Passive sensing radar AI weak target detection method and system

Abstract

The invention provides a passive perception radar AI weak target detection method and system, and the method comprises the steps: obtaining a multi-dimensional electromagnetic signal data set of a target monitoring region, carrying out the signal feature extraction of the multi-dimensional electromagnetic signal data set, and obtaining the time-frequency distribution features and signal mode features corresponding to the multi-dimensional electromagnetic signal data set; based on a pre-trained weak target identification model, carrying out joint analysis processing on the time-frequency distribution characteristics and the signal mode characteristics, and generating weak target confidence parameters of potential weak target reflection signals; generating a weak target detection result of the target monitoring area according to a comparison result of the weak target confidence coefficient parameter and a preset confidence coefficient threshold value; and transmitting the weak target detection result to the target terminal device to trigger an alarm response operation. According to the method, the problem of characteristic loss caused by signal preprocessing and target detection module separation in a traditional method is solved, and the real-time performance of weak target detection and the system robustness are remarkably improved.

Description

Technical Field

[0001] This invention relates to the fields of data processing and artificial intelligence, and in particular to a passive sensing radar AI weak target detection method and system. Background Technology

[0002] With the evolution of radar detection technology, the detection of weak targets in complex electromagnetic environments by passive sensing radar has gradually become a key research direction. This technical field mainly addresses the problem of reliable identification of targets with low signal-to-noise ratio and low cross-sectional area under strong environmental clutter interference. Among existing typical methods, some schemes extract the instantaneous energy distribution characteristics of the target signal through time-frequency analysis, but ignore steady-state characteristics such as signal modulation patterns; others rely on periodic feature detection algorithms, but struggle to cope with the time-varying characteristics of signals in non-stationary environments. These methods have inherent flaws: First, they lack an effective decoupling mechanism for the co-originating characteristics of the target signal and environmental interference, leading to the target signal being easily submerged by background noise during feature extraction; second, the fixed threshold decision-making model is difficult to adapt to the dynamic fluctuations of the signal-to-noise ratio in complex electromagnetic environments, causing an imbalance between false alarm rate and false negative rate; third, the traditional separation architecture of signal processing and target recognition modules causes the correlation information between time-frequency features and pattern features to attenuate during transmission, weakening the ability to distinguish weak targets. These problems severely restrict the effective detection capability of stealth targets and micro-UAVs in key scenarios such as electronic warfare and low-altitude surveillance. Summary of the Invention

[0003] In view of this, embodiments of the present invention provide at least one passive sensing radar AI weak target detection method and system.

[0004] The technical solution of this invention is implemented as follows: On one hand, embodiments of the present invention provide a passive sensing radar AI weak target detection method, including: Acquire a multi-dimensional electromagnetic signal data set of the target monitoring area, wherein the multi-dimensional electromagnetic signal data set includes environmental clutter signals and potential weak target reflection signals; The multi-dimensional electromagnetic signal data set is subjected to signal feature extraction processing to obtain the time-frequency distribution features and signal mode features corresponding to the multi-dimensional electromagnetic signal data set; Based on a pre-trained weak target identification model, the time-frequency distribution features and the signal pattern features are jointly analyzed and processed to generate the weak target confidence parameters of the potential weak target reflection signal. Based on the comparison between the weak target confidence parameter and the preset confidence threshold, the weak target detection result of the target monitoring area is generated; The weak target detection results are transmitted to the target terminal device to trigger an alarm response operation.

[0005] On the other hand, embodiments of the present invention provide a computer system including a memory and a processor, wherein the memory stores a computer program that can run on the processor, and the processor executes the program to implement the steps in the above-described method.

[0006] The passive sensing radar AI weak target detection method provided by this invention acquires a multi-dimensional electromagnetic signal data set containing environmental clutter signals and potential weak target reflection signals, and jointly extracts time-frequency distribution features and signal mode features. It then uses a pre-trained weak target recognition model to perform dual-modal feature fusion analysis to generate dynamic weak target confidence parameters. Finally, it combines preset thresholds to make decisions, generate detection results, and trigger alarm responses. This method effectively distinguishes weak target features in complex background noise through a synchronous processing mechanism of environmental interference signals and target reflection signals, enhancing target identifiability in low signal-to-noise ratio scenarios. The complementary extraction strategy of time-frequency distribution features and signal pattern features overcomes the limitations of single features in representing non-stationary signals, strengthening the correlation modeling ability between the periodic modulation characteristics and instantaneous energy changes of weak target reflection signals. The decision-making mechanism based on dynamic confidence parameters transforms traditional binary detection into interpretable judgments in a continuous probability space, avoiding the poor environmental adaptability of fixed thresholds, while retaining the uncertainty information of the detection results to support multi-level alarm responses. The end-to-end processing link design realizes feature transfer from the raw signal to the decision response. Through deep coupling of time-frequency analysis, pattern recognition, and AI models, it solves the feature loss problem caused by the separation of signal preprocessing and target detection modules in traditional methods, significantly improving the real-time performance and system robustness of weak target detection.

[0007] It should be understood that the above general description and the following detailed description are merely exemplary and explanatory, and are not intended to limit the technical solutions of the present invention. Attached Figure Description

[0008] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present invention and, together with the specification, serve to explain the technical solutions of the present invention.

[0009] Figure 1 This is a schematic diagram illustrating the implementation process of a passive sensing radar AI weak target detection method provided in an embodiment of the present invention.

[0010] Figure 2 This is a schematic diagram of the hardware entity of a computer system provided in an embodiment of the present invention. Detailed Implementation

[0011] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. The described embodiments should not be regarded as limitations on 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.

[0012] In the following description, references to "some embodiments" describe a subset of all possible embodiments; however, it is understood that "some embodiments" may be the same or different subsets of all possible embodiments and may be combined with each other without conflict. The terms "first / second / third" are used merely to distinguish similar objects and do not represent a specific ordering of objects. It is understood that "first / second / third" may be interchanged in a specific order or sequence where permissible, so that the embodiments of the invention described herein can be implemented in an order other than that illustrated or described herein.

[0013] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein is for descriptive purposes only and is not intended to limit the scope of the invention.

[0014] This invention provides a passive sensing radar AI-based weak target detection method, which can be executed by a computer system's processor. The computer system can refer to devices with data processing capabilities, such as servers, laptops, tablets, and desktop computers.

[0015] Figure 1 This is a schematic diagram illustrating the implementation process of a passive sensing radar AI weak target detection method provided in an embodiment of the present invention, as shown below. Figure 1 As shown, the method includes the following steps: Step S100: Obtain a multi-dimensional electromagnetic signal data set of the target monitoring area, wherein the multi-dimensional electromagnetic signal data set includes environmental clutter signals and potential weak target reflection signals.

[0016] The target monitoring area is the spatial range within which weak target detection is required. This range is defined based on actual monitoring needs, such as monitoring a military defense area or an industrial production area. The multi-dimensional electromagnetic signal data set is the sum of electromagnetic signal data with multiple dimensions of information collected within the target monitoring area. These dimensions can include signal frequency, amplitude, phase, and time. Environmental clutter signals are electromagnetic signals within the target monitoring area other than those reflected by potential weak targets. These typically originate from various interference sources in the surrounding environment, such as lightning and atmospheric noise in the natural environment, and radiation from electronic devices in the man-made environment. Potential weak target reflection signals are electromagnetic signals that may be reflected back by weak targets within the target monitoring area. These weak targets may be small aircraft, stealth targets, etc., and due to their weak reflection signal strength, they are easily masked by environmental clutter signals.

[0017] Acquiring a multi-dimensional electromagnetic signal data set of a target monitoring area can be achieved by deploying multiple electromagnetic signal sensors of different types within the monitoring area. For example, in a military defense zone, multiple antenna sensors capable of receiving electromagnetic signals across different frequency ranges can be deployed. These sensors digitize the received electromagnetic signals, converting them into digital signals before transmitting them to a data processing center. At the data processing center, these digital signals are integrated and classified to obtain a multi-dimensional electromagnetic signal data set that includes environmental clutter signals and reflection signals from potential weak targets.

[0018] Step S200: Perform signal feature extraction processing on the multi-dimensional electromagnetic signal data set to obtain the time-frequency distribution features and signal mode features corresponding to the multi-dimensional electromagnetic signal data set.

[0019] Signal feature extraction is the process of extracting information that reflects the essential characteristics of a signal from a multi-dimensional electromagnetic signal dataset. Time-frequency distribution features describe the distribution characteristics of a signal in both time and frequency dimensions, reflecting the distribution of the signal's energy over time and frequency. Signal pattern features describe the established patterns and regularities of a signal, helping to identify the signal's type and origin. Various signal processing algorithms and techniques can be employed to perform signal feature extraction on multi-dimensional electromagnetic signal datasets. For example, Fourier transform can be used to convert the signal from the time domain to the frequency domain, thus obtaining the signal's frequency distribution information; wavelet transform can also be used for multi-resolution analysis to extract features at different scales. Through these processes, the time-frequency distribution features and signal pattern features corresponding to the multi-dimensional electromagnetic signal dataset can be obtained.

[0020] As one implementation, step S200, which involves performing signal feature extraction processing on the multi-dimensional electromagnetic signal data set to obtain the time-frequency distribution features and signal mode features corresponding to the multi-dimensional electromagnetic signal data set, may specifically include the following steps S210~S240: Step S210: Perform noise reduction processing on the environmental clutter signals in the multi-dimensional electromagnetic signal data set to obtain a noise-reduced environmental clutter signal set.

[0021] Signal denoising is the process of removing noise components from a signal, with the aim of improving signal quality and analyzability. In multi-dimensional electromagnetic signal datasets, environmental clutter signals contain a significant amount of noise, which can interfere with subsequent signal analysis and processing. Signal denoising can reduce the impact of noise on environmental clutter signals, resulting in a cleaner dataset.

[0022] Various denoising algorithms can be used to denoise environmental clutter signals, such as mean filtering, median filtering, and wavelet thresholding. Taking wavelet thresholding as an example, the environmental clutter signal is first decomposed into wavelet coefficients at different scales. Then, these wavelet coefficients are processed according to a set threshold, setting wavelet coefficients smaller than the threshold to zero, thereby removing noise components. Finally, wavelet reconstruction is performed on the processed wavelet coefficients to obtain the denoised environmental clutter signal set.

[0023] Step S220: Perform time-frequency analysis processing on the noise-reduced environmental clutter signal set to generate the time-frequency energy distribution characteristics of the environmental clutter signal set.

[0024] Time-frequency analysis is a method that simultaneously analyzes the characteristics of a signal in both time and frequency dimensions, revealing the energy distribution patterns of the signal in these dimensions. Time-frequency energy distribution characteristics describe the energy distribution of an environmental clutter signal array in the time-frequency plane, aiding in the analysis of the characteristics and variation patterns of environmental clutter signals.

[0025] Various time-frequency analysis methods can be used to perform time-frequency analysis on the denoised environmental clutter signal set, such as short-time Fourier transform, Wigner-Ville distribution, and wavelet transform. Taking short-time Fourier transform as an example, the denoised environmental clutter signal is divided into multiple short time intervals, and a Fourier transform is performed on the signal in each short time interval to obtain the spectral information of each time interval. Then, these spectral information are arranged in chronological order to form an energy distribution image on the time-frequency plane, thereby generating the time-frequency energy distribution characteristics of the environmental clutter signal set.

[0026] Step S230: Perform signal pattern recognition processing on the potential weak target reflection signal to obtain the periodicity and modulation characteristics of the potential weak target reflection signal.

[0027] Signal pattern recognition processing is the process of identifying and analyzing the patterns and regularities of signals. Its purpose is to extract feature information from the signals for classification and identification. Periodicity features characterize the repetitive regularity of signals reflected by potential weak targets, reflecting the periodic changes of the signal over time. Modulation features characterize the frequency modulation properties of signals reflected by potential weak targets, reflecting the frequency variation of the signal.

[0028] Various pattern recognition algorithms can be used to process the reflected signals of potential weak targets, such as autocorrelation analysis, cyclostationary analysis, and instantaneous frequency estimation. These algorithms can extract the periodic and modulation characteristics of the reflected signals from potential weak targets.

[0029] As one implementation, step S230, which involves performing signal pattern recognition processing on the potential weak target reflection signal to obtain the periodicity and modulation characteristics of the potential weak target reflection signal, may specifically include the following steps S231-S234: Step S231: Perform signal segmentation processing on the potential weak target reflection signal to obtain multiple signal segment sequences.

[0030] Signal segmentation is the process of dividing the reflection signal of a potential weak target into multiple shorter signal segments according to a set rule. By segmenting the signal, a long signal can be decomposed into multiple shorter signals, facilitating subsequent analysis and processing. Signal segmentation of the reflection signal of a potential weak target can be performed using fixed-length segmentation or adaptive segmentation methods. Taking fixed-length segmentation as an example, the reflection signal of a potential weak target is divided into multiple signal segments of equal length according to a pre-set segment length, thus obtaining multiple signal segment sequences.

[0031] Step S232: Calculate the autocorrelation function for each of the signal segment sequences to generate periodic characteristic parameters corresponding to each of the signal segment sequences; wherein, the periodicity feature is used to characterize the repetitive regularity of the potential weak target reflection signal, and the modulation feature is used to characterize the frequency modulation characteristics of the potential weak target reflection signal.

[0032] The autocorrelation function describes the correlation between signals themselves, reflecting the degree of similarity between signals at different time points. By calculating the autocorrelation function of a segmented sequence of signals, the periodicity information of the signal can be obtained. Periodicity parameters are parameters used to describe the periodicity of a signal, helping to determine the period and repetition pattern of the signal.

[0033] The autocorrelation function of each signal segment sequence can be calculated using the following formula: Let the signal segment sequence be x(n), and its autocorrelation function R xx (m) is defined as Where N is the length of the signal segment sequence and m is the delay time. By calculating the autocorrelation function values ​​at different delay times, the autocorrelation function curve can be obtained. From the autocorrelation function curve, periodic characteristic parameters, such as the signal period and autocorrelation peak value, can be extracted.

[0034] Step S233: Perform statistical analysis on the periodic characteristic parameters to obtain the periodic characteristics.

[0035] Statistical analysis is the process of collecting, organizing, analyzing, and interpreting data, with the aim of extracting useful information and patterns. Statistical analysis of periodic characteristic parameters can reveal the periodic characteristics of potential weak target reflection signals, such as the average period and period variance. Various statistical methods can be used for this analysis, such as mean calculation, variance calculation, and histogram analysis. Taking mean calculation as an example, the periodic characteristic parameters of all signal segment sequences are summed, and then divided by the number of signal segment sequences to obtain the average value of the periodic characteristic parameters, thus yielding the average period of the potential weak target reflection signal. Statistical analysis of periodic characteristic parameters allows for a more accurate description of the periodic characteristics of potential weak target reflection signals.

[0036] Step S234: Perform instantaneous frequency estimation processing on the potential weak target reflection signal to generate the frequency change trajectory of the potential weak target reflection signal.

[0037] Instantaneous frequency estimation is the process of estimating the instantaneous frequency of a signal at each moment. Instantaneous frequency is the frequency of a signal at a specific instant, reflecting the real-time changes in signal frequency. A frequency change trajectory is a curve describing the change of the instantaneous frequency of a potential weak target's reflected signal over time, which helps in analyzing the signal's frequency modulation characteristics. Various instantaneous frequency estimation methods can be used for this process, such as the Hilbert transform method and the zero-crossing method. Taking the Hilbert transform method as an example, firstly, a Hilbert transform is performed on the potential weak target's reflected signal to obtain its analytic signal. Then, the instantaneous frequency is obtained by calculating the phase derivative of the analytic signal. Finally, the change of the instantaneous frequency over time is recorded, generating the frequency change trajectory of the potential weak target's reflected signal.

[0038] Step S235: Perform slope analysis on the frequency change trajectory to obtain the modulation characteristics.

[0039] Slope analysis is the process of calculating and analyzing the slope of a curve, with the aim of revealing the rate of change and trend of the curve. By performing slope analysis on the frequency change trajectory, the modulation characteristics of the reflected signal from a potential weak target can be obtained, such as the rate of frequency change and the modulation slope.

[0040] Numerical differentiation can be used to analyze the slope of frequency change trajectories. First, the frequency change trajectory is discretized, resulting in a series of discrete frequency values ​​and corresponding time points. Then, the frequency difference between two adjacent time points is calculated and divided by the time interval to obtain the rate of frequency change, i.e., the slope, within that interval. By calculating for all adjacent time points, the slope distribution of the frequency change trajectory is obtained. From the slope distribution, modulation features, such as the maximum slope and average slope, can be extracted.

[0041] Step S240: The time-frequency energy distribution feature is fused with the periodicity feature and the modulation feature to generate the signal pattern feature; wherein, the time-frequency distribution feature includes the correlation feature between the time-frequency energy distribution feature and the time-frequency offset of the potential weak target reflection signal.

[0042] Feature fusion is the process of combining and integrating different types of features. Its purpose is to synthesize information from multiple features to obtain a more comprehensive and effective feature representation. Time-frequency offset is the offset of the reflected signal from a potential weak target in time and frequency relative to a reference signal, reflecting the time-frequency positional changes of the signal. The correlation between the time-frequency energy distribution features and the time-frequency offset of the reflected signal from the potential weak target can describe the relationship between the energy distribution and positional changes of the signal in the time-frequency plane.

[0043] Feature fusion of time-frequency energy distribution characteristics with periodic and modulation characteristics can be achieved using various methods, such as weighted summation and feature concatenation. Taking weighted summation as an example, each feature is assigned a weight coefficient, then each feature is multiplied by its corresponding weight coefficient, and finally the weighted features are summed to obtain the signal pattern characteristics. Through feature fusion, information from time-frequency energy distribution characteristics, periodic characteristics, and modulation characteristics can be integrated to generate more representative signal pattern characteristics.

[0044] Step S300: Based on the pre-trained weak target identification model, the time-frequency distribution features and the signal pattern features are jointly analyzed and processed to generate the weak target confidence parameters of the potential weak target reflection signal.

[0045] A pre-trained weak target recognition model is a model trained on a large amount of sample data. It can analyze and judge the input features and output the confidence score of whether a potential weak target's reflected signal is a weak target. The weak target confidence parameter is used to represent the probability that a potential weak target's reflected signal is a weak target. Its value is usually between 0 and 1, and the closer the value is to 1, the greater the probability that it is a weak target.

[0046] Based on a pre-trained weak target recognition model, the joint analysis and processing of time-frequency distribution features and signal pattern features can be carried out in the following way: the time-frequency distribution features and signal pattern features are input into the pre-trained weak target recognition model, the model analyzes and processes these features, and calculates the weak target confidence parameters of the potential weak target reflection signal through the internal neural network structure and algorithm.

[0047] As one implementation method, the pre-trained weak target recognition model is obtained through the following steps S301~S305: Step S301: Obtain a sample electromagnetic signal dataset of the historical monitoring area. The sample electromagnetic signal dataset includes environmental clutter signal samples and weak target reflection signal samples labeled with the presence status of weak targets.

[0048] The historical monitoring area refers to areas that have been monitored in the past, and the electromagnetic signal data from these areas can be used as training samples. The sample electromagnetic signal dataset is a collection of electromagnetic signal data gathered from the historical monitoring area, including environmental clutter signal samples and weak target reflection signal samples. Labeling the presence of weak targets involves marking each sample signal to indicate whether a weak target is present in that signal.

[0049] Obtaining a dataset of sample electromagnetic signals from a historical monitoring area can be achieved by reviewing past monitoring records and extracting relevant electromagnetic signal data from data storage devices. During data extraction, the data needs to be filtered and categorized, separating environmental clutter signal samples from weak target reflection signal samples, and labeling each sample signal to record the presence of weak targets.

[0050] Step S302: Perform signal enhancement processing on the sample electromagnetic signal dataset to obtain the enhanced sample electromagnetic signal dataset.

[0051] Signal enhancement is the process of improving the quality and discernibility of signals through various signal processing techniques. In sample electromagnetic signal datasets, signal quality may be low due to factors such as environmental noise and signal attenuation. Signal enhancement can improve signal features and enhance model training performance. Various methods can be used to enhance sample electromagnetic signal datasets, such as filtering, amplification, and denoising. Taking filtering as an example, bandpass filters can be used to filter sample electromagnetic signals, removing unwanted frequency components and retaining the frequency range relevant to weak target reflection signals, thereby enhancing signal features.

[0052] Step S303: Perform feature extraction processing on the enhanced sample electromagnetic signal dataset to obtain the sample time-frequency distribution features and sample signal mode features corresponding to each sample signal.

[0053] The purpose of feature extraction is to extract information that reflects the essential characteristics of the signal from the enhanced sample electromagnetic signal dataset. The time-frequency distribution characteristics of the samples describe the distribution characteristics of the sample signal in both time and frequency dimensions, while the sample signal pattern characteristics describe the predetermined patterns and regularities of the sample signal. Feature extraction processing of the enhanced sample electromagnetic signal dataset can be performed using methods similar to those in step S200, including multi-scale wavelet transform processing of environmental clutter signal samples and cyclostationary analysis processing of weak target reflection signal samples.

[0054] As one implementation, step S303, which involves feature extraction processing of the enhanced sample electromagnetic signal dataset, may specifically include the following steps S3031~S3035: Step S3031: Perform multi-scale wavelet transform processing on the environmental clutter signal sample to generate multi-resolution energy distribution characteristics of the environmental clutter signal sample.

[0055] Multiscale wavelet transform is a method that can analyze signals at different scales, decomposing signals into sub-signals of different frequencies and resolutions. Multi-resolution energy distribution characteristics describe the energy distribution of environmental clutter signal samples at different scales and resolutions, reflecting the frequency components and energy distribution patterns of the signal.

[0056] Multi-scale wavelet transform processing of environmental clutter signal samples can be performed using the following steps: First, select a suitable wavelet basis function, such as the Daubechies wavelet or Symlets wavelet. Then, perform wavelet decomposition on the environmental clutter signal samples, decomposing the signal into approximate components and detail components at different scales. Next, calculate the energy of each component to obtain the multi-resolution energy distribution characteristics of the environmental clutter signal samples.

[0057] Step S3032: Perform cyclic stationary analysis on the weak target reflection signal sample to generate the cyclic frequency characteristics of the weak target reflection signal sample.

[0058] Cyclostationary analysis is a method used to analyze signals exhibiting cyclostationary characteristics, revealing the periodicity and cyclic nature of the signal. Cyclic frequency characteristics describe the features of weak target reflection signal samples in the cyclic frequency domain, reflecting the signal's periodicity and modulation characteristics. Cyclostationary analysis of weak target reflection signal samples can be performed using methods such as cyclic autocorrelation function (CEC) and cyclic spectrum analysis. Taking the CEC as an example, the CEC of the weak target reflection signal sample is calculated, obtaining the autocorrelation characteristics of the signal at different cyclic frequencies. Then, a Fourier transform is performed on the CEC to obtain the cyclic spectrum, thus generating the cyclic frequency characteristics of the weak target reflection signal sample.

[0059] Step S3033: Perform feature alignment processing on the multi-resolution energy distribution features and the cyclic frequency features to generate the sample time-frequency distribution features; wherein, the feature alignment processing is used to eliminate the time-frequency offset difference between the environmental clutter signal samples and the weak target reflection signal samples.

[0060] Feature alignment is the process of aligning and matching different types of features. Its purpose is to eliminate the time-frequency offset differences between environmental clutter signal samples and weak target reflection signal samples, ensuring they have the same position and scale in the time-frequency plane. Sample time-frequency distribution features are time-frequency features that integrate multi-resolution energy distribution features and cyclic frequency features, providing a more comprehensive description of the time-frequency characteristics of the sample signal. The feature alignment process between multi-resolution energy distribution features and cyclic frequency features can be performed as follows: First, determine the time-frequency offset between the environmental clutter signal samples and the weak target reflection signal samples. Then, adjust the multi-resolution energy distribution features and cyclic frequency features according to the time-frequency offset to align them in the time-frequency plane. Finally, fuse the aligned features to generate the sample time-frequency distribution features.

[0061] Step S3034: Perform high-order cumulative calculation processing on the weak target reflection signal sample to obtain the high-order statistical features of the weak target reflection signal sample.

[0062] Higher-order cumulants are statistical quantities used to describe the higher-order statistical properties of a signal, reflecting its non-Gaussian and non-linear characteristics. Higher-order statistical features describe the higher-order statistical properties of weak target reflected signal samples, providing more information about the signal and helping to improve the model's recognition capabilities.

[0063] The following formula can be used to calculate the higher-order cumulants of weak target reflection signal samples: Let the weak target reflection signal sample be x(n), and its k-th order cumulant... Defined as: Where E represents the expectation and p represents different combinations. By calculating higher-order cumulants of different orders, the higher-order statistical characteristics of weak target reflection signal samples can be obtained.

[0064] Step S3035: Perform dimensionality compression processing on the higher-order statistical features to generate the sample signal pattern features.

[0065] Dimensionality compression is the process of compressing high-dimensional feature data into low-dimensional feature data. Its purpose is to reduce the dimensionality of features, lower computational complexity, and retain the main information of the features. Sample signal pattern features are high-order statistical features after dimensionality compression, which can more effectively represent the patterns and regularities of weak target reflection signal samples.

[0066] Various methods can be used to compress the dimensionality of higher-order statistical features, such as Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA). Taking PCA as an example, the covariance matrix of the higher-order statistical features is first calculated, and then the eigenvalues ​​and eigenvectors of the covariance matrix are solved. The eigenvectors with larger eigenvalues ​​are selected as principal components, and the higher-order statistical features are projected onto the principal components to obtain the dimensionality-reduced sample signal pattern features.

[0067] Step S304: Construct an initial weak target recognition model, which includes a feature fusion layer and a classification decision layer.

[0068] The initial weak target identification model is used to identify and classify the reflection signals of potential weak targets. It consists of a feature fusion layer and a classification decision layer. The function of the feature fusion layer is to fuse the time-frequency distribution features and signal pattern features of the input samples to obtain a comprehensive feature representation. The function of the classification decision layer is to make a classification decision based on the fused features and output the confidence level of whether the reflection signal of the potential weak target is a weak target.

[0069] The initial weak target recognition model can be constructed using deep learning methods, such as multilayer perceptrons (MLPs) and convolutional neural networks (CNNs). Taking a multilayer perceptron as an example, the feature fusion layer can consist of multiple fully connected layers used to perform nonlinear transformations and fusion of the input features. The classification decision layer can be an output layer, using a softmax function to convert the output into a probability distribution, thereby obtaining the weak target confidence parameters.

[0070] Step S305: Based on the time-frequency distribution features of the samples and the signal pattern features of the samples, iteratively train the initial weak target recognition model until the error between the output of the classification decision layer and the labeling result of the weak target existence state is less than a preset threshold, thereby obtaining the pre-trained weak target recognition model.

[0071] Iterative training is the process of updating the model's parameters through multiple iterations, gradually bringing the model's output closer to the labeled results. The preset threshold is a pre-defined error value; when the error between the output of the classification decision layer and the labeled results for the weak target's presence is less than this threshold, the model training is considered to have achieved satisfactory results.

[0072] Based on the time-frequency distribution features and signal pattern features of the samples, iterative training of the initial weak target recognition model can employ optimization algorithms such as gradient descent. The specific steps are as follows: First, input the time-frequency distribution features and signal pattern features of the samples into the initial weak target recognition model to obtain the model's output. Then, calculate the error between the output and the labeled weak target presence state, for example, using the cross-entropy loss function. Next, calculate the gradient of the model parameters based on the error and update the model parameters using the optimization algorithm. Repeat the above steps until the error is less than a preset threshold, obtaining the pre-trained weak target recognition model.

[0073] As one implementation method, step S300, based on a pre-trained weak target recognition model, performs joint analysis and processing on the time-frequency distribution features and the signal pattern features to generate weak target confidence parameters of the potential weak target reflection signal, specifically including the following steps S310~S340: Step S310: Input the time-frequency distribution features into the feature encoding layer of the pre-trained weak target recognition model to obtain the first intermediate feature vector.

[0074] The feature encoding layer is a network layer in a pre-trained weak target recognition model. Its function is to encode and transform the input time-frequency distribution features, mapping them to a low-dimensional feature space to obtain the first intermediate feature vector. The first intermediate feature vector is the feature representation of the time-frequency distribution features after processing by the feature encoding layer, and it can more effectively represent the information of the time-frequency distribution features. The time-frequency distribution features are input into the feature encoding layer of the pre-trained weak target recognition model, and neural network structures such as fully connected layers and convolutional layers can be used. Taking a fully connected layer as an example, the time-frequency distribution features are used as input, and the first intermediate feature vector is obtained through the linear transformation of the fully connected layer and the non-linear transformation of the activation function.

[0075] Step S320: Input the signal pattern features into the feature transformation layer of the pre-trained weak target recognition model to obtain the second intermediate feature vector.

[0076] The feature transformation layer is another network layer in the pre-trained weak target recognition model. Its function is to transform and process the input signal pattern features, converting them into a feature representation suitable for subsequent analysis, thus obtaining the second intermediate feature vector. The second intermediate feature vector is the feature representation of the signal pattern features after processing by the feature transformation layer, and it can better reflect the characteristics of the signal pattern features. Inputting the signal pattern features into the feature transformation layer of the pre-trained weak target recognition model can employ a neural network structure similar to the feature encoding layer. For example, fully connected layers or convolutional layers can be used to process the signal pattern features to obtain the second intermediate feature vector.

[0077] Step S330: Perform weighted fusion processing on the first intermediate feature vector and the second intermediate feature vector to obtain the target fused feature vector; wherein, the weight coefficients of the weighted fusion processing are adjusted according to the correlation between the time-frequency distribution features and the signal pattern features.

[0078] Weighted fusion is the process of fusing the first and second intermediate feature vectors according to predetermined weights. Its purpose is to integrate the information from the two feature vectors to obtain a more comprehensive and effective feature representation. Weighting coefficients are used to control the proportion of the first and second intermediate feature vectors in the fusion process, and are adjusted based on the correlation between time-frequency distribution features and signal mode features. The target fused feature vector is the feature vector obtained after weighted fusion, integrating information from both time-frequency distribution features and signal mode features. The weighted fusion of the first and second intermediate feature vectors can be performed using the following steps: First, calculate the correlation between the time-frequency distribution features and signal mode features, for example, using a correlation coefficient. Then, adjust the weighting coefficients according to the correlation; feature vectors with higher correlation are assigned greater weights. Finally, multiply the first and second intermediate feature vectors by their respective weighting coefficients, and then sum the weighted feature vectors to obtain the target fused feature vector.

[0079] As one implementation, step S330, which involves weighted fusion of the first intermediate feature vector and the second intermediate feature vector to obtain the target fused feature vector, may specifically include the following steps S331-S338: Step S331: Perform time dimension distribution analysis on the first intermediate feature vector to generate the time-series correlation weights corresponding to the time-frequency distribution features.

[0080] Temporal distribution analysis is the process of analyzing the distribution of the first intermediate feature vector along the time dimension, with the aim of uncovering the temporal correlation information of time-frequency distribution features. Temporal correlation weights are weights used to represent the importance of time-frequency distribution features at different points in time, reflecting the correlation of time-frequency distribution features along the time dimension.

[0081] Various methods can be used to perform time-dimensional distribution analysis on the first intermediate feature vector, such as autocorrelation analysis and time series analysis. Taking autocorrelation analysis as an example, the autocorrelation coefficient of the first intermediate feature vector at different time delays is calculated to obtain the autocorrelation function. Based on the results of the autocorrelation function, the correlation of the time-frequency distribution characteristics at different time points is determined, thereby generating time-series correlation weights.

[0082] Step S332: Perform frequency dimension distribution analysis on the second intermediate feature vector to generate frequency domain correlation weights corresponding to the signal mode features.

[0083] Frequency-dimensional distribution analysis is the process of analyzing the distribution of the second intermediate feature vector along the frequency dimension. Its purpose is to uncover the frequency-related information of signal mode features. Frequency domain correlation weights are weights used to represent the importance of signal mode features at different frequencies, reflecting the correlation of signal mode features along the frequency dimension.

[0084] Frequency-dimensional distribution analysis of the second intermediate eigenvector can be performed using methods similar to those used for time-dimensional distribution analysis, such as spectral analysis and power spectral density analysis. Taking spectral analysis as an example, a Fourier transform is performed on the second intermediate eigenvector to obtain its spectrum. Based on the spectrum results, the energy distribution of the signal mode characteristics at different frequencies is determined, thereby generating frequency domain correlation weights.

[0085] Step S333: Construct an intermediate weight matrix based on the time-series correlation weight and the frequency-domain correlation weight, and extract the environmental interference features of the potential weak target reflection signal.

[0086] The intermediate weight matrix is ​​a matrix that combines temporal correlation weights and frequency domain correlation weights, and can be used to weight and fuse the first and second intermediate eigenvectors. Environmental interference features are characteristics of the reflected signals from potential weak targets that are affected by environmental interference, reflecting the degree of influence of environmental interference on the signal.

[0087] The intermediate weight matrix can be constructed based on temporal correlation weights and frequency domain correlation weights using the following method: combine the temporal correlation weights and frequency domain correlation weights to form a two-dimensional matrix, namely the intermediate weight matrix. Environmental interference features of potential weak target reflection signals can be extracted using filtering, denoising, and other methods to separate environmental interference components from the signal, thus obtaining the environmental interference features.

[0088] Step S334: Perform interference suppression processing on the intermediate weight matrix according to the environmental interference characteristics to generate an optimized weight matrix.

[0089] Interference suppression is a process of adjusting the intermediate weight matrix to reduce the impact of environmental interference on the weighted fusion process. The optimized weight matrix, obtained after interference suppression, more accurately reflects the importance of time-frequency distribution characteristics and signal mode features. Interference suppression of the intermediate weight matrix based on environmental interference characteristics can be achieved using the following method: First, determine the intensity and distribution of the interference based on the environmental interference characteristics. Then, adjust the elements in the intermediate weight matrix that are significantly affected by interference, reducing their weights. Finally, the optimized weight matrix is ​​obtained.

[0090] Step S335: Perform cross-projection calculation on the first intermediate feature vector and the second intermediate feature vector based on the optimized weight matrix to obtain the first projected feature vector and the second projected feature vector.

[0091] Cross-projection calculation is the process of projecting the first and second intermediate eigenvectors according to an optimized weight matrix. Its purpose is to fuse and transform the information of the two eigenvectors. The first and second projected eigenvectors are eigenvectors obtained after cross-projection calculation, and they respectively contain partial information of time-frequency distribution characteristics and signal mode characteristics.

[0092] The cross-projection calculation of the first and second intermediate eigenvectors based on the optimized weight matrix can be performed using the following formula: Let the first intermediate eigenvector be x, the second intermediate eigenvector be y, and the optimized weight matrix be W, then the first projected eigenvector... Second projection feature vector .

[0093] Step S336: Perform multi-level feature interaction processing on the first projection feature vector and the second projection feature vector to generate an interactive feature vector; wherein, the multi-level feature interaction processing includes: splitting the first projection feature vector into multiple sub-vector segments, performing segment-by-segment convolution processing on each sub-vector segment and the second projection feature vector respectively to generate a set of convolutional feature segments, and performing cross-segment attention allocation processing on each convolutional feature segment to generate the interactive feature vector.

[0094] Multi-level feature interaction processing is a process of multi-level and multi-dimensional interaction between the first and second projected feature vectors, with the aim of uncovering the potential correlation information between the two feature vectors. The interactive feature vector is the feature vector obtained after multi-level feature interaction processing, which integrates the information of the first and second projected feature vectors.

[0095] The first projected feature vector can be divided into multiple sub-vector segments according to a set length or rule. Each sub-vector segment can be convolved with the second projected feature vector segment by segment using the convolution operation in a convolutional neural network, resulting in a set of convolutional feature segments. Cross-segment attention allocation can be applied to each convolutional feature segment using an attention mechanism, assigning different attention weights to each segment to highlight important feature information, thereby generating an interactive feature vector.

[0096] Step S337: Perform a dot product operation between the interactive feature vector and the optimized weight matrix to generate a fusion transition vector.

[0097] The fusion transition vector is a vector obtained after dot product operation, and it is an intermediate result in the process of generating the target fusion feature vector.

[0098] Step S338: Perform nonlinear transformation processing on the fusion transition vector to generate the target fusion feature vector.

[0099] Nonlinear transformation is a process of nonlinearly mapping the fused transition vector, aiming to increase the expressive power and complexity of the features. The target fused feature vector is the feature vector obtained after nonlinear transformation, integrating information from time-frequency distribution features and signal pattern features, and possessing stronger expressive power. Activation functions such as ReLU and Sigmoid can be used to perform nonlinear transformation on the fused transition vector. Taking ReLU as an example, ReLU transformation is applied to each element of the fused transition vector, setting elements less than 0 to 0 and leaving elements greater than 0 unchanged, thereby generating the target fused feature vector.

[0100] Step S340: Input the target fusion feature vector into the confidence calculation layer of the pre-trained weak target recognition model to generate the weak target confidence parameters.

[0101] The confidence calculation layer is the last network layer in the pre-trained weak target recognition model. Its function is to calculate the confidence level of a potential weak target's reflected signal as a weak target based on the input target fusion feature vector. The weak target confidence parameter is the output of the confidence calculation layer, representing the probability that the potential weak target's reflected signal is a weak target.

[0102] The target fusion feature vector is input into the confidence calculation layer of the pre-trained weak target recognition model. This can be achieved using neural network structures such as fully connected layers or softmax layers. Taking the softmax layer as an example, the target fusion feature vector is input into the softmax layer, and the output is converted into a probability distribution through the softmax function to obtain the weak target confidence parameters.

[0103] Step S400: Based on the comparison result between the weak target confidence parameter and the preset confidence threshold, generate the weak target detection result of the target monitoring area.

[0104] The preset confidence threshold is a pre-defined threshold used to determine whether a potential weak target's reflected signal is indeed a weak target. When the weak target confidence parameter is greater than the preset confidence threshold, the potential weak target's reflected signal is considered a weak target; when the weak target confidence parameter is less than the preset confidence threshold, the potential weak target's reflected signal is considered not a weak target. The weak target detection result is generated based on the comparison between the weak target confidence parameter and the preset confidence threshold, and includes the presence and related information of weak targets within the target monitoring area.

[0105] The generation of weak target detection results for a target monitoring area can be achieved by comparing the weak target confidence parameter with a preset confidence threshold using the following steps: First, compare the weak target confidence parameter with the preset confidence threshold. Then, determine whether the potential weak target reflection signal is indeed a weak target based on the comparison result. Finally, organize and record the existence of weak targets and related information to generate the weak target detection results.

[0106] As one implementation, step S400, which generates a weak target detection result for the target monitoring area based on the comparison result between the weak target confidence parameter and a preset confidence threshold, may specifically include the following steps S410~S480: Step S410: Perform multidimensional feature correlation analysis on the time-frequency distribution features and the signal mode features to generate a joint feature matrix.

[0107] Multidimensional feature correlation analysis is the process of analyzing the correlation between time-frequency distribution features and signal mode features across multiple dimensions, with the aim of uncovering potential correlation information between the two features. The joint feature matrix is ​​a matrix obtained after multidimensional feature correlation analysis, integrating information from both time-frequency distribution features and signal mode features.

[0108] Various methods can be used to perform multidimensional feature correlation analysis on time-frequency distribution features and signal mode features, such as correlation analysis and principal component analysis. Taking correlation analysis as an example, the correlation coefficients of time-frequency distribution features and signal mode features in different dimensions are calculated to obtain a correlation coefficient matrix. Based on the correlation coefficient matrix, the correlation relationship between the two features is determined, thereby generating a joint feature matrix.

[0109] Step S420: Perform confidence calibration based on the joint feature matrix and the historical environmental interference feature set to generate the corrected weak target confidence parameters.

[0110] The historical environmental interference feature set is a collection of environmental interference features gathered in the target monitoring area in the past, reflecting the environmental interference situation in the target monitoring area. Confidence calibration is the process of adjusting the weak target confidence parameter based on the historical environmental interference feature set, aiming to reduce the impact of environmental interference on the weak target detection results. The corrected weak target confidence parameter, obtained after confidence calibration, more accurately reflects the probability that the reflected signal from a potential weak target is indeed a weak target.

[0111] The confidence level calibration based on the joint feature matrix and the historical environmental interference feature set can be performed as follows: First, the joint feature matrix is ​​matched and compared with the historical environmental interference feature set to determine the degree of influence of the current environmental interference. Then, the confidence parameters of the weak target are adjusted according to the degree of influence to obtain the corrected confidence parameters of the weak target.

[0112] Step S430: Perform stratified verification processing on the corrected weak target confidence parameters and the preset confidence threshold to obtain primary detection results and secondary verification parameters.

[0113] Layered verification is a process of comparing and verifying the corrected weak target confidence parameters with a preset confidence threshold multiple times. Its purpose is to improve the accuracy and reliability of weak target detection results. The primary detection result is obtained from the first comparison, providing an initial assessment of whether the potential weak target's reflection signal is indeed a weak target. Secondary verification parameters are used to further verify the primary detection results, providing more information and evidence.

[0114] The stratified validation process, which compares the corrected weak target confidence parameters with a preset confidence threshold, can be performed using the following steps: First, a preliminary comparison is made between the corrected weak target confidence parameters and the preset confidence threshold to obtain a primary detection result. Then, based on the primary detection result and the specific characteristics of the corrected weak target confidence parameters, secondary validation parameters are determined. For example, a validation interval can be defined based on the magnitude and fluctuation of the confidence parameters, and the secondary validation parameters can be set as the boundary values ​​of this validation interval.

[0115] Step S440: Perform multi-source signal backtracking processing on the primary detection results to extract the feature trajectory of the potential weak target reflection signal that matches the secondary verification parameters.

[0116] Multi-source signal backtracking processing is a process of backtracking and analyzing primary detection results. Its purpose is to extract feature trajectories of potential weak target reflection signals that match secondary verification parameters from multiple data sources. The feature trajectory describes the temporal and spatial variations of the potential weak target reflection signal and can provide more information about the weak target. Multi-source signal backtracking processing of primary detection results can be performed as follows: First, identify multiple data sources related to the primary detection results, such as sensors at different locations and monitoring data from different time periods. Then, extract feature information of potential weak target reflection signals that match secondary verification parameters from these data sources. Finally, integrate and analyze this feature information to obtain the feature trajectory of the potential weak target reflection signal.

[0117] Step S450: Based on the feature trajectory, perform spatial coverage analysis on the primary detection results to generate a confidence distribution map of the target area.

[0118] Spatial coverage analysis is the process of analyzing the spatial coverage of the characteristic trajectories of reflected signals from potential weak targets. Its purpose is to determine the possible spatial range of weak targets within the target monitoring area. The target area confidence distribution map is a map generated based on the spatial coverage analysis results, representing the confidence distribution of weak targets at different locations within the target monitoring area. Spatial coverage analysis based on characteristic trajectories of primary detection results can be performed using the following method: First, determine the possible spatial locations and ranges of reflected signals from potential weak targets based on the characteristic trajectories. Then, divide the target monitoring area into multiple smaller areas and calculate the confidence of weak targets within each smaller area. Finally, visualize these confidence values ​​to generate the target area confidence distribution map.

[0119] Step S460: Determine the weak target presence marker based on the peak region in the target area confidence distribution map, and generate spatial location estimation parameters by combining the time-frequency offset of the potential weak target reflection signal.

[0120] The peak region is an area with a high confidence value in the target area confidence distribution map, indicating a higher probability of the presence of a weak target within that area. The weak target presence indicator is used to indicate whether a weak target exists within the target monitoring area and can be determined based on the peak region. Spatial location estimation parameters are used to estimate the spatial location of the weak target, combining information from the time-frequency offset of the potential weak target's reflected signal and the target area confidence distribution map. Determining the weak target presence indicator based on the peak region in the target area confidence distribution map can be achieved using the following method: A confidence threshold is set; if the confidence value of the peak region is greater than this threshold, a weak target is considered to exist within the area, and the weak target presence indicator is set to "exist"; otherwise, the weak target presence indicator is set to "not exist". Generating spatial location estimation parameters by combining the time-frequency offset of the potential weak target's reflected signal can be achieved using the following steps: First, determine the weak target's position in time and frequency based on the time-frequency offset. Then, combine this with the peak region in the target area confidence distribution map to determine the possible spatial location of the weak target. Finally, integrate the time, frequency, and spatial location information to generate the spatial location estimation parameters.

[0121] Step S470: Perform multi-dimensional interference compensation processing on the spatial location estimation parameters to obtain optimized spatial location parameters; wherein, the multi-dimensional interference compensation processing includes: acquiring the environmental reflection characteristic dataset of the target monitoring area, extracting real-time reflection interference features corresponding to the spatial location estimation parameters, constructing an interference compensation model based on the environmental reflection characteristic dataset, suppressing the real-time reflection interference features through the interference compensation model, generating compensated coordinate offsets, and performing reverse correction on the spatial location estimation parameters based on the coordinate offsets. Multi-dimensional interference compensation processing is a process of compensating for multi-dimensional interference in the spatial location estimation parameters. Its purpose is to reduce the impact of environmental interference on spatial location estimation and improve the accuracy of spatial location estimation. The environmental reflection characteristic dataset is a dataset recording the environmental reflection characteristics within the target monitoring area, which can reflect the impact of environmental reflection on the signal. The real-time reflection interference features are the reflection interference features at the current moment corresponding to the spatial location estimation parameters, which can reflect the impact of current environmental reflection on the spatial location estimation of weak targets. The interference compensation model is a model constructed based on the environmental reflection characteristic dataset, which can be used to suppress and compensate for real-time reflection interference features. The compensated coordinate offset is the coordinate offset obtained after processing by the interference compensation model, which represents the degree of influence of environmental reflection interference on the spatial location estimation parameters.

[0122] Multi-dimensional interference compensation processing for spatial location estimation parameters can be performed using the following steps: First, obtain a dataset of environmental reflectance characteristics of the target monitoring area. Then, extract real-time reflectance interference features corresponding to the spatial location estimation parameters from this dataset. Next, construct an interference compensation model based on the environmental reflectance characteristic dataset, such as using a linear regression model or a neural network model. Suppress the real-time reflectance interference features using the interference compensation model to generate compensated coordinate offsets. Finally, perform a reverse correction on the spatial location estimation parameters based on the compensated coordinate offsets to obtain the optimized spatial location parameters.

[0123] As one implementation method, the spatial position parameters are obtained using the following steps S471~S475: Step S471: Perform multi-channel phase difference measurement processing on the potential weak target reflection signal to obtain the phase difference parameters between each receiving channel.

[0124] Multichannel phase difference measurement processing is the process of measuring the phase difference of the reflected signal from a potential weak target across multiple receiving channels. Its purpose is to obtain phase difference information between each receiving channel, which is crucial for subsequently determining the spatial location of the weak target. The phase difference parameter describes the phase difference between the signals in each receiving channel; different phase differences reflect different signal propagation paths and are therefore correlated with the spatial location of the weak target.

[0125] Multi-channel phase difference measurement of potential weak target reflection signals can be achieved in the following ways. If multiple receiving antennas serve as receiving channels, the potential weak target reflection signals received by each antenna will have phase differences due to different propagation paths. First, the signals received by each receiving channel are preprocessed, such as by filtering to remove noise interference, and then converted into digital signals for subsequent processing. Next, a reference channel is selected, and the signals from other channels are compared with the signal from the reference channel in terms of phase. A correlation algorithm can be used to calculate the correlation between the two signals, and the phase difference is determined by the peak position of the correlation. For example, a cross-correlation function can be used to calculate the correlation between signals from different channels. Let the reference channel signal be x(n), and the other channel signals be y(n). The cross-correlation function... Where N is the signal length and m is the delay time. The phase difference between two signals can be obtained by finding the peak position of the cross-correlation function. By performing this process on all receiving channels, the phase difference parameters between each receiving channel can be obtained.

[0126] Step S472: Generate a multidimensional feature vector based on the phase difference parameter and the time-frequency change rate of the potential weak target reflection signal.

[0127] A multidimensional feature vector is a vector containing information from multiple dimensions. It integrates information from phase difference parameters and the time-frequency change rate of the reflected signal from a potential weak target, providing a more comprehensive description of the characteristics of the potential weak target and offering richer evidence for subsequent localization optimization. The time-frequency change rate is the rate of change of the reflected signal from a potential weak target in time and frequency, reflecting the dynamic characteristics of the signal. Generating a multidimensional feature vector based on the phase difference parameters and the time-frequency change rate of the reflected signal from a potential weak target can be achieved using the following method: First, arrange the phase difference parameters into a vector according to the order of the receiving channels; this vector reflects the phase relationship between the channels. Then, calculate the time-frequency change rate of the reflected signal from the potential weak target, for example, by differentiating the instantaneous frequency and instantaneous amplitude of the signal. Include relevant parameters of the time-frequency change rate (such as the instantaneous frequency change rate and instantaneous amplitude change rate) as elements of the vector. Finally, concatenate the phase difference vector and the relevant vector of the time-frequency change rate to form a multidimensional feature vector.

[0128] Step S473: Input the multidimensional feature vector into the pre-trained localization optimization model for multidimensional parameter fusion processing to generate an optimized feature vector.

[0129] The pre-trained localization optimization model is a model trained on a large amount of sample data. It can analyze and process the input multi-dimensional feature vector, fuse parameters from multiple dimensions, remove redundant information, and extract more representative and discriminative features to generate an optimized feature vector. The optimized feature vector is the feature vector obtained after processing by the localization optimization model, which is more conducive to accurately estimating the spatial location of weak targets.

[0130] Deep learning models, such as Convolutional Neural Networks (CNNs) or Recurrent Neural Networks (RNNs), can be used to input multidimensional feature vectors into a pre-trained localization optimization model for multidimensional parameter fusion. Taking CNNs as an example, a localization optimization model can consist of multiple convolutional layers, pooling layers, and fully connected layers. First, the multidimensional feature vectors are input into the convolutional layers, which extract local features by performing convolution operations with the input vectors using convolution kernels. The pooling layers downsample the output of the convolutional layers, reducing the dimensionality of the data while retaining important feature information. After processing by multiple convolutional and pooling layers, the features are further abstracted and integrated. Finally, the fully connected layers fuse the features extracted by the previous layers, outputting an optimized feature vector. During training, a large amount of labeled sample data is used, and the model parameters are continuously adjusted through the backpropagation algorithm, enabling the model to accurately convert multidimensional feature vectors into optimized feature vectors.

[0131] Step S474: Perform three-dimensional spatial mapping processing based on the azimuth and distance components in the optimized feature vector to obtain the initial spatial position parameters.

[0132] Azimuth and range components are crucial information in optimizing feature vectors, describing the position of weak targets in 3D space. Azimuth represents the angle of the weak target relative to a reference direction, and range represents the distance between the weak target and the receiving device. 3D spatial mapping converts the azimuth and range components into coordinates in 3D space. This process yields the initial position information of the weak target in 3D space, i.e., the initial spatial position parameters.

[0133] Three-dimensional spatial mapping based on the azimuth and range components in the optimized feature vector can be achieved using a transformation from spherical coordinates to rectangular coordinates. If the azimuth is... Angle of elevation is (If considering three-dimensional space, in addition to the horizontal azimuth, elevation information is also needed. The elevation angle can be extracted from the optimized feature vector or obtained through other methods.) The distance is r. In a Cartesian coordinate system, the coordinates (x, y, z) of a weak target can be calculated using the following formula: Through this transformation, the azimuth and range components in the optimized feature vector can be converted into coordinates in three-dimensional space, thus obtaining the initial spatial position parameters.

[0134] Step S475: Perform environmental interference compensation processing on the initial spatial position parameters to generate the spatial position parameters; wherein, the environmental interference compensation processing includes: obtaining a historical environmental interference feature set of the target monitoring area, extracting real-time environmental interference features corresponding to the initial spatial position parameters, performing weight matching on the real-time environmental interference features based on the historical environmental interference feature set, generating interference compensation coefficients, and performing reverse correction on the coordinate offset of the initial spatial position parameters according to the interference compensation coefficients.

[0135] Environmental interference compensation processing aims to reduce the impact of environmental interference on initial spatial location parameters and improve the accuracy of spatial location estimation. The historical environmental interference feature set is a collection of environmental interference features gathered in the target monitoring area over the past, encompassing environmental interference information from different locations and times. Real-time environmental interference features are the environmental interference characteristics corresponding to the initial spatial location parameters at the current moment, reflecting the actual impact of the current environment on the spatial location estimation of weak targets. The interference compensation coefficient is a coefficient generated based on the matching results of the historical environmental interference feature set and the real-time environmental interference features, used to measure the degree of influence of environmental interference on the initial spatial location parameters.

[0136] Environmental interference compensation processing for initial spatial location parameters can be performed according to the following steps. First, obtain a historical set of environmental interference features for the target monitoring area. This set can be stored in a database and contains interference feature information at different locations, such as signal attenuation and multipath effects. Then, determine the corresponding location based on the initial spatial location parameters and extract the interference features near that location from the historical set of environmental interference features as a reference. Simultaneously, obtain the real-time environmental interference features corresponding to the initial spatial location parameters through real-time monitoring. Next, perform weight matching on the real-time environmental interference features based on the historical set of features. Similarity calculation methods can be used, such as calculating the Euclidean distance or cosine similarity between the real-time environmental interference features and each feature in the historical set of features. Assign a weight to each historical interference feature based on the similarity, and then normalize these weights to obtain the interference compensation coefficient. Finally, perform a reverse correction on the coordinate offset of the initial spatial location parameters based on the interference compensation coefficient. If the initial spatial location parameters are... The coordinate offset is If the interference compensation coefficient is k, then the corrected spatial location parameters are: .

[0137] Step S480: Associate and encapsulate the weak target presence identifier, the optimized spatial location parameters, and the target area confidence distribution map to generate the weak target detection result.

[0138] Association encapsulation is the process of integrating and packaging three crucial pieces of information: weak target presence identifiers, optimized spatial location parameters, and target area confidence distribution maps. Its purpose is to combine these related information into a complete, easily transmitted, and processed result—the weak target detection result. The weak target presence identifier clearly indicates the presence of a weak target within the target monitoring area; the optimized spatial location parameters accurately determine the weak target's location in three-dimensional space; and the target area confidence distribution map shows the confidence distribution of weak targets at different locations within the target monitoring area. Association encapsulating these elements provides comprehensive information for subsequent decision-making and processing. This association encapsulation of weak target presence identifiers, optimized spatial location parameters, and target area confidence distribution maps can be achieved using data structures. For example, a structure or class can be created, with the weak target presence identifier, optimized spatial location parameters, and target area confidence distribution map as member variables.

[0139] Step S500: Transmit the weak target detection result to the target terminal device to trigger an alarm response operation.

[0140] The target terminal device is the device that receives the weak target detection results. It can be a computer in the monitoring center, a mobile terminal, etc. It is used to receive and process the weak target detection results and trigger corresponding alarm response operations based on the results. The alarm response operation is a series of measures taken based on the weak target detection results, such as issuing an alarm sound, displaying alarm information, and activating emergency plans, to remind relevant personnel to pay attention to the possible presence of weak targets in the target monitoring area.

[0141] Transmitting weak target detection results to the target terminal device can employ various communication methods, such as wired communication (e.g., Ethernet, fiber optics) or wireless communication (e.g., Wi-Fi, Bluetooth, 4G / 5G). The specific communication method needs to be selected based on the type of target terminal device and the actual application scenario. For example, if the target terminal device is a computer in a monitoring center and the devices are located within the same local area network, Ethernet can be used for data transmission; if the target terminal device is a mobile terminal, 4G / 5G wireless communication can be used for data transmission. During transmission, the weak target detection results need to be encoded and encapsulated for transmission over the communication network. For example, the weak target detection results can be encoded using JSON format, converted into a string, and then sent to the target terminal device over the network. After receiving the data, the target terminal device decodes and parses it, extracts the relevant information from the weak target detection results, and then triggers an alarm response based on this information.

[0142] As one implementation, step S500, which involves transmitting the weak target detection result to the target terminal device to trigger an alarm response, may specifically include the following steps S510-S550: Step S510: Perform multi-dimensional information association processing on the weak target detection results to generate composite alarm data containing spatial distribution features, confidence change features, and environmental interference association features.

[0143] Multi-dimensional information association processing is the process of associating and integrating multiple dimensions of information from weak target detection results. Its purpose is to fuse different types of information to generate more comprehensive and valuable composite alarm data. Spatial distribution characteristics describe the spatial location and distribution of weak targets within the target monitoring area, which can be reflected through optimized spatial location parameters and a target area confidence distribution map. Confidence change characteristics reflect the change in the confidence level of weak targets over time, which can be obtained by analyzing the confidence parameters in multiple detection results. Environmental interference association characteristics represent the correlation between weak target detection results and environmental interference, such as the degree of influence of environmental interference on weak target detection.

[0144] The following methods can be used to perform multi-dimensional information association processing on weak target detection results. First, optimize spatial location parameters and target area confidence distribution maps are extracted from the weak target detection results to construct spatial distribution features. The optimized spatial location parameters can be converted into visual coordinate points and marked on a map. Combined with the target area confidence distribution map, the spatial distribution and confidence levels of weak targets are displayed. Then, time series analysis is performed on the weak target confidence parameters from multiple detection results to calculate statistics such as the rate of change and fluctuation range of confidence, generating confidence change features. Finally, the impact of environmental interference on weak target detection is analyzed. For example, by comparing detection results under different environmental conditions, the correlation between environmental interference and weak target detection results is determined, generating environmental interference association features. The spatial distribution features, confidence change features, and environmental interference association features are integrated to generate composite alarm data.

[0145] Step S520: Select the data transmission path and determine the data compression level according to the communication protocol type and real-time network latency parameters of the target terminal device.

[0146] The communication protocol type refers to the communication protocol supported by the target terminal device. Different communication protocols have different characteristics and applicable scenarios. For example, TCP is characterized by reliable transmission, while UDP is characterized by low latency. Real-time network latency parameters are the data transmission delay time under the current network environment, reflecting the real-time performance of the network. The data transmission path is the transmission path of the composite alarm data from the sending end to the target terminal device. Different transmission paths may have different performance indicators such as bandwidth and latency. The data compression level is the degree to which the composite alarm data is compressed. Different compression levels affect the data compression ratio and the transmission efficiency after compression.

[0147] The following strategies can be used to select the data transmission path and determine the data compression level based on the target terminal device's communication protocol type and real-time network latency parameters. First, understand the communication protocol types supported by the target terminal device and select a suitable transmission path based on the characteristics of the communication protocol and real-time network latency parameters. If the real-time network latency is high and the target terminal device supports the TCP protocol, a transmission path with higher reliability can be selected to ensure accurate data transmission. If the real-time network latency is low and the real-time requirements for data transmission are high, the UDP protocol can be selected, along with a low-latency transmission path. Then, determine the data compression level based on the transmission path's bandwidth and real-time network latency parameters. If the transmission path's bandwidth is small, or the real-time network latency is high, a higher data compression level can be selected to reduce the amount of data transmitted and improve transmission efficiency. If the transmission path's bandwidth is large and the real-time network latency is low, a lower data compression level can be selected to reduce the time overhead of data compression and decompression.

[0148] Step S530: Perform layered compression processing on the composite alarm data based on the data compression level to generate a compressed data stream.

[0149] Layered compression is a process of progressively compressing composite alarm data according to different compression levels. Its purpose is to minimize the amount of data transmitted while ensuring data recoverability. Different compression levels correspond to different compression algorithms and parameters, allowing for flexible adjustment of the compression degree based on actual conditions. The compressed data stream, obtained after layered compression, can be transmitted more efficiently in communication networks.

[0150] Layered compression of composite alarm data based on data compression levels can employ various compression algorithms, such as the LZ77 algorithm, Huffman coding algorithm, and Deflate algorithm. Taking the Deflate algorithm as an example, it combines the advantages of the LZ77 algorithm and the Huffman coding algorithm, achieving high compression efficiency. In layered compression processing, different compression parameters, such as the compression window size and Huffman coding table, are determined according to the data compression level. First, the composite alarm data is preprocessed, such as converting the data to binary format. Then, appropriate compression parameters are selected based on the data compression level to compress the data. For example, at lower data compression levels, a smaller compression window and Huffman coding table can be used; at higher data compression levels, a larger compression window and a more complex Huffman coding table can be used. After multiple compression processes, a compressed data stream is generated.

[0151] Step S540: Send the compressed data stream to the target terminal device through the data transmission path, triggering the target terminal device to perform the following operations: parse the compressed data stream and restore the composite alarm data, generate a dynamic heat map based on the spatial distribution characteristics, generate a time axis overlay curve based on the confidence change characteristics, and generate interference suppression suggestion parameters in combination with the environmental interference correlation characteristics.

[0152] The data transmission path is the path selected in step S520 for data transmission, which can be a wired network or a wireless network. The compressed data stream is sent to the target terminal device through the data transmission path, and network communication technology can be used to transmit data from the sender to the receiver. After receiving the compressed data stream, the target terminal device needs to perform parsing and restoration operations to recover the original composite alarm data. Parsing the compressed data stream can employ a decompression algorithm corresponding to the compression algorithm; for example, for data compressed using the Deflate algorithm, a corresponding decompression library can be used for decompression.

[0153] Generating dynamic heatmaps based on spatial distribution characteristics visualizes the spatial distribution of weak targets within the target monitoring area. Dynamic heatmaps intuitively display the distribution density and confidence level of weak targets, using different colors and brightness levels to represent different confidence values. Visualization tools (such as Python's Matplotlib library and JavaScript's D3.js library) can be used to generate dynamic heatmaps. Generating time-axis overlay curves based on confidence level change characteristics displays the change in weak target confidence over time as a curve. Time-axis overlay curves help analysts observe the trend of weak target confidence changes and determine the stability and variation patterns of weak targets. Time series analysis and visualization techniques can be used to generate time-axis overlay curves. Generating interference suppression suggestion parameters by combining environmental interference correlation characteristics proposes corresponding interference suppression suggestions based on the correlation between environmental interference and weak target detection results. For example, if a certain environmental interference is found to have a significant impact on weak target detection, corresponding measures can be suggested to suppress the interference, such as adjusting the position of monitoring equipment or using filtering techniques.

[0154] Step S550: Integrate the dynamic heat map, time axis overlay curve, and interference suppression suggestion parameters into the interactive alarm interface for visualization.

[0155] The interactive alarm interface is used to display alarm information and provides interactive functions, allowing users to interact with the alarm information, such as zooming in and out of the map and viewing detailed data. By integrating dynamic heatmaps, timeline overlay curves, and interference suppression suggestion parameters into the interactive alarm interface for visualization, different types of alarm information can be combined to provide users with a comprehensive and intuitive display of alarm information.

[0156] Integrating dynamic heatmaps, overlay timeline curves, and interference suppression recommendations into an interactive alarm interface can be achieved using front-end development technologies such as HTML, CSS, and JavaScript. First, HTML and CSS are used to construct the layout and style of the alarm interface, designing appropriate containers to display the dynamic heatmaps, overlay timeline curves, and interference suppression recommendations. Then, JavaScript is used to write the interactive logic, enabling user interaction with the alarm information. For example, when a user clicks on a region on the dynamic heatmap, detailed information about that region is displayed; when a user drags a time point on the overlay timeline curve, the confidence value for that time point is displayed, and so on. Through such integration and visualization, users can more easily obtain and analyze alarm information and take timely appropriate measures.

[0157] It is understood that the various algorithms involved in the above descriptions of the embodiments of the present invention, such as noise reduction algorithms, LZ77 algorithms, Huffman coding algorithms, Deflate algorithms, etc., can all be obtained from relevant content in the prior art. To save space, they will not be elaborated on in the embodiments of the present invention. In addition, those skilled in the art can supplement the details based on common knowledge in the art when implementing the solution of the present invention. For example, they can use normalization to eliminate dimensional conflicts before feature fusion, use interpolation to eliminate dimensional differences, reasonably set the threshold based on historical data, experience or business scenario requirements, train the model based on a general model training method, etc. The present invention will not provide redundant descriptions of overly detailed implementation processes here.

[0158] Figure 2 A hardware entity diagram of a computer system provided as an embodiment of the present invention, such as... Figure 2 As shown, the hardware entity of the computer system 1000 includes a processor 1001 and a memory 1002, wherein the memory 1002 stores a computer program that can run on the processor 1001, and the processor 1001 executes the program to implement the steps in the method of any of the above embodiments.

[0159] The memory 1002 stores computer programs that can run on the processor. The memory 1002 is configured to store instructions and applications that can be executed by the processor 1001. It can also cache data to be processed or already processed (e.g., image data, audio data, voice communication data, and video communication data) of the processor 1001 and various modules in the computer system 1000. It can be implemented by flash memory or random access memory (RAM).

[0160] When the processor 1001 executes the program, it implements the steps of the passive sensing radar AI weak target detection method described above. The processor 1001 typically controls the overall operation of the computer system 1000.

[0161] This invention provides a computer storage medium storing one or more programs that can be executed by one or more processors to implement the steps of the passive sensing radar AI weak target detection method as described in any of the above embodiments.

[0162] It should be noted that the descriptions of the above storage medium and device embodiments are similar to those of the above method embodiments, and have similar beneficial effects. For technical details not disclosed in the storage medium and device embodiments of the present invention, please refer to the descriptions of the method embodiments of the present invention for understanding. The processor described above can be at least one of the following: Application-Specific Integrated Circuit (ASIC), Digital Signal Processor (DSP), Digital Signal Processing Device (DSPD), Programmable Logic Device (PLD), Field Programmable Gate Array (FPGA), Central Processing Unit (CPU), Controller, Microcontroller, and Microprocessor. It is understood that the electronic device implementing the above processor function can also be other types, and the embodiments of the present invention do not specifically limit it.

[0163] The aforementioned computer storage media / memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), magnetic random access memory (FRAM), flash memory, magnetic surface memory, optical disc, or compact disc read-only memory (CD-ROM), etc.; or it can be various terminals that include one or any combination of the above-mentioned memories, such as mobile phones, computers, tablet devices, personal digital assistants, etc.

[0164] It should be understood that the phrase "one embodiment" or "an embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of the invention. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. It should be understood that in the various embodiments of the invention, the sequence number of the above steps / processes does not imply the order of execution; the order of execution of each step / process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the invention. The sequence numbers of the above embodiments of the invention are merely descriptive and do not represent the superiority or inferiority of the embodiments. It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0165] In the several embodiments provided by this invention, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.

[0166] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units. They may be located in one place or distributed across multiple network units. Some or all of the units may be selected to achieve the purpose of this embodiment according to actual needs.

[0167] In addition, in the various embodiments of the present invention, each functional unit can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the integrated unit can be implemented in hardware or in the form of hardware plus software functional units.

[0168] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media that can store program code, such as mobile storage devices, read-only memory (ROM), magnetic disks, or optical disks.

[0169] Alternatively, if the integrated units of this invention are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to related technologies, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROM, magnetic disks, or optical disks.

[0170] The above description is merely an embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A passive sensing radar AI weak target detection method, characterized in that, include: Acquire a multi-dimensional electromagnetic signal data set of the target monitoring area, wherein the multi-dimensional electromagnetic signal data set includes environmental clutter signals and potential weak target reflection signals; The multi-dimensional electromagnetic signal data set is subjected to signal feature extraction processing to obtain the time-frequency distribution features and signal mode features corresponding to the multi-dimensional electromagnetic signal data set; Based on a pre-trained weak target identification model, the time-frequency distribution features and the signal pattern features are jointly analyzed and processed to generate the weak target confidence parameters of the potential weak target reflection signal. Based on the comparison between the weak target confidence parameter and the preset confidence threshold, the weak target detection result of the target monitoring area is generated; The weak target detection results are transmitted to the target terminal device to trigger an alarm response operation.

2. The method according to claim 1, characterized in that, The multi-dimensional electromagnetic signal data set is subjected to signal feature extraction processing to obtain the time-frequency distribution features and signal mode features corresponding to the multi-dimensional electromagnetic signal data set, including: The environmental clutter signals in the multi-dimensional electromagnetic signal data set are subjected to signal denoising processing to obtain a denoised environmental clutter signal set. The noise-reduced environmental clutter signal set is subjected to time-frequency analysis processing to generate the time-frequency energy distribution characteristics of the environmental clutter signal set; The potential weak target reflection signal is processed by signal pattern recognition to obtain the periodicity and modulation characteristics of the potential weak target reflection signal; The time-frequency energy distribution feature is fused with the periodic feature and the modulation feature to generate the signal pattern feature; wherein, the time-frequency distribution feature includes the correlation feature between the time-frequency energy distribution feature and the time-frequency offset of the potential weak target reflection signal.

3. The method according to claim 2, characterized in that, The potential weak target reflection signal is subjected to signal pattern recognition processing to obtain the periodicity and modulation characteristics of the potential weak target reflection signal, including: The potential weak target reflection signal is processed by signal segmentation to obtain multiple signal segment sequences; Autocorrelation function is calculated for each of the signal segment sequences to generate periodic characteristic parameters corresponding to each of the signal segment sequences; The periodic characteristics are obtained by performing statistical analysis on the periodicity parameters. Instantaneous frequency estimation processing is performed on the reflection signal of the potential weak target to generate the frequency change trajectory of the reflection signal of the potential weak target; The modulation characteristics are obtained by performing slope analysis on the frequency change trajectory. The periodicity feature is used to characterize the repetitive regularity of the reflected signal from the potential weak target, and the modulation feature is used to characterize the frequency modulation characteristics of the reflected signal from the potential weak target.

4. The method according to claim 1, characterized in that, The pre-trained weak target recognition model is obtained through the following steps: Acquire a sample electromagnetic signal dataset of the historical monitoring area, the sample electromagnetic signal dataset containing environmental clutter signal samples and weak target reflection signal samples labeled with the presence status of weak targets; The sample electromagnetic signal dataset is subjected to signal enhancement processing to obtain an enhanced sample electromagnetic signal dataset; The enhanced sample electromagnetic signal dataset is subjected to feature extraction processing to obtain the sample time-frequency distribution features and sample signal mode features corresponding to each sample signal; An initial weak target recognition model is constructed, which includes a feature fusion layer and a classification decision layer; Based on the time-frequency distribution features of the samples and the signal pattern features of the samples, the initial weak target recognition model is iteratively trained until the error between the output of the classification decision layer and the labeling result of the weak target existence state is less than a preset threshold, thus obtaining the pre-trained weak target recognition model.

5. The method according to claim 4, characterized in that, The feature extraction process for the enhanced sample electromagnetic signal dataset includes: The environmental clutter signal sample is subjected to multi-scale wavelet transform processing to generate multi-resolution energy distribution characteristics of the environmental clutter signal sample. Cyclic stationary analysis is performed on the weak target reflection signal samples to generate the cyclic frequency characteristics of the weak target reflection signal samples; The multi-resolution energy distribution features and the cyclic frequency features are aligned to generate the sample time-frequency distribution features. The high-order cumulant calculation process is performed on the weak target reflection signal sample to obtain the high-order statistical characteristics of the weak target reflection signal sample; The higher-order statistical features are subjected to dimensionality compression to generate the sample signal pattern features; The feature alignment process is used to eliminate the time-frequency offset difference between the environmental clutter signal sample and the weak target reflection signal sample.

6. The method according to claim 1, characterized in that, Based on a pre-trained weak target identification model, the time-frequency distribution features and the signal pattern features are jointly analyzed and processed to generate weak target confidence parameters for the reflection signals of potential weak targets, including: The time-frequency distribution features are input into the feature encoding layer of the pre-trained weak target recognition model to obtain the first intermediate feature vector; The signal pattern features are input into the feature transformation layer of the pre-trained weak target recognition model to obtain the second intermediate feature vector; The first intermediate feature vector and the second intermediate feature vector are weighted and fused to obtain the target fused feature vector. The target fusion feature vector is input into the confidence calculation layer of the pre-trained weak target recognition model to generate the weak target confidence parameters; The weighting coefficients of the weighted fusion process are adjusted according to the correlation between the time-frequency distribution features and the signal pattern features.

7. The method according to claim 6, characterized in that, The first intermediate feature vector and the second intermediate feature vector are weighted and fused to obtain the target fused feature vector, including: Perform time dimension distribution analysis on the first intermediate feature vector to generate the time-series correlation weights corresponding to the time-frequency distribution features; Perform frequency dimension distribution analysis on the second intermediate feature vector to generate frequency domain correlation weights corresponding to the signal mode features; An intermediate weight matrix is ​​constructed based on the time-series correlation weight and the frequency-domain correlation weight, and environmental interference features of the potential weak target reflection signal are extracted. Based on the environmental interference characteristics, the intermediate weight matrix is ​​subjected to interference suppression processing to generate an optimized weight matrix; Based on the optimized weight matrix, cross-projection calculation is performed on the first intermediate feature vector and the second intermediate feature vector to obtain the first projected feature vector and the second projected feature vector. Multi-level feature interaction processing is performed on the first projection feature vector and the second projection feature vector to generate an interactive feature vector. The interaction feature vector is multiplied by the optimized weight matrix to generate a fusion transition vector. The fusion transition vector is subjected to nonlinear transformation to generate the target fusion feature vector; The multi-level feature interaction processing includes: splitting the first projected feature vector into multiple sub-vector segments, performing segment-by-segment convolution processing on each sub-vector segment and the second projected feature vector to generate a set of convolutional feature segments, performing cross-segment attention allocation processing on each convolutional feature segment, and generating the interaction feature vector.

8. The method according to claim 1, characterized in that, Based on the comparison between the weak target confidence parameter and the preset confidence threshold, a weak target detection result for the target monitoring area is generated, including: Multidimensional feature correlation analysis is performed on the time-frequency distribution features and the signal mode features to generate a joint feature matrix; Based on the joint feature matrix and the set of historical environmental interference features, a confidence calibration process is performed to generate the corrected weak target confidence parameters. The corrected weak target confidence parameters are subjected to hierarchical verification with the preset confidence threshold to obtain primary detection results and secondary verification parameters. The primary detection results are subjected to multi-source signal backtracking processing to extract the feature trajectory of the potential weak target reflection signal that matches the secondary verification parameters; Based on the feature trajectory, spatial coverage analysis is performed on the primary detection results to generate a confidence distribution map of the target area. Weak target presence markers are determined based on the peak regions in the target area confidence distribution map, and spatial location estimation parameters are generated by combining the time-frequency offset of the potential weak target reflection signal. The spatial location estimation parameters are subjected to multi-dimensional interference compensation processing to obtain optimized spatial location parameters; The weak target presence identifier, the optimized spatial location parameters, and the target region confidence distribution map are associated and encapsulated to generate the weak target detection result. The multi-dimensional interference compensation process includes: acquiring an environmental reflection characteristic dataset of the target monitoring area, extracting real-time reflection interference features corresponding to the spatial location estimation parameters, constructing an interference compensation model based on the environmental reflection characteristic dataset, suppressing the real-time reflection interference features through the interference compensation model, generating a compensated coordinate offset, and performing reverse correction on the spatial location estimation parameters based on the coordinate offset.

9. The method according to claim 8, characterized in that, The spatial location parameters are determined using the following steps: The potential weak target reflection signal is processed by multi-channel phase difference measurement to obtain the phase difference parameters between each receiving channel; A multidimensional feature vector is generated based on the phase difference parameter and the time-frequency change rate of the potential weak target reflection signal; The multidimensional feature vector is input into a pre-trained localization optimization model for multidimensional parameter fusion processing to generate an optimized feature vector; The initial spatial position parameters are obtained by performing three-dimensional spatial mapping processing based on the azimuth and distance components in the optimized feature vector. The initial spatial location parameters are subjected to environmental interference compensation processing to generate the spatial location parameters; wherein, the environmental interference compensation processing includes: obtaining a historical environmental interference feature set of the target monitoring area, extracting real-time environmental interference features corresponding to the initial spatial location parameters, performing weight matching on the real-time environmental interference features based on the historical environmental interference feature set to generate interference compensation coefficients, and performing reverse correction on the coordinate offset of the initial spatial location parameters according to the interference compensation coefficients.

10. A computer system comprising a memory and a processor, the memory storing a computer program executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the method according to any one of claims 1 to 9.

Images