the velocity resolution of 0.125 meters / second, resulting in a coarse estimated velocity of -3.0 meters / second. Its corresponding transmitting subarray combination number is 1, and its beam pointing center angle is -30 degrees. This angle is used as the coarse estimated angle. This information is compiled into target coarse estimation information. The data structure of the target coarse estimation information includes distance, velocity, and angle fields. Optionally, the system stores the coarse estimation information calculated for all marker units in a single scan as a list. In some embodiments, the coarse estimated angle in the target coarse estimation information is used as the basis for the calculation. The system controls the full array antenna to perform precise narrow beam velocity pointing detection. The system analyzes the target coarse estimation information and reads the coarse angle value. For example, if the coarse angle is read as -30 degrees, the system disables the fast wide beam phase scan operation mode and switches to the full array cooperative working mode. The system calculates the phase delay of each antenna element in the full array antenna. The phase delay is used to ensure that the electromagnetic waves radiated by all 256 antenna elements are in phase and superimposed in the direction of the coarse angle -30 degrees. In specific implementation, the calculated phase delay is loaded into each RF channel of the full array antenna, and a narrow beamwidth FM continuous wave signal is transmitted. The beamwidth of the narrow beamwidth FM continuous wave signal is 5 degrees, so that the main lobe of the narrow beamwidth is aligned with the spatial direction indicated by the coarse angle -30 degrees. It can be understood that the precise narrow beam velocity pointing detection operation has higher angular resolution and signal gain compared with the fast wide beam phase scan operation.

[0027] See Figure 3This is a power distribution map of ambient noise floor and a dynamic detection threshold map. The power distribution of the blank area is fitted using a histogram to obtain the mean and standard deviation, serving as a quantitative representation of the ambient noise floor. The threshold, based on false alarm probability calculation, adapts to changes in ambient noise floor, ensuring constant false alarm rate characteristics. The threshold separates the noise floor from the target echo; cells with a value higher than 29.4 are considered suspected targets, providing a basis for subsequent coarse estimation information extraction. The dynamic threshold can offset the effects of ambient clutter and equipment thermal noise, avoiding false alarms / missed detections caused by fixed thresholds. It provides a quantifiable threshold standard for extracting coarse target estimation information, serving as a crucial bridge from the spectrum to target information. The histogram clearly displays the statistical characteristics of the noise floor, facilitating system debugging, parameter optimization, and result verification.

[0028] In one embodiment of the invention, to locate the signal corresponding to the target in the echo data of a precise narrow-beam velocity pointing detection operation, the system first receives the baseband signal data returned by the operation. A range-dimensional Fast Fourier Transform (FFT) and a velocity-dimensional FFT are performed on the baseband signal data to generate a high-resolution, fine-grained two-dimensional range-velocity mapping. A coarse range and a coarse velocity range are obtained from previously obtained target coarse estimation information. A local rectangular region defined by the coarse range and the coarse velocity range is extracted from the fine-grained range-velocity mapping. A short-time Fourier Transform is performed along the time axis on the signal data within this local rectangular region to obtain a time-frequency distribution matrix, which is used as the target's micro-Doppler spectral characteristics. The micro-Doppler spectral characteristics are analyzed to generate a heatmap, including calculating the energy intensity at each time-frequency point in the micro-Doppler spectral characteristics. A morphological filtering algorithm is used to remove isolated noise points and sudden interference pulses from the time-frequency distribution matrix. The filtered time-frequency distribution matrix is ​​then subjected to energy normalization to eliminate echo amplitude differences caused by varying target distances. The normalized time-frequency distribution matrix is ​​converted into a pseudo-color image representation, where color intensity represents energy level. The generated pseudo-color images are arranged according to time and frequency sequences and plotted as a two-dimensional thermal image, which is the target micro-Doppler thermogram.

[0029] In specific implementation, after the low-altitude UAV detection system performs a precise narrow-beam pointing detection operation, the system receives the baseband signal data returned by the precise narrow-beam pointing detection operation. The baseband signal data is a two-dimensional complex matrix containing 1024 range gates and 256 pulse repetition periods. A range-dimensional fast Fourier transform is performed on the baseband signal data, transforming the sampling sequence within each pulse repetition period to the range frequency domain, generating a range spectrum. Then, a velocity-dimensional fast Fourier transform is performed on the range spectrum of the 256 pulse repetition periods along the slow time dimension, generating a high-resolution, fine-grained two-dimensional range-velocity mapping spectrum. This fine-grained range-velocity mapping spectrum is a 1024-row, 256-column matrix, and its range resolution is higher than that obtained by the fast wide-beam phase scan operation. In some embodiments, the coarse range and coarse velocity range are obtained from the target coarse estimation information. For example, according to the example in Embodiment 3, the coarse range is 225.0 meters, and the coarse velocity is -3.0 meters / second. A... With a distance search window range of ±20 meters and a velocity search window range of ±2 meters per second, the coarsely estimated distance range is 205 to 245 meters, and the coarsely estimated velocity range is -5.0 to -1.0 meters per second. In practice, in the fine distance-velocity two-dimensional mapping map, the search window range is converted into the row and column index intervals of the matrix based on the distance and velocity resolution. A local rectangular region defined by the coarsely estimated distance and velocity ranges is extracted. This local rectangular region may correspond to matrix data blocks with row indices 136 to 164 and column indices 120 to 136. The signal data within the local rectangular region is subjected to a short-time Fourier transform along the time axis. The data of 256 pulse repetition cycles is divided into several segments. Each segment is windowed and then subjected to a Fourier transform to obtain the time-frequency distribution matrix. The time-frequency distribution matrix is ​​used as a micro-Doppler spectral feature. It can be understood that the dimension of the time-frequency distribution matrix is ​​the number of time points multiplied by the number of frequency points, which includes the micro-Doppler modulation information caused by the motion of the target rotor and other components.

[0030] In the specific implementation, the micro-Doppler spectral characteristics are analyzed to generate the target micro-Doppler heat map. The process includes calculating the energy intensity of each time-frequency point in the micro-Doppler spectral characteristics. The energy intensity is calculated by the square of the modulus of the corresponding element in the complex matrix. A morphological filtering algorithm is used to remove isolated noise points and sudden interference pulses in the time-frequency distribution matrix. The morphological filtering uses a 3x3 structuring element to perform an opening operation on the two-dimensional matrix representing the energy intensity by first eroding and then dilating. Referring to Table 2, the changes in the energy intensity value of a local region (3x3) in the time-frequency distribution matrix before and after the morphological filtering process are shown.

[0031] Table 2: Comparison of Energy Intensity Filtering Processes at Time and Frequency Points Optionally, the filtered time-frequency distribution matrix is ​​subjected to energy normalization. The maximum-minimum normalization method is used to eliminate amplitude differences caused by distance attenuation. The normalization formula is: in: Indicates the time index after normalization and frequency index The energy value at that location, Indicates the time index after filtering and frequency index The original energy value at that location, This represents the maximum energy value in the filtered matrix. This represents the minimum energy value in the filtered matrix. In some embodiments, the normalized time-frequency distribution matrix is ​​converted into a pseudo-color image representation, and the normalized energy value is represented by a predefined color mapping table. The image is mapped to red, green, and blue color values, with the color intensity representing the energy level. For example, low energy is mapped to dark blue, and high energy is mapped to bright red. The generated pseudo-color image is then arranged according to time and frequency sequences, with the time sequence as the horizontal axis and the frequency sequence as the vertical axis, to create a two-dimensional thermal image. This two-dimensional thermal image is the target micro-Doppler thermal image, which can be used to visually observe the time-frequency structure of micro-Doppler features.

[0032] See Figure 4 This is a micro-Doppler spectral feature extraction and time-frequency distribution map, visually presenting the time-frequency domain energy distribution characteristics of the target echo. The map shows positive and negative dual-band energy bands (approximately -2Hz to +2Hz), consistent with the micro-motion characteristics of a UAV rotor / propeller. When the rotor rotates, the blades move towards / away from the radar, generating positive and negative Doppler frequency shifts, a typical identifying feature of UAV targets. The energy band exhibits a parabolic trajectory of first rising and then falling, with the peak appearing in the approximately 7.5-10 second interval, reflecting the dynamic changes in the target's radial velocity. The overall background is predominantly dark blue / purple, representing a low-energy noise floor, indicating high differentiation between the effective signal and noise, suggesting good signal processing (filtering and denoising). Rotor micro-Doppler features are a core distinguishing feature between UAVs and birds or fixed-wing aircraft; this heatmap provides crucial input features for subsequent pattern recognition models.

[0033] In one embodiment of the present invention, the target micro-Doppler heatmap is input into a pre-trained pattern recognition model for type identification. First, the target micro-Doppler heatmap is standardized to meet the specifications of the input layer of the pre-trained pattern recognition model. Texture features, edge features, and periodic features are extracted from the standardized heatmap. The extracted texture features, edge features, and periodic features are concatenated into a comprehensive feature vector. This feature vector is input into a pre-trained convolutional neural network classifier. The activation values ​​of the output layer of the convolutional neural network classifier are read, and the aircraft type is determined based on the category label corresponding to the highest probability. To integrate various information and generate a final report, a target data structure needs to be created, which defines a distance field, a velocity field, an angle field, and a type field. The coarsely estimated distance value is read from the target coarse estimation information and filled into the distance field of the target data structure. The coarsely estimated velocity value is read from the target coarse estimation information and filled into the velocity field of the target data structure. The coarsely estimated angle value is read from the target coarse estimation information and filled into the angle field of the target data structure. The aircraft type determination result output by the pattern recognition model is filled into the type field of the target data structure. Finally, the current system timestamp and sensor number information are appended to serialize the complete target data structure into a standard data message, which is the final target status report.

[0034] In practical implementation, after generating the target micro-Doppler heatmap, the low-altitude UAV detection system performs size standardization on the heatmap. The input layer specification of the pre-trained pattern recognition model requires an image size of 128 pixels by 128 pixels, while the original size of the generated target micro-Doppler heatmap is 256 pixels by 256 pixels. A bilinear interpolation algorithm is used to scale the target micro-Doppler heatmap to 128 pixels by 128 pixels to conform to the input layer specification of the pre-trained pattern recognition model. In some embodiments, texture features, edge features, and perimeter features of the size-standardized target micro-Doppler heatmap are extracted. Periodic features are obtained by calculating the local binary pattern histogram of the image, texture features are obtained by calculating the gradient magnitude of the image using the Sobel operator, and periodic features are obtained by performing a Fourier transform on each column of the image and analyzing its spectral peaks. It can be understood that the extracted texture feature vector, edge feature vector, and periodic feature vector are concatenated into a comprehensive feature vector. The comprehensive feature vector contains multi-dimensional information of the target micro-Doppler heat map. For example, the dimension of the texture feature vector is 256, the dimension of the edge feature vector is 128, the dimension of the periodic feature vector is 64, and the dimension of the comprehensive feature vector after concatenation is 448.

[0035] In practice, the comprehensive feature vector is input into a pre-trained convolutional neural network (CNN) classifier. The CNN classifier has been trained using a dataset containing micro-Doppler heatmaps of multiple aircraft types. The CNN classifier consists of three convolutional layers, two pooling layers, and two fully connected layers. The activation values ​​of the output layer of the CNN classifier are read. The output layer is a fully connected layer with four nodes, corresponding to four aircraft types: quadcopter drones, fixed-wing drones, helicopters, and birds. The activation values ​​of the output layer are converted into a probability distribution using the Softmax function. The aircraft type is determined based on the category label corresponding to the highest probability. For example, if the activation values ​​of the four nodes of the output layer are [0.05, 0.02, 0.91, 0.02], then the category label corresponding to the highest probability of 0.91 is "helicopter," and the aircraft type is determined to be "helicopter."

[0036] In some embodiments, to integrate various types of information and generate the final target status report, a target data structure needs to be created. This target data structure defines a distance field, a velocity field, an angle field, and a type field. The coarse distance value of 225.0 is read from the target coarse estimation information and filled into the distance field of the target data structure. The coarse velocity value of -3.0 is read from the target coarse estimation information and filled into the velocity field of the target data structure. The coarse angle value of -30.0 is read from the target coarse estimation information and filled into the angle field of the target data structure. This can be understood as filling the aircraft type determination result "helicopter" into the type field of the target data structure. The current system timestamp and sensor number information are appended. The current system timestamp is "2023-10-". 2714:30:05.123", sensor number "RADAR_001", the target data structure is serialized into a standard data message, which is encapsulated in JSON format as the final target status report. An example of the final target status report content is: {"timestamp":"2023-10-2714:30:05.123","sensor_id":"RADAR_001","range_m":225.0,"velocity_mps":-3.0,"angle_deg":-30.0,"type":"helicopter"}. Optionally, the velocity value in the target data structure can be obtained by smoothing the coarsely estimated velocity values ​​from multiple consecutive frames using Kalman filtering. The smoothing formula is: in: This represents the smoothed velocity estimate at time k. This represents the smoothed velocity estimate at time k-1. This represents the Kalman gain at time k. This represents the rough speed value read from the target coarse estimation information at time k, which is then smoothed and filled into the speed field of the target data structure.

[0037] See Figure 5 This is a micro-Doppler thermal image with normalized dimensions, a core feature carrier for UAV target recognition. The image shows a clear arc-shaped main energy band, with a frequency range concentrated between 0 and 3 Hz, and a peak occurring between 15 and 20 seconds, consistent with the periodic micro-motion characteristics of UAV rotors / propellers. The energy intensity has been normalized to the 0-1 range, completely eliminating amplitude differences caused by distance attenuation and transmission power fluctuations, meeting the "energy normalization processing" requirements of the patent. The dark blue background (energy ≈ 0) indicates no significant strong interference, demonstrating excellent preprocessing effects such as morphological filtering and noise removal, effectively preserving target features. The time-frequency clustering, signal-to-noise ratio, and feature clarity can be directly used to evaluate the resolution, anti-interference capability, and target recognition accuracy of the detection system. The shape of the energy trajectory can help determine the target's maneuvering state, providing supplementary information for the final target status report.

[0038] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A method for detecting low-altitude unmanned aerial vehicles (UAVs), characterized in that, include: Acquire an optical image frame sequence of the monitored airspace, perform background difference operation on the optical image frame sequence, and then perform fast wide-beam phase scanning operation to generate a two-dimensional range-velocity mapping map; Incoherent accumulation processing is performed on the distance-velocity two-dimensional mapping map to estimate the ambient noise power level. A detection threshold is set based on the ambient noise power level, and peak points exceeding the detection threshold are extracted to obtain target rough estimation information including coarse distance, coarse velocity, and coarse angle. Based on the coarse angle in the target coarse estimation information, control the full array antenna to perform a precise narrow beam pointing detection operation, and re-align the detection beam with the target direction; In the echo data of the precise narrow beam velocity pointing detection operation, the distance-velocity coordinate interval corresponding to the target coarse estimation information is located, and the micro-Doppler spectral features within the distance-velocity coordinate interval are extracted; Analyze the micro-Doppler spectral characteristics, filter out random noise components, and generate the target micro-Doppler heatmap; Input the target microDoppler heat map into the pre-trained pattern recognition model and output the aircraft type determination result; By integrating the estimated distance, estimated speed, and estimated angle from the target coarse estimation information, as well as the aircraft type determination result, a final target status report is generated.

2. The low-altitude unmanned aerial vehicle (UAV) detection method as described in claim 1, characterized in that, The process of acquiring an optical image frame sequence of the monitored airspace, performing background subtraction on the optical image frame sequence, and then performing a fast wide-beam phase scan to generate a two-dimensional range-velocity mapping map includes: The optical image frame sequence of the monitored airspace is captured by an image sensing device, and background subtraction operation is performed on the optical image frame sequence to extract the moving foreground pixel clusters; Based on the spatial distribution density of the moving foreground pixel clusters, the monitoring airspace is divided into multiple sector-shaped detection sectors; The radio frequency transmitting array is driven to perform a fast wide beam phase scan operation. The radio frequency transmitting array is decomposed into multiple transmitting sub-array combinations. Each transmitting sub-array combination controls multiple array antennas. Each transmitting sub-array combination is aligned with a different sector detection area and transmits a frequency-modulated continuous wave signal. The echo reflection signal of the frequency-modulated continuous wave signal is received, and a distance-dimensional fast Fourier transform and a velocity-dimensional fast Fourier transform are performed on the echo reflection signal to generate a two-dimensional distance-velocity mapping spectrum. The step of capturing an optical image frame sequence of the monitored spatial domain using an image sensing device, performing background subtraction on the optical image frame sequence, and extracting moving foreground pixel clusters includes: The image sensing device is controlled to continuously acquire visible light images of the monitored airspace at a fixed frame rate; Construct a background model, which stores static scene images when there are no moving targets; The currently acquired visible light image is subtracted pixel by pixel from the background model to obtain the difference image; The difference image is binarized, and regions with pixel gray values ​​greater than a threshold are marked as motion regions; Connectivity analysis is performed on the motion region to remove tiny noise points with an area smaller than a preset minimum threshold, thus obtaining the motion foreground pixel cluster.

3. The low-altitude unmanned aerial vehicle (UAV) detection method as described in claim 2, characterized in that, The driving radio frequency transmitting array performs a fast wide-beam phase scan operation, which decomposes the radio frequency transmitting array into multiple transmitting subarray combinations. Each transmitting subarray combination controls multiple array antennas, and each transmitting subarray combination is aligned with a different sector detection area and transmits a frequency-modulated continuous wave signal, including: Determine the physical aperture size and the total number of antenna elements of the radio frequency transmitting array; Calculate the number of sector detection areas to be divided according to the horizontal coverage of the monitored airspace; The antenna elements of the radio frequency transmitting array are equally grouped to form multiple transmitting subarray combinations, each transmitting subarray combination containing the same number of antenna elements; Each transmitter subarray combination is configured with an independent beamforming coefficient, so that the main lobe of the beam generated by each transmitter subarray combination points to a different sector of the detection field. Each transmitter subarray is assigned a non-overlapping transmission time slot or a non-interfering modulation frequency, and the frequency-modulated continuous wave signal is transmitted in a time-division multiplexing or frequency-division multiplexing manner.

4. The low-altitude unmanned aerial vehicle (UAV) detection method as described in claim 3, characterized in that, Receiving the echo reflection signal of the frequency-modulated continuous wave signal, performing a range-dimensional fast Fourier transform and a velocity-dimensional fast Fourier transform on the echo reflection signal to generate a two-dimensional range-velocity mapping spectrum, including: The echo reflection signal is de-modulated to obtain an intermediate frequency analog signal; The intermediate frequency analog signal is converted into a digital sampling sequence; Perform a distance-dimensional Fast Fourier Transform along the fast time dimension on the digital sampling sequence to obtain distance-oriented spectral data; Perform a velocity-dimensional fast Fourier transform along the slow time dimension on the distance-oriented spectral data to obtain velocity-oriented spectral data; The velocity spectrum data is rearranged according to distance and velocity units to construct a two-dimensional matrix structure, forming the distance-velocity two-dimensional mapping spectrum.

5. The low-altitude unmanned aerial vehicle (UAV) detection method as described in claim 4, characterized in that, Incoherent accumulation processing is performed on the two-dimensional distance-velocity mapping to estimate the ambient noise power level. A detection threshold is set based on this ambient noise power level, and peak points exceeding the detection threshold are extracted to obtain coarse target estimation information containing coarse distance, coarse velocity, and coarse angle, including: Select a blank area in the two-dimensional distance-velocity mapping that does not contain the target echo, and statistically analyze the power value distribution of all units within the blank area; Calculate the statistical mean and standard deviation of the power value distribution, and use them as characterization parameters of the environmental background noise power level; Based on the ambient noise floor power level and the preset false alarm probability, a dynamic detection threshold is calculated using a constant false alarm rate detection algorithm; Traverse the entire distance-velocity two-dimensional mapping spectrum and mark the positions of all cells whose power values ​​are higher than the dynamic detection threshold; For each marked unit position, the corresponding coarse estimated distance, coarse estimated velocity, and coarse estimated angle are calculated by combining the beam pointing information of the transmitting subarray combination to which it belongs, and compiled into the target coarse estimation information.

6. The low-altitude unmanned aerial vehicle (UAV) detection method as described in claim 5, characterized in that, Based on the coarse angle in the target coarse estimation information, control the full array antenna to perform a precise narrow beam pointing detection operation, and re-align the detection beam with the target direction, including: Analyze the target coarse estimation information and read the coarse estimation angle value; Discontinue the fast wide-beam phase scanning operation mode and switch to the full array cooperative working mode; Calculate the phase delay of each antenna element in the full array antenna. The phase delay is used to make the electromagnetic waves radiated by all antenna elements superimpose in phase in the coarsely estimated angular direction. The calculated phase delay is applied to the radio frequency channel of the full array antenna; Transmit a frequency-modulated continuous wave signal with a narrow beamwidth, aligning the main lobe of the narrow beamwidth with the spatial direction indicated by the estimated angle.

7. The low-altitude unmanned aerial vehicle (UAV) detection method as described in claim 6, characterized in that, In the echo data of the precise narrow beam velocity pointing detection operation, the range-velocity coordinate interval corresponding to the target coarse estimation information is located, and the micro-Doppler spectral features within the range-velocity coordinate interval are extracted, including: Receive the baseband signal data returned by the precise narrow beam velocity pointing detection operation; Perform range-dimensional fast Fourier transform and velocity-dimensional fast Fourier transform on the baseband signal data to generate a high-resolution, fine-grained two-dimensional range-velocity mapping spectrum. Obtain the rough distance range and rough speed range from the target rough estimation information; In the fine distance-velocity two-dimensional mapping map, a local rectangular region defined by the coarse distance range and the coarse velocity range is extracted; A short-time Fourier transform is performed on the signal data within the local rectangular region along the time axis to obtain a time-frequency distribution matrix, which is then used as the micro-Doppler spectral feature.

8. The low-altitude unmanned aerial vehicle (UAV) detection method as described in claim 7, characterized in that, Analyzing the micro-Doppler spectral characteristics, filtering out random noise components, and generating a target micro-Doppler heatmap includes: Calculate the energy intensity at each time-frequency point in the micro-Doppler spectral characteristics; A morphological filtering algorithm is used to remove isolated noise points and sudden interference pulses from the time-frequency distribution matrix; The filtered time-frequency distribution matrix is ​​subjected to energy normalization to eliminate amplitude differences caused by distance attenuation; The normalized time-frequency distribution matrix is ​​converted into a pseudo-color image representation, where the color intensity represents the energy level. The pseudo-color images are arranged according to time and frequency sequences to form a two-dimensional thermal image, namely the target micro-Doppler thermal image.

9. The low-altitude unmanned aerial vehicle (UAV) detection method as described in claim 8, characterized in that, Input the target micro-Doppler heatmap into a pre-trained pattern recognition model, and output the aircraft type determination result, including: The target microDoppler heatmap is size-normalized to conform to the input layer specifications of the pre-trained pattern recognition model; Extract the texture features, edge features, and periodic features of the target microDoppler thermogram; The texture features, edge features, and periodic features are concatenated into a feature vector; The feature vector is input into a pre-trained convolutional neural network classifier; The activation values ​​of the output layer of the convolutional neural network classifier are read, and the aircraft type determination result is determined based on the category label corresponding to the highest probability.

10. The low-altitude unmanned aerial vehicle (UAV) detection method as described in claim 9, characterized in that, By integrating the coarse distance, coarse speed, and coarse angle from the target coarse estimation information, along with the aircraft type determination result, a final target status report is generated, including: Create the target data structure, defining the distance field, velocity field, angle field, and type field; Read the estimated distance value from the target coarse estimation information and fill it into the distance field of the target data structure; Read the rough speed value from the target rough estimate information and fill it into the speed field of the target data structure; Read the rough estimate angle value from the target rough estimate information and fill it into the angle field of the target data structure; The aircraft type determination result is filled into the type field of the target data structure; The target data structure is serialized into a standard data message by appending the current system timestamp and sensor number information, which serves as the final target status report.

Images