Altimeter signal detection method and system based on dynamic parameter OS-CFAR

Abstract

The invention discloses an altimeter signal detection method and system based on a dynamic parameter OS-CFAR, and the method comprises the steps: firstly collecting a radar signal, carrying out the preprocessing, obtaining frequency spectrum data, and calculating the frequency spectrum entropy and the signal-clutter-noise ratio; then, forming a two-dimensional feature vector as the input of a Gaussian mixture model, performing an offline training stage and an online execution stage, and generating an environment adaptive OS-CFAR parameter combination; calculating a detection threshold factor based on the parameter combination, and identifying a target signal; and finally, performing multi-dimensional target verification on the target signal to obtain an altimeter signal. The method makes up the defect that a traditional single power feature is difficult to distinguish a strong target from distributed interference by constructing a mixed feature of the spectrum entropy and the signal-clutter-noise ratio, introduces an environment feature perception and parameter dynamic mapping mechanism, realizes adaptive adjustment of detection parameters along with the environment, and meanwhile, adopts an offline training and online reasoning mode to improve the detection accuracy. And the calculation efficiency and the stability are considered while the environment self-adaptive capability is guaranteed.

Description

Technical Field

[0001] This invention relates to signal detection methods, belonging to the field of target detection algorithms, and particularly to a method and system for altimeter signal detection based on dynamic parameter OS-CFAR. Background Technology

[0002] In actual testing, complex and ever-changing environmental conditions often cause significant fluctuations in the dynamic range of signals. When the signal strength is too high, it is prone to saturation distortion, while when the strength is too low, effective detection becomes difficult. Therefore, effective signal recognition relies on the flexible configuration of adaptive detection thresholds. However, traditional OS-CFAR detectors use fixed parameter designs and cannot dynamically adjust the detection strategy according to environmental characteristics, resulting in a severe lack of adaptability in complex and dynamic environments.

[0003] In addition to the limitations of the detector itself, the signal recognition process also faces multiple interference challenges. For example, after FFT spectrum analysis, false spectral peaks formed by random noise, multipath reflection signals in low-altitude flight scenarios, and interference from multiple scattering centers caused by complex terrain will all seriously affect the accuracy of altitude measurement and further increase the difficulty of effective signal recognition.

[0004] A closer look at the shortcomings of traditional detection schemes reveals that fixed-threshold detection has particularly prominent drawbacks: First, fixed thresholds are difficult to adapt to the changing clutter characteristics of complex terrains such as cities, mountains, and sea surfaces, resulting in poor environmental adaptability. Second, false detection events are prone to occur in low-altitude multipath propagation environments, making it difficult to effectively control the false alarm rate. Third, the detection sensitivity remains constant, unable to adaptively switch between clean and complex environments. To address these shortcomings, deep learning-based CFAR technology has been introduced and applied. While this method uses neural networks to classify environments and exhibits superior detection performance in complex environments, it faces significant bottlenecks in practical engineering applications: On the one hand, model training relies on large-scale labeled datasets and has high computational complexity, making it difficult to meet real-time processing requirements; on the other hand, the model maintenance and update process is complex, and its black-box nature cannot meet the deterministic requirements of high-reliability systems, severely restricting the engineering implementation of this technology on embedded platforms. Therefore, there is an urgent need for a height gauge signal detection method that can adapt to environmental changes, has low computational complexity, and stable detection performance to overcome the aforementioned shortcomings of existing technologies. Summary of the Invention

[0005] The purpose of this invention is to overcome the above-mentioned defects and problems in the prior art and to provide a method and system for altimeter signal detection based on dynamic parameter OS-CFAR that can adapt to environmental changes and balance efficiency and stability.

[0006] To achieve the above objectives, the technical solution of the present invention is: a method for detecting altimeter signals based on dynamic parameter OS-CFAR, comprising:

[0007] The radar signal is collected and preprocessed to obtain spectrum data, and the spectrum entropy and signal-to-clutter ratio are calculated based on the spectrum data.

[0008] The spectral entropy and signal-to-noise ratio are used as a two-dimensional feature vector as input to the Gaussian mixture model for offline training and online execution, generating an environment-adaptive OS-CFAR parameter combination.

[0009] Based on the environment-adaptive OS-CFAR parameter combination, the detection threshold factor is calculated, and the target signal is identified by point-by-point detection of the spectrum data based on the detection threshold factor.

[0010] After performing multi-dimensional target verification on the target signal, the final detection result is output, and the altimeter signal is obtained.

[0011] Optionally, the calculation of spectral entropy and signal-to-noise ratio specifically includes:

[0012] The spectral entropy is calculated using the following expression:

[0013] ;

[0014] ;

[0015] in: For spectral entropy, Normalized power spectral density; For the first Power spectral density at each frequency point; This is the index of the frequency point currently being calculated; This refers to the traversal index in the summation operation; This represents the total number of frequency points in the spectrum data.

[0016] Calculating the signal-to-noise ratio (SNR) specifically includes:

[0017] The power spectrum is smoothed using an adaptive sliding window process based on the spectral data to obtain the smoothed power spectrum value, which is expressed as follows:

[0018] ;

[0019] in: For the first Power spectrum values ​​after smoothing at each frequency point; This represents the total number of sample points in the sliding window. For the number of window points; For the summation sequence number; For the first The original power spectrum values ​​at each frequency point; The current frequency point number being processed; Sampling rate; To achieve smooth physical bandwidth; The number of FFT points;

[0020] The spectral segments with a continuous spectral width greater than twice the window length in the smoothed power spectrum are divided into signal bands, and the remaining spectral segments are divided into clutter noise bands.

[0021] The signal-to-noise ratio (SNR) is obtained by performing power integral calculations on the signal band and clutter noise band separately and then calculating the ratio.

[0022] Optionally, the offline training phase and the online execution phase specifically include:

[0023] During the offline training phase, environment clustering is performed based on two-dimensional feature vectors to determine representative environment states. The representative environment states are then mapped to the corresponding ordered statistical constant false alarm rate (OS-CFAR) parameter combinations. An environment-parameter mapping strategy library is constructed and embedded in the Gaussian mixture model.

[0024] During the online execution phase, the feature vector of the current environment state is extracted and input into the solidified Gaussian mixture model. The parameters of the environment-parameter mapping strategy library are called to perform responsibility weighting and coupling constraint verification, and generate an environment-adaptive OS-CFAR parameter combination.

[0025] Optionally, the environmental clustering specifically includes:

[0026] Two-dimensional feature vectors are input into a Gaussian mixture model, and the EM algorithm is used for environmental clustering to determine the cluster center point representing each environmental state; the representative environmental states include pure, moderate, complex, and harsh.

[0027] Using the cluster centers corresponding to each environmental state as the object, the cluster centers with the highest signal-to-noise ratio and the lowest spectral entropy value are classified as pure environmental states, and the cluster centers with the lowest signal-to-noise ratio and the highest spectral entropy value are classified as harsh environmental states.

[0028] Between cluster centers in clean and harsh environments, cluster centers with relatively high signal-to-noise ratios are classified as medium environments, and cluster centers with relatively low signal-to-noise ratios are classified as complex environments, based on their centrality.

[0029] Optionally, the online execution phase specifically includes:

[0030] The system collects and extracts feature vectors of the current environmental state in real time, and inputs them into the solidified Gaussian mixture model to output a responsibility vector. ;in, For each component The current environmental state represented belongs to the first Each represents a probability weight of the environmental state;

[0031] Recall the corresponding OS-CFAR parameter combination representing the environment state from the environment-parameter mapping policy library. Furthermore, through responsibility-weighted and coupling constraint verification, an environment-adaptive OS-CFAR parameter combination is generated, the expression of which is as follows:

[0032] ; ; ;

[0033] in: This represents the number of training units; The number of protection units; For sorting indexes.

[0034] Optionally, the coupling constraint specifically includes:

[0035] The training unit The value range is from 16 to 64, and the adjustment step is an integer multiple of 8;

[0036] The protection unit The value range is 1 to 6, the adjustment step is 1, and the sum of the protection window and the training window does not exceed the sliding window processing capacity.

[0037] The sorting index The value range is 0.3. Up to 0.8 Adjust the step size to 1, and when When it increases from 32 to 48, The upper limit was automatically adjusted from 25 to 38.

[0038] Optionally, the calculation of the detection threshold factor based on the environment-adaptive OS-CFAR parameter combination specifically includes:

[0039] Based on a preset false alarm probability Training Unit With sorted index The detection threshold factor is solved by inversely calculating the exact model in the number field, and its expression is as follows:

[0040] ;

[0041] in: To detect threshold factors.

[0042] Optionally, the step of performing point-by-point detection of the spectral data based on a detection threshold factor to identify the target signal specifically includes:

[0043] Centered on the current detection unit, set up on both sides of the spectrum data Each protection unit is equipped with a protective element, and the protective element is installed on the outside of the protective element. Each reference unit is selected on one side. One reference unit;

[0044] The power values ​​around the reference cell are sorted in ascending order based on the detection threshold factor, and the first value is selected. The minimum power value is used as an estimate of the current background noise power;

[0045] Based on detection threshold factor Compared with the estimated current background noise power Calculate the detection threshold and the power value of each detection unit With detection threshold Comparison:

[0046] like If the signal is positive, it is identified as the target signal, and the frequency and amplitude at that point are recorded.

[0047] like If the value is not recorded, it is considered noise and will not be recorded.

[0048] Optionally, after performing multi-dimensional target verification on the target signal, the final detection result is output to obtain the altimeter signal, specifically including:

[0049] Perform the following target verification on the target signal and eliminate target signals that do not meet the verification conditions:

[0050] Amplitude verification: Remove target signals whose amplitude is less than 10% of the maximum amplitude;

[0051] Frequency range verification: Remove target signals whose frequency is not within the frequency threshold;

[0052] Local peak verification: Remove target signals with insignificant local peaks;

[0053] Historical continuity verification: Remove target signals that deviate from historical detection results by more than 30%;

[0054] The target signal that meets the above verification conditions is taken as the final detection result, and the corresponding frequency and amplitude information is output. The distance is then converted according to the frequency difference to obtain the platform's height above the ground.

[0055] A height gauge signal detection system based on dynamic parameter OS-CFAR, which is applied to the above-mentioned method, the system comprising:

[0056] The input vector acquisition module is used to collect radar signals and preprocess them to obtain spectrum data, and calculate the spectrum entropy and signal-to-clutter ratio based on the spectrum data.

[0057] The adaptive parameter generation module is used to construct a two-dimensional feature vector from the spectral entropy and signal-to-noise ratio as input to the Gaussian mixture model for offline training and online execution, generating an environment-adaptive OS-CFAR parameter combination.

[0058] The target signal recognition module is used to calculate the detection threshold factor based on the environment-adaptive OS-CFAR parameter combination, and to perform point-by-point detection on the spectrum data based on the detection threshold factor to identify the target signal;

[0059] The altimeter signal acquisition module is used to perform multi-dimensional target verification on the target signal and output the final detection result to obtain the altimeter signal.

[0060] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0061] This invention relates to a method and system for altimeter signal detection based on dynamic parameter OS-CFAR. The method first acquires and preprocesses radar signals to obtain spectral data and calculates spectral entropy and signal-to-clutter ratio (SNR). Then, the spectral entropy and SNR are used to construct a two-dimensional feature vector as input to a Gaussian mixture model (Gaussian mixture model), which undergoes offline training and online execution phases to generate an environment-adaptive OS-CFAR parameter combination. Next, based on the parameter combination, a detection threshold factor is calculated, and the target signal is identified. Finally, the target signal undergoes multi-dimensional target verification to obtain the altimeter signal. In application, this design effectively overcomes the shortcomings of traditional single power features in distinguishing between strong targets and distributed interference by constructing a hybrid feature of spectral entropy and SNR. Furthermore, it introduces an environmental feature perception and dynamic parameter mapping mechanism to achieve adaptive adjustment of detection parameters according to environmental complexity. Simultaneously, it adopts a model learning pre-paradigm, completing the complex training process offline, while only performing lightweight deterministic inference in the online phase. Ultimately, it balances computational efficiency and stability while ensuring environmental adaptability. Attached Figure Description

[0062] Figure 1 This is a flowchart of the method of the present invention.

[0063] Figure 2 This is a flowchart of the altimeter signal processing in Embodiment 1 of the present invention.

[0064] Figure 3 This is a schematic diagram of the OS-CFAR detection principle in Embodiment 1 of the present invention.

[0065] Figure 4 This is a system structure diagram of the present invention.

[0066] Figure 5 This is a structural diagram of the device of the present invention.

[0067] In the diagram: Input vector acquisition module 1, adaptive parameter generation module 2, target signal recognition module 3, altimeter signal acquisition module 4, processor 5, memory 6, computer program code 61. Detailed Implementation

[0068] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0069] Example 1:

[0070] See Figure 1 A method for detecting altimeter signals based on dynamic parameter OS-CFAR, applicable to LFMCW system altimeters, is executed after completing an 8192-point complex FFT, specifically including:

[0071] The radar signal is collected and preprocessed to obtain spectrum data, and the spectrum entropy and signal-to-clutter ratio are calculated based on the spectrum data.

[0072] In this embodiment, the intermediate frequency echo signal of the radar is acquired, analog-to-digital conversion (ADC) is performed, and the digital signal is preprocessed such as DC removal and windowing to reduce spectral leakage; the preprocessed signal is subjected to fast Fourier transform (FFT) to obtain spectral data, and the two are normalized.

[0073] Furthermore, spectral entropy, used to characterize the degree of disorder in the spectral distribution, is calculated as follows:

[0074] ;

[0075] ;

[0076] in: For spectral entropy, To normalize the power spectral density, satisfying ; For the first Power spectral density at each frequency point; This is the index of the frequency point currently being calculated; This refers to the traversal index in the summation operation; This represents the total number of frequency points in the spectrum data.

[0077] The signal-to-noise ratio (SNR) is calculated using an adaptive estimation algorithm based on the power spectral density distribution characteristics, specifically including:

[0078] To stabilize the background estimation, an adaptive sliding window smoothing process is applied to the power spectrum. The width of the smoothing window adapts to the sampling rate and the number of FFT points to maintain a constant physical smoothing bandwidth. The expression for this process is as follows:

[0079] ;

[0080] in: For the first Power spectrum values ​​after smoothing at each frequency point; This represents the total number of sample points for the symmetrical sliding window; The number of points in the window is rounded down. For the summation sequence number; For the first The original power spectrum values ​​at each frequency point; The current frequency point number being processed; Sampling rate; The desired smoothing physical bandwidth (in Hz), whether preset or adaptive, is the core parameter controlling the degree of smoothing. The number of FFT points;

[0081] This design allows for changes in sampling rate. When changing, the number of window points Automatic adjustment to ensure that smoothing operations always apply to approximately the same desired smoothing physical bandwidth. This ensures the stability of the algorithm's performance.

[0082] The spectral segments with a continuous spectral width greater than twice the window length in the smoothed power spectrum are divided into signal bands, and the remaining spectral segments are divided into clutter noise bands.

[0083] The signal-to-noise ratio (SNR) is obtained by performing power integral calculations on the signal band and clutter noise band separately and then calculating the ratio.

[0084] The normalized spectral entropy and signal-to-noise ratio are used to form a two-dimensional feature vector, which is then used as the input to the Gaussian mixture model for both offline training and online execution.

[0085] In this embodiment, to eliminate the inherent differences in physical dimensions and numerical ranges between spectral entropy and signal-to-noise ratio (SNR), and to make the feature distribution more consistent with the probabilistic assumptions of the Gaussian mixture model, a statistical Z-Score normalization method is used to normalize the data. This method solidifies global statistical parameters offline and applies the same standard online, ensuring consistency in the feature space, thereby improving the system's robustness to outliers and overall stability. Specifically, it includes:

[0086] During the offline training phase, parameter statistics, data processing, and model solidification are performed.

[0087] Parameter statistics: Collect the set of spectral entropy values ​​for all historical training samples. Set of signal-to-noise ratio values And calculate their global mean respectively. Compared with global standard deviation .

[0088] Data processing: Using the statistical parameters described above, all training samples are standardized to generate a standardized feature set for model training. For the _____, ... The conversion formula for each training sample is as follows:

[0089] ;

[0090] Model training and solidification: using standardized feature sets Train a Gaussian mixture model and build an environment-parameter mapping policy library. Finally, statistical parameters... It is stored in the system memory along with the trained Gaussian mixture model and policy library.

[0091] Online execution phase: parameter calling and real-time standardization are performed; during real-time signal processing, the standards established in the offline phase are followed;

[0092] Parameter retrieval: Read parameters directly from the hard drive. .

[0093] Real-time calculation: The raw feature values ​​obtained by real-time calculation of the current spectrum. and Real-time standardization is performed using the same formula as in the offline phase:

[0094] ;

[0095] The feature values ​​obtained after the above standardization process , Constructed as a two-dimensional feature vector As input to the Gaussian mixture model, it is used for the subsequent offline training and online execution phases of the model itself.

[0096] In this embodiment, an environment perception and parameter adaptive mapping mechanism based on a Gaussian mixture model (GMM) is adopted. Through a separate architecture for offline training and online execution, intelligent optimization of OS-CFAR detection parameters is achieved. During the offline training phase, unsupervised clustering of massive historical environmental feature data identifies four most representative environmental states (clean, moderate, complex, and harsh). The clustering process automatically discovers the inherent structure of the environmental feature space through probability density modeling. Each environmental state corresponds to a two-dimensional Gaussian distribution, and the statistical characteristics and correlations of spectral entropy and signal-to-noise ratio under each state are analyzed. During the online execution phase, the current environmental feature vector is extracted in real time and input into the pre-defined GMM model. A continuous responsibility vector is calculated and output, where each component precisely represents the probability weight of the current environment belonging to each environmental state. Based on this responsibility vector, a weighted linear synthesis and coupling constraints are used to generate an adaptive OS-CFAR parameter combination in real time, as detailed below:

[0097] Offline training phase: Environment clustering is performed based on two-dimensional feature vectors to determine representative environment states, and the representative environment states are mapped to the corresponding ordered statistical constant false alarm rate (OS-CFAR) parameter combinations. An environment-parameter mapping strategy library is constructed and embedded in the Gaussian mixture model.

[0098] Furthermore, during the offline training phase, the Gaussian mixture model performs environment clustering based on the EM algorithm, which is an iterative optimization algorithm aimed at maximizing the log-likelihood function of the observed data. Its steps are as follows:

[0099] Initialize and set the parameters of the Gaussian mixture model, including the weights of each Gaussian component, the mean vector, and the covariance matrix;

[0100] Perform the E-step, for a given set of parameters, to calculate the responsibility weight for each sample belonging to each Gaussian component. , which is the probability weight of a new sample belonging to the Kth environmental cluster ([0,1], the larger the value, the more it resembles that cluster), is expressed as follows:

[0101] ;

[0102] in: For the sample Obey the first Two-dimensional Gaussian probability density of Gaussian components ; , , The first The weights, mean vectors, and covariance matrices of each Gaussian component; , , The first The weights, mean vectors, and covariance matrices of each Gaussian component;

[0103] Execute M steps, based on responsibility weights The weights, mean vector, and covariance matrix of each Gaussian component are updated as follows:

[0104] ; ; ;

[0105] ;

[0106] Repeat the E-step and M-step until the parameters converge, and obtain Gaussian components of 4 categories. The mean vector of the Gaussian component of each category represents the typical characteristics of the environmental state represented by that category (i.e., pure, moderate, complex, harsh).

[0107] Based on the physical characteristics of each cluster center (i.e., the mean vector), the mapping relationship between them and four typical environmental states is automatically established, and an environment-parameter mapping strategy library is constructed. The specific steps are as follows:

[0108] Cluster centroid analysis: Calculate the mean vector for each Gaussian component. This characterizes the typical features of this type of environmental state, namely the average level of spectral entropy and signal-to-noise ratio under this environment.

[0109] Automatic state matching: Clusters with both the highest signal-to-noise ratio (SNR) and the lowest spectral entropy are matched as "clean" environmental states; clusters with both the lowest SNR and the highest spectral entropy are matched as "harsh" environmental states; for cluster centers between "clean" and "harsh" environmental states, the center point between the clean and harsh environmental states is first determined, and based on the degree of centrality, clusters with a relatively high SNR compared to the center point (i.e., closer to the clean environmental state with the center point as the reference point) are matched as "medium" environmental states; clusters with a relatively low SNR compared to the center point (i.e., closer to the harsh environmental state with the center point as the reference point) are matched as "complex" environmental states.

[0110] In the online execution phase: the feature vector of the current environment state is extracted and input into the solidified Gaussian mixture model. The parameters of the environment-parameter mapping strategy library are called to generate an environment-adaptive OS-CFAR parameter combination. When a new feature vector is input, the GMM calculates the probability that it belongs to each component and performs weighted fusion based on the environment label and parameter combination probability corresponding to the component, relying on the parameter combination in the parameter strategy library, thereby achieving "environment adaptation".

[0111] Furthermore, the feature vector of the current environmental state is collected and extracted in real time, and input into the solidified Gaussian mixture model to output the responsibility vector. ;in, For each component The current environmental state represented belongs to the first Each represents a probability weight of the environmental state;

[0112] Recall the corresponding OS-CFAR parameter combination representing the environment state from the environment-parameter mapping policy library. Furthermore, through responsibility-weighted and coupling constraint verification, an environment-adaptive OS-CFAR parameter combination is generated, the expression of which is as follows:

[0113] ; ; ;

[0114] in: This represents the number of training units; The number of protection units; For sorting indexes.

[0115] The environment-adaptive OS-CFAR parameter combination in this solution includes: , and , The size of the reference cell determines the reliability of the background statistics. The reference cell refers to a cell at a specific location in the spectrum data. It is an index or positional concept. What needs to be sorted are the signal power values ​​corresponding to these cells. Protect the target energy from leakage and ensure the purity of the detection unit; The sorting position of the ordered statistics is controlled to balance sensitivity and robustness; the three together constitute a complete detection statistic, and constant false alarm rate control is achieved through the detection threshold factor.

[0116] In this embodiment, the mapping mechanism representing environmental features and OS-CFAR parameters is established based on a quantitative model of environmental features and detection performance, wherein the adjustment of each parameter follows a clear driving logic:

[0117] Training Unit The signal-to-noise ratio (SCNR) is adaptively adjusted, increasing the number of reference units in low SCNR environments to improve noise estimation accuracy and reducing the number of reference units in high SCNR environments to improve processing efficiency.

[0118] Protection Unit Based on the dynamic configuration of the spectrum entropy value, the protection unit is increased to prevent target energy leakage in a high spectrum entropy environment, and the protection unit is reduced to improve range resolution in a low spectrum entropy environment;

[0119] Sort index Based on the adjustment of the overall environmental complexity, a larger K value is used to suppress clutter interference in complex environments with high spectral entropy or low SCNR, while a smaller K value is selected to improve detection sensitivity in clean environments with low spectral entropy and high SCNR.

[0120] The above parameters are jointly optimized through system-level coupling constraints, the specific constraint system of which is as follows:

[0121] Training Unit The value range is from 16 to 64, and the adjustment step is an integer multiple of 8. It is subject to basic boundary constraints to ensure that it does not exceed the system design safety range.

[0122] Protection Unit The value range is 1 to 6, and the adjustment step is 1. The system capacity constraint ensures that the sum of the protection window and the training window does not exceed the sliding window processing capacity L_window to prevent data index overflow.

[0123] Sort index The value range is 0.3. Up to 0.8 Adjust the step size to 1, subject to parameter linkage constraints to ensure a coordinated ratio with the training unit size. When it increases from 32 to 48, The upper limit is automatically adjusted from 25 to 38 to avoid parameter mismatch.

[0124] This constraint system, through the synergistic constraints between parameters, jointly ensures the detection performance and operational stability of the system under different environments.

[0125] Based on the environment-adaptive OS-CFAR parameter combination, the detection threshold factor is calculated, and the target signal is identified by point-by-point detection of the spectrum data based on the detection threshold factor.

[0126] Furthermore, the detection threshold factor is calculated, specifically including:

[0127] To achieve the preset false alarm probability (For example This scheme is based on training units. With sorted index The required detection threshold factor is solved by back-calculating the exact model in the number domain. ,in , , and The following relationship must be satisfied:

[0128] ;

[0129] in: To detect threshold factors.

[0130] Furthermore, the step of performing point-by-point detection of the spectral data based on a detection threshold factor to identify the target signal specifically includes:

[0131] In applications, the current detection unit refers to the smallest data or signal processing unit selected at the current moment to compare with a preset threshold and complete the detection decision. Its definition and form will vary depending on the detection scenario, while the design dimension of the threshold needs to match the dimension of the current detection unit.

[0132] During point-by-point detection, with the current detection unit as the center, two arrays are set on both sides of its spectrum data. Several protection units are selected to prevent target energy from contaminating the background estimation; outside the protection units, on each side... Reference units, totaling One reference cell is used for background power estimation;

[0133] The power values ​​around the reference cell are sorted in ascending order based on the detection threshold factor to obtain a sequence. And select the first Minimum value As an estimate of the current background noise power;

[0134] Based on detection threshold factor Compared with the estimated current background noise power Calculate the detection threshold ;

[0135] The power value of each detection unit With detection threshold Comparison:

[0136] like If the signal is positive, it is identified as the target signal, and the frequency and amplitude at that point are recorded.

[0137] like If the value is not recorded, it is considered noise and will not be recorded.

[0138] After performing multi-dimensional target verification on the target signal, the final detection result is output, and the altimeter signal is obtained.

[0139] Furthermore, after performing multi-dimensional target verification on the target signal, the final detection result is output to obtain the platform's ground clearance information, specifically including:

[0140] Perform the following target verification on the target signal and eliminate target signals that do not meet the verification conditions:

[0141] Amplitude verification: Remove target signals whose amplitude is less than 10% of the maximum amplitude;

[0142] Frequency range verification: Remove target signals whose frequency is not within the frequency threshold;

[0143] Local peak verification: Remove target signals with insignificant local peaks;

[0144] Historical continuity verification: Remove target signals that deviate from historical detection results by more than 30%;

[0145] The target signal that meets the above verification conditions is taken as the final detection result, and the corresponding frequency and amplitude information is output. The distance is then converted according to the frequency difference to obtain the platform's height above the ground.

[0146] In this embodiment, the altimeter adopts the LFMCW system. A standard linear frequency modulated continuous wave baseband signal is generated through a digital generation device, which is then converted into an analog LFMCW baseband signal by a D / A converter. After upconversion with a point frequency source, a frequency modulated continuous wave microwave signal is obtained. After being amplified by a solid-state power amplifier, it is transmitted through a transmitting antenna.

[0147] The LFMCW system employs triangular wave modulation with consecutive positive and negative slopes. By averaging two consecutive frames of data, it can automatically cancel out altimeter errors caused by Doppler frequency shift, thus improving altimeter accuracy. The software flow for altimeter signal processing and data processing is as follows: Figure 2 As shown.

[0148] After the FMCW signal is reflected from the ground / sea surface, the receiving antenna receives the target echo. The receiver amplifies the echo, performs bandpass filtering and I / Q analog detection, and converts it into a digital signal via A / D conversion. The FPGA performs complex FFT processing on 8k data points to obtain the spectrum of the ground / sea surface echo. The echo spectrum is then sent to the DSP for data processing to obtain the frequency difference between the over-detection threshold, the maximum amplitude echo, and the transmitted signal. The frequency difference is then used to calculate the distance, thereby measuring the height of the platform above the ground / surface.

[0149] The core objective of the Ordered Statistical Constant False Alarm Rate (OS-CFAR) algorithm is to maintain a constant false alarm probability in complex environments. Its implementation process can be summarized into three stages: noise estimation, threshold calculation, and target decision. It sorts the echo power of each reference cell and selects a middle value as the average clutter power of the target cell to calculate the threshold. It is suitable for multi-target environments. The OS-CFAR detection principle is as follows: Figure 3 As shown.

[0150] This solution achieves three major advantages in the field of OS-CFAR detection technology: First, it creates a new paradigm of dual-dimensional environmental perception, introducing spectral entropy, an information-theoretic feature, to form an environmental fingerprint together with SCNR, thus solving the fundamental defect that traditional single power features cannot distinguish between strong targets and distributed interference. Second, it automatically mines the inherent patterns of the environment through unsupervised clustering of Gaussian mixture models (GMMs), and replaces the traditional hard classification decision by weighted fusion of multiple sets of preset parameters through the output of continuous responsibility vectors. This allows the detection parameters to continuously and smoothly adapt to the environment, eliminating performance fluctuations caused by step jumps in parameters. Third, by completing the complex model learning process in advance, the innovative paradigm of performing only lightweight, deterministic inference computation in the online stage balances computational efficiency, system stability, and engineering practicality while maintaining environmental adaptability, facilitating deployment in embedded radar systems. These three innovations together constitute a technological leap from a "static configuration tool" to an "environmentally perceptive intelligent agent," providing a completely new solution for radar detection in complex electromagnetic environments.

[0151] Example 2:

[0152] See Figure 4 A height gauge signal detection system based on dynamic parameter OS-CFAR, the system being applied to the method described in Example 1, the system comprising:

[0153] Input vector acquisition module 1 is used to acquire radar signals and perform preprocessing to obtain spectrum data, and calculate spectrum entropy and signal-to-clutter ratio based on the spectrum data;

[0154] The adaptive parameter generation module 2 is used to construct a two-dimensional feature vector from the spectral entropy and signal-to-noise ratio as input to the Gaussian mixture model for offline training and online execution, generating an environment-adaptive OS-CFAR parameter combination.

[0155] The target signal recognition module 3 is used to calculate the detection threshold factor based on the environment adaptive OS-CFAR parameter combination, and to perform point-by-point detection on the spectrum data based on the detection threshold factor to identify the target signal;

[0156] The altimeter signal acquisition module 4 is used to perform multi-dimensional target verification on the target signal and output the final detection result to obtain the altimeter signal.

[0157] Furthermore, for the specific implementation steps of the input vector acquisition module 1, adaptive parameter generation module 2, target signal recognition module 3, and altimeter signal acquisition module 4, please refer to the corresponding description in Embodiment 1, which will not be repeated here.

[0158] Example 3:

[0159] See Figure 5 A height meter signal detection device based on dynamic parameter OS-CFAR, the device includes a processor 5 and a memory 6;

[0160] The memory 6 is used to store computer program code 61 and transmit the computer program code 61 to the processor 5;

[0161] The processor 5 is used to execute the altimeter signal detection method based on dynamic parameter OS-CFAR as described in Embodiment 1 according to the instructions in the computer program code 61.

[0162] This embodiment also includes a computer-readable storage medium storing computer-executable instructions. When the computer-executable instructions are executed on a computer, the altimeter signal detection method based on dynamic parameter OS-CFAR described in Embodiment 1 is implemented.

[0163] Generally, the computer instructions for implementing the method of the present invention can be carried on any combination of one or more computer-readable storage media. Non-transitory computer-readable storage media can include any computer-readable medium except for the signal itself, which is temporarily propagating.

[0164] Computer-readable storage media can be, for example, but not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EKROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0165] Computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof. These programming languages ​​include object-oriented programming languages—such as Java, Smarttalk, and C++—as well as conventional procedural programming languages—such as the "C" language or similar programming languages. In particular, Python, suitable for neural network computation, and platform frameworks such as TensorFlow and PyTorch can be used. 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 or to an external computer (e.g., via the Internet using an Internet service provider) through any type of network, including a local area network (LAN) or a wide area network (WAN).

[0166] The aforementioned equipment and non-transitory computer-readable storage media can be found in the detailed description of a method for detecting altimeter signals based on dynamic parameter OS-CFAR and its beneficial effects, which will not be repeated here.

[0167] Although embodiments of the present invention have been shown and described above, it should be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for detecting altimeter signals based on dynamic parameter OS-CFAR, characterized in that, include: The radar signal is collected and preprocessed to obtain spectrum data, and the spectrum entropy and signal-to-clutter ratio are calculated based on the spectrum data. The spectral entropy and signal-to-noise ratio are used as a two-dimensional feature vector as input to the Gaussian mixture model for offline training and online execution, generating an environment-adaptive OS-CFAR parameter combination. Based on the environment-adaptive OS-CFAR parameter combination, the detection threshold factor is calculated, and the target signal is identified by point-by-point detection of the spectrum data based on the detection threshold factor. After performing multi-dimensional target verification on the target signal, the final detection result is output, and the altimeter signal is obtained.

2. The altimeter signal detection method based on dynamic parameter OS-CFAR according to claim 1, characterized in that: The calculation of spectral entropy and signal-to-noise ratio specifically includes: The spectral entropy is calculated using the following expression: ； ； in: For spectral entropy, Normalized power spectral density; For the first Power spectral density at each frequency point; This is the index of the frequency point being calculated. This refers to the traversal index in the summation operation; This represents the total number of frequency points in the spectrum data. The calculation of signal-to-noise ratio (SNR) specifically includes: The power spectrum is smoothed using an adaptive sliding window process based on the spectral data to obtain the smoothed power spectrum value, which is expressed as follows: ； in: For the first Power spectrum values ​​after smoothing at each frequency point; This represents the total number of sample points in the sliding window. For window points; For the summation sequence number; For the first The original power spectrum values ​​at each frequency point; The current frequency point number being processed; Sampling rate; To achieve smooth physical bandwidth; The number of FFT points; The spectral segments with a continuous spectral width greater than twice the window length in the smoothed power spectrum are divided into signal bands, and the remaining spectral segments are divided into clutter noise bands. The signal-to-noise ratio (SNR) is obtained by performing power integral calculations on the signal band and clutter noise band separately and then calculating the ratio.

3. The altimeter signal detection method based on dynamic parameter OS-CFAR according to claim 1, characterized in that: The offline training phase and the online execution phase specifically include: During the offline training phase, environment clustering is performed based on two-dimensional feature vectors to determine representative environment states. The representative environment states are then mapped to the corresponding ordered statistical constant false alarm rate (OS-CFAR) parameter combinations. An environment-parameter mapping strategy library is constructed and embedded in the Gaussian mixture model. During the online execution phase, the feature vector of the current environment state is extracted and input into the solidified Gaussian mixture model. The parameters of the environment-parameter mapping strategy library are called to perform responsibility weighting and coupling constraint verification, and generate an environment-adaptive OS-CFAR parameter combination.

4. The altimeter signal detection method based on dynamic parameter OS-CFAR according to claim 3, characterized in that: The environmental clustering process specifically includes: Two-dimensional feature vectors are input into a Gaussian mixture model, and the EM algorithm is used for environmental clustering to determine the cluster center point representing each environmental state; the representative environmental states include pure, moderate, complex, and harsh. Using the cluster centers corresponding to each environmental state as the object, the cluster centers with the highest signal-to-noise ratio and the lowest spectral entropy value are classified as pure environmental states, and the cluster centers with the lowest signal-to-noise ratio and the highest spectral entropy value are classified as harsh environmental states. Between cluster centers in clean and harsh environments, cluster centers with relatively high signal-to-noise ratios are classified as medium environments, and cluster centers with relatively low signal-to-noise ratios are classified as complex environments, based on their centrality.

5. The altimeter signal detection method based on dynamic parameter OS-CFAR according to claim 3, characterized in that: The online execution phase specifically includes: The system collects and extracts feature vectors of the current environmental state in real time, and inputs them into the solidified Gaussian mixture model to output a responsibility vector. ;in, For each component The current environmental state represented belongs to the first Each represents a probability weight of the environmental state; Recall the corresponding OS-CFAR parameter combination representing the environment state from the environment-parameter mapping policy library. Furthermore, through responsibility-weighted and coupling constraint verification, an environment-adaptive OS-CFAR parameter combination is generated, the expression of which is as follows: ； ； ； in: This represents the number of training units; The number of protection units; For sorting indexes.

6. The altimeter signal detection method based on dynamic parameter OS-CFAR according to claim 5, characterized in that: The coupling constraints specifically include: The training unit The value range is from 16 to 64, and the adjustment step is an integer multiple of 8; The protection unit The value range is 1 to 6, the adjustment step is 1, and the sum of the protection window and the training window does not exceed the sliding window processing capacity. The sorting index The value range is 0.

3. Up to 0.8 Adjust the step size to 1, and when When it increases from 32 to 48, The upper limit was automatically adjusted from 25 to 38.

7. The altimeter signal detection method based on dynamic parameter OS-CFAR according to claim 6, characterized in that: The calculation of the detection threshold factor based on the environment-adaptive OS-CFAR parameter combination specifically includes: Based on a preset false alarm probability Training Unit With sorted index The detection threshold factor is solved by inversely calculating the exact model in the number field, and its expression is as follows: ； in: To detect threshold factors.

8. The altimeter signal detection method based on dynamic parameter OS-CFAR according to claim 7, characterized in that: The step of performing point-by-point detection of spectral data based on a detection threshold factor to identify the target signal specifically includes: Centered on the current detection unit, set up on both sides of the spectrum data Each protection unit is equipped with a protective element, and the protective element is installed on the outside of the protective element. Each reference unit is selected on one side. One reference unit; The power values ​​around the reference cell are sorted in ascending order based on the detection threshold factor, and the first value is selected. The minimum power value is used as an estimate of the current background noise power; Based on detection threshold factor Compared with the estimated current background noise power Calculate the detection threshold and the power value of each detection unit With detection threshold Comparison: like If the signal is positive, it is identified as the target signal, and the frequency and amplitude at that point are recorded. like If the value is not recorded, it is considered noise and will not be recorded.

9. The altimeter signal detection method based on dynamic parameter OS-CFAR according to claim 1, characterized in that: After performing multi-dimensional target verification on the target signal, the final detection result is output to obtain the altimeter signal, specifically including: Perform the following target verification on the target signal and eliminate target signals that do not meet the verification conditions: Amplitude verification: Remove target signals whose amplitude is less than 10% of the maximum amplitude; Frequency range verification: Remove target signals whose frequency is not within the frequency threshold; Local peak verification: Remove target signals with insignificant local peaks; Historical continuity verification: Remove target signals that deviate from historical detection results by more than 30%; The target signal that meets the above verification conditions is taken as the final detection result, and the corresponding frequency and amplitude information is output. The distance is then converted according to the frequency difference to obtain the platform's height above the ground.

10. A height gauge signal detection system based on dynamic parameter OS-CFAR, characterized in that, The system is applied to the method according to any one of claims 1-9, the system comprising: The input vector acquisition module (1) is used to acquire radar signals and perform preprocessing to obtain spectrum data, and calculate spectrum entropy and signal-to-noise ratio based on the spectrum data; The adaptive parameter generation module (2) is used to construct a two-dimensional feature vector from the spectral entropy and signal-to-noise ratio as the input of the Gaussian mixture model, and to generate an environment-adaptive OS-CFAR parameter combination for the offline training stage and the online execution stage. The target signal recognition module (3) is used to calculate the detection threshold factor based on the environment adaptive OS-CFAR parameter combination, and to perform point-by-point detection on the spectrum data based on the detection threshold factor to identify the target signal; The altimeter signal acquisition module (4) is used to perform multi-dimensional target verification on the target signal and output the final detection result to obtain the altimeter signal.

Images