Coal mine tunnel complex environment millimeter wave radar multi-domain interference identification method and device

Abstract

This application provides a method and apparatus for multi-domain interference identification using millimeter-wave radar in complex environments such as coal mine roadways. It relates to the field of intelligent detection and anti-interference technology, and includes: acquiring millimeter-wave radar echo signals and establishing target, interference, and noise component models; extracting features in the power spectrum, Doppler-time, range-time, and range-Doppler domains to capture interference spectrum, dust scattering, and multipath reflection features; uniformly projecting features from different domains into a representation space to ensure comparability and fusion; constructing a multi-domain fusion identification framework for target feature extraction and interference component separation; and introducing velocity consistency, spatial redundancy suppression, and temporal smoothing constraints to ensure detection stability. This invention enables robust detection and stable identification of vehicle-mounted millimeter-wave radar in coal mine roadways under interference environments such as dust, multipath interference, and electromagnetic radiation, reducing false alarm and false detection rates and improving application performance in coal mine tunneling scenarios.

Description

Technical Field

[0001] This application relates to the field of detection technology in complex environments of coal mine roadways, and in particular to a method and device for identifying multi-domain interference of millimeter-wave radar in complex environments of coal mine roadways. Background Technology

[0002] With increasing coal mining depth and the growing complexity of mining environments, the working conditions in coal mine roadways are becoming increasingly complex. Traditional coal mine safety monitoring technologies, such as optical imaging, infrared imaging, and laser scanning, often suffer from limitations when faced with factors like coal dust, gases, moisture, and metal reflections. This is especially true in harsh environments such as low light, high dust levels, and high humidity, where traditional sensors and monitoring technologies frequently suffer from signal attenuation, information loss, and significant errors. In contrast, millimeter-wave radar demonstrates unique advantages in coal mine roadways, capable of penetrating obstacles such as coal dust, moisture, and smoke, and unaffected by lighting conditions, making it suitable for target detection in complex coal mine environments. However, effectively identifying and separating various interference sources to improve the radar system's target detection capabilities and robustness remains a major challenge for this technology.

[0003] While existing technologies have made some progress in the application of millimeter-wave radar, many shortcomings remain. For example, a traditional technology proposes using multiple cascaded millimeter-wave radar chips to effectively improve target detection accuracy in complex environments like mines, particularly in driver assistance. It identifies target distance and velocity information by combining multidimensional FFT processing of echo signals with the CFAR algorithm, providing crucial data support for underground vehicles. However, this technology primarily focuses on the safety and controllability of underground vehicles, lacking in-depth analysis and processing of signal attenuation and distortion caused by various interference sources (such as coal dust and gases) in coal mine roadways. Systematic research on interference suppression in complex environments and effective identification of multiple interference sources is still lacking.

[0004] Another traditional technique involves collecting coal mine environmental data and analyzing point cloud information to accurately identify mine boundaries, including frontal, lateral, and overhead boundaries, providing data support for safe mine production. While this technique has certain advantages in boundary identification in coal mine environments, it still has shortcomings. This method primarily relies on radar to detect the position and height of objects to achieve boundary detection. However, it does not effectively address interference sources in the complex environment of coal mines (such as coal dust, metal reflection, and multipath propagation). The impact of signal attenuation and multipath effects on target detection has not been fully analyzed and resolved.

[0005] Although existing traditional technologies have made progress in radar signal processing and target recognition, most have not yet fully solved the problem of multiple interferences in the complex environment of coal mine roadways, especially in how to simultaneously suppress multiple interference sources and improve the target recognition capability of radar systems. Summary of the Invention

[0006] In view of the above problems, this application proposes a method and device for multi-domain interference identification of millimeter-wave radar in complex environments of coal mine roadways, overcoming the shortcomings of the prior art.

[0007] In a first aspect, embodiments of this application provide a method for multi-domain interference identification using millimeter-wave radar in complex environments of coal mine roadways, including:

[0008] The echo signal data of the tunnel is obtained by using a continuous wave millimeter wave radar deployed in the underground roadway to collect the echo signal.

[0009] Based on the echo signal data, feature extraction and interference suppression are carried out sequentially in the power spectral density domain, range-time domain, Doppler time domain, and range-Doppler domain to construct a multi-domain link feature consisting of spectrum cleaning, energy compensation, dynamic suppression, and trajectory separation.

[0010] The feature matrices corresponding to the multi-domain link features are uniformly represented, and the features of different domains are mapped to a unified representation space, so that the feature matrices are comparable and fusionable in terms of scale, dimension and physical meaning. Then, the features of each domain are weighted and fused to form a unified feature vector after fusion, which is used for target recognition and interference separation.

[0011] Based on the unified feature vector, multiple physical consistency constraints based on motion laws and spatial structures are introduced to output the final recognition result, so that the recognition result has physical rationality and temporal continuity.

[0012] Secondly, embodiments of this application provide a millimeter-wave radar multi-domain interference identification device for complex environments in coal mine roadways, including:

[0013] The acquisition module is used to acquire echo signals using a continuous wave millimeter wave radar deployed in the underground roadway, and obtain echo signal data of the roadway.

[0014] The extraction module is used to perform feature extraction and interference suppression in the power spectral density domain, range-time domain, Doppler time domain, and range-Doppler domain according to the echo signal data, and to construct a multi-domain link feature consisting of spectrum cleaning, energy compensation, dynamic suppression and trajectory separation.

[0015] The unified module is used to perform unified representation processing on the feature matrix corresponding to the multi-domain link features, map the features of different domains to a unified representation space, so that the feature matrix is ​​comparable and fusionable in scale, dimension and physical meaning, and then perform weighted fusion of the features of each domain to form a fused unified feature vector for target recognition and interference separation.

[0016] The constraint recognition module is used to introduce multiple physical consistency constraints based on motion laws and spatial structures based on the unified feature vector and output the final recognition result, so that the recognition result has physical rationality and temporal continuity.

[0017] The multi-domain interference identification method for millimeter-wave radar in complex coal mine roadways proposed in this application can effectively cope with multiple interference sources in such environments. Coal mine roadways are complex environments, with common interferences including coal dust, moisture, gases, and metal reflections. Traditional signal processing methods cannot effectively extract target features in such environments. This application combines multiple signal processing methods, including power spectral density (PSD) analysis, Doppler-time domain (DT) analysis, range-time domain (RT) analysis, and range-Doppler domain (RD) analysis, to effectively distinguish between target signals and interference signals. This method effectively addresses the complex interference environment in coal mine roadways, significantly improving the target identification capability and robustness of millimeter-wave radar systems in complex environments. It provides a new solution for safety monitoring in coal mine roadways, with broad application prospects and high practicality. Attached Figure Description

[0018] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which:

[0019] Figure 1 This is a flowchart of a millimeter-wave radar multi-domain interference identification method for complex environments in coal mine roadways, according to an embodiment of this application.

[0020] Figure 2 This is a block diagram of a millimeter-wave radar multi-domain interference identification device for complex environments in coal mine roadways, as described in this application. Detailed Implementation

[0021] The embodiments of this application will now be described in detail. Examples of these embodiments are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0022] This application proposes a multi-domain interference identification method for millimeter-wave radar in complex environments of coal mine roadways. This method primarily addresses the problems encountered in such environments and proposes a multi-domain interference identification method for millimeter-wave radar in complex coal mine roadways. (Reference) Figure 1 The flowchart shown includes the following steps:

[0023] Step 101: Use a continuous wave millimeter-wave radar deployed in the underground roadway to collect echo signals and obtain echo signal data of the roadway.

[0024] In this embodiment of the invention, a continuous wave (FMCW) millimeter-wave radar is deployed in the underground roadway to collect echo signals and obtain echo signal data of the roadway.

[0025] Continuous wave (FMCW) millimeter-wave radar transmits a continuous frequency modulated signal for target detection; the frequency range of this signal is... f 0, the frequency modulation slope is Δ f The duration is T Continuous wave (FMCW) millimeter-wave radar also receives echo signals reflected back from targets.

[0026] Continuous wave (FMCW) millimeter-wave radar transmits continuous wave signals and receives their reflected echo signals. t ) is represented as:

[0027]

[0028] In the above formula, A The amplitude of the echo signal. f 0 is the center frequency, Δ f For frequency modulation slope, T For duration, t For time variables, j It is the imaginary unit.

[0029] The echo signal is preprocessed, which includes removing background noise and DC components to improve the signal-to-noise ratio (SNR) of the echo signal. The echo signal is then filtered and denoised to obtain echo signal data. Finally, the acquired echo signal data is stored in time series to provide a data foundation for subsequent data analysis and feature extraction.

[0030] Step 102: Based on the echo signal data, feature extraction and interference suppression are carried out sequentially in the power spectral density domain, range-time domain, Doppler time domain, and range-Doppler domain to construct a multi-domain link feature consisting of spectrum cleaning, energy compensation, dynamic suppression, and trajectory separation.

[0031] After obtaining the echo signal data, feature extraction and interference suppression are performed sequentially in the power spectral density domain (PSD), range-time domain (R–T), Doppler-time domain (D–T), and range-Doppler domain (R–D). That is, feature extraction is performed in multiple domains based on the echo signal data. Specifically, this includes:

[0032] Spectral feature extraction and electromagnetic interference anomaly identification are performed on echo signal data in the power spectral density domain (PSD) to identify and suppress spectral anomalies of electromagnetic interference.

[0033] Energy feature extraction and dust scattering anomaly identification are performed on the echo signal data processed in the power spectral density domain in the distance-time domain (R-T) to extract the energy distribution law of dust scattering and achieve energy compensation.

[0034] Dynamic feature extraction and dust disturbance anomaly identification are performed on echo signal data processed in the Doppler-time domain (DT) and the distance-time domain (R-T) domain to identify and suppress the dynamic fluctuation characteristics of dust disturbance.

[0035] Feature trajectory extraction and multipath reflection anomaly identification are performed on the echo signal data processed in the Doppler-time domain (DT) in the range-Doppler domain (R-D) to separate the sub-trajectory structure of multipath reflection and enhance the target response.

[0036] In one embodiment of this application, a preferred method for extracting spectral features and identifying electromagnetic interference anomalies from echo signal data in the power spectral density domain includes:

[0037] First, power spectrum feature extraction is performed, and then a Fast Fourier Transform is applied to the echo signal data to calculate the echo signal at various frequencies. f The power spectral density PSD(f) is expressed as follows:

[0038]

[0039] In the above formula, The received echo signal, The Fourier transform operator is used to reflect the distribution of electromagnetic energy in the frequency dimension.

[0040] Next, electromagnetic interference anomaly identification is performed to identify abnormal frequency bands and calculate the local mean of the power spectrum. with standard deviation And set an adaptive threshold T f as follows:

[0041]

[0042] In the above formula, k For adjustment coefficients, when At that time, determine the frequency point f i Anomalies are identified by their frequency, forming anomaly sets. .

[0043] Calculate anomalous energy And count the percentage of abnormal frequencies. This describes the intensity and range of electromagnetic interference, identifies the concentrated frequency bands of electromagnetic interference through threshold detection, and provides a location basis for subsequent filtering and clearing.

[0044] Next, spectral compensation and feature output are performed to remove abnormal frequency bands. A band-stop filter is used to perform band-stop filtering on the abnormal frequency points, and its expression is:

[0045]

[0046] In the above formula, F s Sampling frequency, r This is the stopband sharpness factor. For abnormal frequencies, the purified signal after filtering to remove electromagnetic interference is expressed as follows:

[0047]

[0048] Based on the expression of the purified signal after electromagnetic interference removal and after normalizing the whitening spectrum, the following formula is obtained:

[0049]

[0050] The purified signal after electromagnetic interference removal suppresses the narrowband electromagnetic peak, which serves as the input signal for subsequent distance-time domain operations, while simultaneously extracting electromagnetic interference features. The purified signal and electromagnetic interference characteristics after electromagnetic interference removal These results are all used as input signals for the weighted fusion of multi-domain features in step 103.

[0051] In one embodiment of this application, a preferred method for extracting energy features and identifying dust scattering anomalies from echo signal data processed in the power spectral density domain in the distance-time domain includes:

[0052] First, distance-time energy features are extracted to analyze the energy distribution of the signal in the distance and time dimensions, which is then applied to the purified signal after electromagnetic interference removal. Performing a distance-dimensional Fourier transform, the energy matrix is ​​expressed as follows:

[0053]

[0054] In the above formula, r is the distance variable. This represents the Fourier transform along the distance dimension.

[0055] Next, wavelets are used to decompose the signal into low-frequency background. With high frequency components Low-frequency background This refers to the energy scattered by dust, specifically the high-frequency components. This is the target signal, and its expression is as follows:

[0056]

[0057] Next, anomaly identification of dust scattering is performed, and the statistical energy change law along the distance direction is used to fit the dust attenuation model as follows:

[0058]

[0059] In the above formula, Let I0 be the average energy at a distance r, I0 be the initial energy, and α be the dust attenuation coefficient. This indicates an abnormal dust concentration or attenuation.

[0060] Simultaneously define the dust energy ratio DER To measure the proportion of dust scattering energy in the total signal:

[0061]

[0062] The dust energy ratio (DER) is used to identify and quantify the masking and attenuation effects of dust on signal energy.

[0063] Then, dust energy compensation and characteristic output are performed to execute low-frequency energy subtraction and normalization compensation, thereby obtaining the energy compensation matrix. Its expression is as follows:

[0064]

[0065] In the above formula, To prevent division by zero, a small constant is typically 1e. 6.

[0066] The contrast enhancement rate was calculated after final compensation. CIR The expression is as follows:

[0067]

[0068] Obtain the energy compensation matrix To significantly enhance target echo contrast, this energy compensation matrix serves as the input signal in the Doppler time domain for detecting dynamic disturbances and simultaneously extracting dust features. This serves as the input signal for the weighted fusion of multi-domain features in step 103.

[0069] In one embodiment of this application, a preferred method for dynamic feature extraction and dust disturbance anomaly identification of echo signal data processed in the distance-time domain in the Doppler time domain includes:

[0070] First, Doppler time dynamic features are extracted to further identify the time-varying characteristics of target motion and dust disturbance. Then, time-Doppler analysis is performed on the signal after dust energy compensation. This process uses short-time Fourier transform to achieve local spectral expansion in the time domain, thereby obtaining the joint distribution of signal energy in time and Doppler frequency shift, expressed as follows:

[0071]

[0072] In the above formula, This indicates performing a local Fourier transform on the signal within a time window, where t is time. The Doppler frequency shift reflects the radial velocity change of the scatterer.

[0073] The target radial velocity can be calculated from the Doppler frequency shift, and its expression is as follows:

[0074]

[0075] In the above formula, f 0 is the center frequency. c The speed of light (usually 3 × 10⁻⁶) 8 (m / s). Through this transformation, the signal is mapped onto the time-Doppler plane. If a continuously moving target presents as a time-continuous, frequency-narrow energy ridge, it indicates smooth velocity changes and stable reflection. Dust disturbances, on the other hand, form short-duration, broadband, and randomly distributed energy patches, indicating that the reflector particles are small in size and move randomly.

[0076] Next, dust disturbance anomaly identification is performed to quantitatively distinguish whether there is dust disturbance anomaly. Three types of dynamic criteria are defined: trajectory continuity index, local bandwidth index, and energy stability index.

[0077] Among them, trajectory continuity index for:

[0078]

[0079] In the above formula, L c The continuous length of the Doppler ridge line. T For analysis of time windows. When R c →1, the target trajectory is continuous, indicating stable motion; when R c 1 indicates energy breakage, signifying the presence of dust disturbance.

[0080] Local bandwidth index for:

[0081]

[0082] In the above formula, and The 95% and 5% frequency positions of the cumulative energy distribution at that moment, when the target's motion energy is concentrated. Very small; energy diffuses when dust is disturbed. Significantly increased.

[0083] Energy stability index for:

[0084]

[0085] In the above formula, and Let represent the time variance and mean, respectively. When the signal energy fluctuates drastically over time... An increase indicates frequent short-term disturbances.

[0086] Based on the three indicators of trajectory continuity, local bandwidth, and energy stability, the disturbance criterion is established as follows: If or or This area is then identified as a dust disturbance zone. β and γ are empirical thresholds, calibrated by measured samples. Continuous targets have high continuity, narrow band, and stable energy characteristics; dust disturbances have low continuity, wide band, and random energy characteristics.

[0087] Then, dynamic perturbation suppression and feature output are performed to adaptively suppress the identified dust perturbation area. The soft mask function is constructed as follows:

[0088]

[0089] In the above formula, σ( () is the Sigmoid function. a 1, a 2, a 3 is an adjustment coefficient used to balance the effects of continuity and diffusion.

[0090] The masking function is used for spectral suppression, and its expression is as follows:

[0091]

[0092] when The target trajectory is preserved during the process; when Suppress dust energy.

[0093] To eliminate short-term jumps in the target velocity, Kalman filtering is applied to the frequency shift of the main ridge line, and its expression is as follows:

[0094]

[0095] In the above formula, This is the current estimated frequency; For predicting frequencies; For observation frequency; Kalman gain; H This is the observation matrix.

[0096] The smoothed velocity reflects the true motion state of the target without being affected by random disturbances, and is expressed as:

[0097]

[0098] To quantify the inhibition effect, a dynamic residual energy ratio is introduced:

[0099]

[0100] In the above formula, For the sum of the absolute values ​​of matrices, when A lower value indicates that the dust disturbance energy has been effectively weakened;

[0101] The final output features are dynamic features. As the input signal for the weighted fusion of multi-domain features in step 103, the transplanted spectrum The signal is transmitted to the distance Doppler domain to further enable multipath reflection anomaly identification.

[0102] In one embodiment of this application, a preferred method for feature trajectory extraction and multipath reflection anomaly identification of echo signal data processed in the Doppler time domain in the range-Doppler domain includes:

[0103] After dust disturbance suppression in the time Doppler domain, the dynamic noise of the echo signal is reduced, but multipath reflection subtrajectories caused by tunnel walls, equipment metal structures, etc., may still exist. If multipath reflection subtrajectories exist in the echo signal, feature trajectory extraction and multipath reflection anomaly identification are performed in the range Doppler domain to further distinguish between the true target main trajectory and the multipath subtrajectories.

[0104] First, range-Doppler trajectory features are extracted to simultaneously reflect the target's spatial distance distribution and velocity characteristics. This is then applied to the input spectral map. Performing a two-dimensional fast Fourier transform (2D-FFT) yields the range-Doppler energy spectrum, expressed as follows:

[0105]

[0106] In the above formula, FFTr( ) performs a one-dimensional FFT along the distance dimension (fast time) to obtain the distance gate energy distribution; FFTt( This involves performing a one-dimensional FFT along the time dimension (slow time) to extract the Doppler frequency shift. r Displacement distance variable (m); f D The Doppler frequency shift (Hz) is related to the target radial velocity v as follows: c The speed of light; f 0 represents the radar carrier frequency; A spectrum (dB) reflecting the energy distribution of the target in the range-velocity space.

[0107] On the plane of the distance to the Doppler domain, the energy distribution of a real target usually forms a single main ridge. If there is reflection from a wall or equipment, a secondary trajectory parallel to or offset from the main ridge will appear. The energy distribution spectrum visually reveals the spatial motion coupling characteristics of the target echo.

[0108] Next, multipath reflection anomaly identification is performed to identify reflection signals caused by walls, ground, or equipment by quantifying the offset of the primary and secondary trajectories in terms of distance and Doppler values. Specifically, this includes:

[0109] First, extract the peak coordinates of the main trajectory and the secondary trajectory, and calculate their distance and frequency offset:

[0110]

[0111] In the above formula, , The peak position corresponding to the main trajectory; , If the peak position corresponding to the trajectory, or The spatial or velocity characteristics of the secondary trajectory are significantly different from those of the primary trajectory, indicating multipath reflection interference. and These are the range and the Doppler threshold, respectively. These values ​​are determined by the system's geometric layout and empirical data. Multipath propagation typically originates from secondary reflections of radar waves from walls or mechanical structures, resulting in longer propagation paths and increased range. Meanwhile, the phase delay causes a pseudo-velocity component ( "Mimetic" refers to the fact that the velocity component is a non-physical real motion velocity caused by phase delay. Since multipath reflection causes the signal propagation path to become longer, thus introducing additional phase delay, this part of the phase offset will be mismapped as a velocity value that is not generated by the actual motion of the target during Doppler frequency calculation. Therefore, multipath identification is achieved by joint offset detection.

[0112] Next, multipath separation and feature output are performed to separate the real target from the multipath reflection components in the mixed signal, so as to preserve the main trajectory and achieve robust target representation. The observed range Doppler signal is regarded as a superposition of multiple scattering sources (referring to the main target and multipath reflections) through a multipath mixing model, and its expression is as follows:

[0113]

[0114] In the above formula, X The observed signal matrix is ​​obtained by the radar receiver on the range-velocity plane; , where is the source signal vector, corresponding to the main target echo and the multipath reflection echo, respectively; A This is a hybrid matrix that describes the propagation attenuation and phase shift along different paths; N This is the noise term.

[0115] The main signal was isolated by independent component analysis. and secondary signals :

[0116]

[0117] In the above formula, These represent the separated main trajectory signal and sub-trajectory signal, respectively. The main trajectory signal... Used for trajectory smoothing and feature extraction, sub-trajectory signal Used for multipath quantization.

[0118] To smooth the main trajectory state and constrain continuity, a recursive Kalman filter is used for temporal smoothing to address velocity abrupt changes caused by multipath propagation. The expression for this filter is as follows:

[0119]

[0120] In the above formula, r k For distance, v k For speed;

[0121] The state and observation equations are defined as follows:

[0122]

[0123] The following equation is obtained by recursively updating using Kalman filtering:

[0124]

[0125] In the above formula, F Represented as a state transition matrix, it is determined by the sampling period and the motion model; H The observation matrix; It is by Extracted observation vectors; The Kalman gain is adaptively updated based on the error covariance.

[0126] After filtering, a smooth trajectory sequence is obtained. And calculate the continuity index, the expression of which is:

[0127]

[0128] In the above formula, e To prevent small constants with denominators of zero (usually taken as 10) -6 ),when The good continuity of the trajectory indicates that the main trajectory is stable and free from multipath disturbances.

[0129] In this step, Kalman filtering is used for temporal smoothing of the local main trajectory. In subsequent step 104, global Kalman consistency constraints will be further introduced at the fusion level.

[0130] The final output is the distance-Doppler domain feature vector: The feature result serves as the input signal for the multi-domain feature weighted fusion module in step 103. Together with the outputs from the power spectral density domain, distance-time domain, and Doppler time domain, it constructs a unified representation space, thereby achieving highly robust identification of targets in the roadway.

[0131] Step 103: Perform unified representation processing on the feature matrices corresponding to the multi-domain link features, map the features of different domains to a unified representation space, so that the feature matrices are comparable and fusionable in scale, dimension and physical meaning, and then perform weighted fusion of the features of each domain to form a unified feature vector after fusion, which is used for target recognition and interference separation.

[0132] After extracting the link features from each domain in the aforementioned steps, the corresponding feature matrices are subjected to unified representation processing, thereby mapping the features of different domains to a unified representation space. This makes the feature matrices comparable and fusionable in terms of scale, dimension, and physical meaning. Then, the features of each domain are weighted and fused to form a unified feature vector, which is used for target recognition and interference separation.

[0133] In one embodiment of this application, a preferred method for uniformly projecting features extracted from different domains onto a representation space, making the feature matrices comparable and fusionable in terms of scale, dimension, and physical meaning, and then weightedly fusing the features from each domain to form a unified feature vector includes:

[0134] First, the features from each domain are weighted and fused to construct a multi-domain fusion deep recognition framework, forming a unified feature vector after fusion, which is used for target recognition and interference separation. Specifically, this includes:

[0135] First, multi-domain features are summarized, and the outputs from each domain are combined to obtain four types of feature vectors, whose expressions are as follows:

[0136]

[0137] The four types of feature vectors mentioned above correspond to electromagnetic interference spectrum, dust energy distribution, dynamic disturbance and multipath trajectory characteristics, respectively, and together reflect the main interference and motion elements of millimeter-wave radar signals in the complex environment of coal mine roadways.

[0138] Next, feature standardization and representation space mapping are performed to eliminate the differences in feature units and dynamic ranges between different domains. This requires the standardization operator Φ( Standardize the scale:

[0139]

[0140] In the above formula, m i , s i The first i The mean and standard deviation of the domain features. These are the normalized eigenvectors.

[0141] Subsequently, canonical correlation analysis was used to project the multi-domain features onto a unified representation space:

[0142]

[0143] In the above formula, W i For the projection matrices of each domain, maximize cross-domain correlation. Learned; To maintain cross-domain comparability and physical consistency for the mapped feature representation.

[0144] Next, information gain weights are calculated to quantify the contribution of each domain feature to target recognition, and the information gain is calculated as follows:

[0145]

[0146] In the above formula, Represents the target label entropy; Let the target entropy be given the features of the i-th domain. This represents the information gain of the feature domain.

[0147] The weights of each domain are then determined based on the information gain as follows:

[0148]

[0149] Next, multi-domain feature weighted fusion is performed to weight and fuse the projected features from each domain according to weight coefficients, forming a unified representation feature vector:

[0150]

[0151] After fusion, the total characteristic energy index is calculated as follows:

[0152]

[0153] In the above formula, Energy intensity and fusion confidence level are used to reflect overall characteristics.

[0154] Fusion feature vectors As the output of multi-domain fusion recognition, it constitutes the input features for subsequent physical consistency constraints.

[0155] Finally, key physical features are extracted and observation variables are generated to extract features with clear physical meaning from the multi-domain representation matrix. The main peak position, energy centroid, or main trajectory center of each domain matrix corresponds to the main features of distance, velocity, and energy, respectively.

[0156]

[0157] In the above formula, Mi( ) is the first i The energy matrix of the domain.

[0158] To achieve consistent representation across multiple domains, the eigenvalues ​​are normalized and weighted according to information gain, as shown in the following expression:

[0159]

[0160] In the above formula, , The normalized distance and velocity characteristics, The energy of the main peak in each domain is used; after weighting, three physical characteristics are obtained: fusion distance, fusion speed, and fusion energy.

[0161] The final fusion result is defined as the observed input:

[0162]

[0163] The observed input is directly passed to step 104 for physical consistency constraints and trajectory recognition optimization.

[0164] Step 104: Based on the unified feature vector, introduce multiple physical consistency constraints based on motion laws and spatial structures and output the final recognition result to make the recognition result physically reasonable and temporally continuous.

[0165] After obtaining a unified feature vector, multiple physical consistency constraints based on motion laws and spatial structure are introduced to output the final recognition result, ensuring that the recognition result has physical rationality and temporal continuity. The multiple physical consistency of motion laws and spatial structure includes: velocity consistency check, spatial redundancy suppression, temporal smoothing, and continuity.

[0166] A preferred speed consistency check includes:

[0167] After completing the multi-domain feature fusion, to ensure that the target's trajectory conforms to physical laws, it is necessary to impose consistent constraints on velocity changes. First, the instantaneous Doppler frequency components of the target's main trajectory are extracted from the spectrum in the range-Doppler domain. f D,k The peak position of the energy distribution in the Doppler domain spectrum is taken from the distance, and its calculation method is as follows:

[0168]

[0169] In the above formula, The distance-Doppler spectrum obtained by Fourier transform; r k The distance position corresponds to the main trajectory of the current frame; the instantaneous Doppler main peak frequency reflects the instantaneous velocity of the target.

[0170] For sequence f D,k A smoothing correction is performed to avoid pseudo-frequency fluctuations caused by noise spikes (where "pseudo" refers to frequency fluctuations that are not physically real frequency changes caused by noise spikes). The expression for this correction is as follows:

[0171]

[0172] In the above formula, T f The frequency variation threshold is set by the radar system bandwidth and sampling rate, and is corrected accordingly. Used for subsequent speed consistency calculations.

[0173] According to the radar Doppler effect, instantaneous radial velocity v kCalculated from the frequency components of the main trajectory:

[0174]

[0175] In the above formula, To observe radial velocity; c The speed of light;

[0176] The rate of change of velocity between adjacent frames is defined as follows:

[0177]

[0178] If Δ v k > T v If the velocity change does not conform to the law of continuous motion, a weighted correction is applied as follows:

[0179]

[0180] In the above formula For smoothing coefficients, T v This is the speed jump threshold.

[0181] Revised This serves as a spatial redundancy and temporal smoothing and continuity process following the observation velocity input, ensuring that the main trajectory velocity sequence is continuous and stable in the time dimension.

[0182] Spatial redundancy suppression includes:

[0183] Considering that the deployment of radar on metal components or at multiple nodes in the tunnel may lead to duplicate target detections, clustering and redundancy reduction are performed on the observation point set after velocity consistency verification. This specifically includes:

[0184] Let the set of detection points be:

[0185]

[0186] In the above formula, For fusion observation distance (m); This is the observation rate after consistency correction; This corresponds to the energy characteristic (dB).

[0187] Redundancy detection is performed using the density clustering (DBSCAN) method based on spatial velocity similarity.

[0188]

[0189] In the above formula, Δ r 0, Δ v 0 represents the normalized scaling parameter; This is the clustering threshold used to determine whether objects belong to the same physical target cluster.

[0190] The energy-weighted merging of intra-cluster clicks is as follows:

[0191]

[0192] Obtain a unique target representation This is to achieve redundant detection elimination and feature aggregation.

[0193] Temporal smoothing and continuity include:

[0194] To ensure the temporal continuity of the target trajectory, the observation sequence after spatial redundancy suppression is... Global smoothing is performed using a recursive Kalman filter, and the global state vector is defined as follows:

[0195]

[0196] In the above formula, The target distance is estimated globally. The target velocity is estimated globally; This is the global state vector.

[0197] The state and observation model is as follows:

[0198]

[0199] In the above formula, This is the global state transition matrix. T s The sampling period; The observation matrix; This refers to the process noise term; This is for observing the noise term.

[0200] The update equation for global smoothing using Kalman filtering recursively is:

[0201]

[0202] In the above formula, The Kalman gain matrix is ​​adaptively updated based on the error covariance. This represents the smoothed global trajectory state.

[0203] The trajectory continuity index is defined as:

[0204]

[0205] In the above formula, To prevent small constants with a denominator of zero, when At that time, it indicates that the global trajectory is smooth and consistent.

[0206] The final recognition results are output as follows:

[0207] After multiple physical consistency constraints, the final identification status of the output target is as follows:

[0208]

[0209] In the above formula, This is the final estimated target distance; The final estimated target speed; To integrate energy characteristics; It serves as an indicator of trajectory continuity.

[0210] The final identification result comprehensively represents the target's spatial location, motion state, energy intensity, and temporal stability. This is the final identification result output by the system and can be directly used for target classification, trajectory monitoring, and stability assessment.

[0211] Based on the above-mentioned millimeter-wave radar multi-domain interference identification method for complex environments in coal mine roadways, this application also proposes a millimeter-wave radar multi-domain interference identification device for complex environments in coal mine roadways, referring to... Figure 2 The block diagram shown includes:

[0212] The acquisition module 210 is used to acquire echo signals using a continuous wave millimeter wave radar deployed in the underground roadway to obtain echo signal data of the roadway.

[0213] The extraction module 220 is used to perform feature extraction and interference suppression in the power spectral density domain, range-time domain, Doppler time domain, and range-Doppler domain according to the echo signal data, and construct a multi-domain link feature consisting of spectrum cleaning, energy compensation, dynamic suppression and trajectory separation.

[0214] The unified module 230 is used to perform unified representation processing on the feature matrix corresponding to the multi-domain link features, map the features of different domains to a unified representation space, so that the feature matrix is ​​comparable and fusionable in scale, dimension and physical meaning, and then perform weighted fusion of the features of each domain to form a fused unified feature vector for target recognition and interference separation.

[0215] The constraint recognition module 240 is used to introduce multiple physical consistency constraints based on motion laws and spatial structures based on the unified feature vector and output the final recognition result so that the recognition result has physical rationality and temporal continuity.

[0216] Optionally, the acquisition module 210 is specifically used for:

[0217] The continuous-wave millimeter-wave radar is deployed in the underground tunnels to transmit continuous frequency-modulated signals for target detection and to receive echo signals reflected back from the targets. The frequency range of the frequency-modulated signals is [missing information]. f 0, the frequency modulation slope is Δ f The duration is T ;

[0218] The echo signal Echo ( t ) is represented as:

[0219]

[0220] In the above formula, A is the amplitude of the echo signal. f 0 is the center frequency, Δ f For frequency modulation slope, T For duration, t For time variables, j The imaginary unit;

[0221] The echo signal is preprocessed, which includes removing background noise and DC component to improve the signal-to-noise ratio of the echo signal, and the echo signal is filtered and denoised to obtain the echo signal data.

[0222] The echo signal data is stored in time series to provide a data foundation for subsequent data analysis and feature extraction.

[0223] Optionally, the extraction module 220 includes:

[0224] The spectral feature extraction and electromagnetic interference anomaly identification submodule is used to perform spectral feature extraction and electromagnetic interference anomaly identification on the echo signal data in the power spectral density domain, so as to identify and suppress spectral anomalies of electromagnetic interference.

[0225] The energy feature extraction and dust scattering anomaly identification submodule is used to extract energy features and identify dust scattering anomalies in the echo signal data processed in the power spectral density domain in the distance-time domain, so as to extract the energy distribution law of dust scattering and realize energy compensation.

[0226] The dynamic feature extraction and dust disturbance anomaly identification submodule is used to perform dynamic feature extraction and dust disturbance anomaly identification on the echo signal data processed in the distance-time domain in the Doppler time domain, so as to identify and suppress the dynamic fluctuation characteristics of dust disturbance.

[0227] The feature trajectory feature extraction and multipath reflection anomaly identification submodule is used to perform feature trajectory feature extraction and multipath reflection anomaly identification on the echo signal data after Doppler time domain processing in the range Doppler domain, so as to separate the sub-trajectory structure of multipath reflection and enhance the target response.

[0228] Optionally, the spectral feature extraction and electromagnetic interference anomaly identification submodule is specifically used for:

[0229] First, power spectrum feature extraction is performed, and then a Fast Fourier Transform is applied to the echo signal data to calculate the echo signal at various frequencies. f The power spectral density PSD(f) is expressed as follows:

[0230]

[0231] In the above formula, The received echo signal, The Fourier transform operator is used to describe the distribution of electromagnetic energy in the frequency dimension.

[0232] Next, electromagnetic interference anomaly identification is performed to identify abnormal frequency bands and calculate the local mean of the power spectrum. with standard deviation And set the adaptive threshold Tf as follows:

[0233]

[0234] In the above formula, k For adjustment coefficients, when At that time, determine the frequency point f i Anomalies are identified by their frequency, forming anomaly sets. ;

[0235] Calculate anomalous energy And count the percentage of abnormal frequencies. This describes the intensity and range of electromagnetic interference, identifies the concentrated frequency bands of electromagnetic interference through threshold detection, and provides a location basis for subsequent filtering and clearing.

[0236] Then perform spectrum compensation and feature output.

[0237] To remove abnormal frequency bands, a band-stop filter is used to perform band-stop filtering on the abnormal frequency points. The expression is as follows:

[0238]

[0239] In the above formula, F s Sampling frequency, r This is the stopband sharpness factor. For abnormal frequencies, the purified signal after filtering to remove electromagnetic interference is expressed as follows:

[0240]

[0241] Based on the expression of the purified signal after electromagnetic interference removal and after normalizing the whitening spectrum, the following formula is obtained:

[0242]

[0243] The purified signal after electromagnetic interference removal suppresses the electromagnetic narrowband peak, which serves as the input signal in the subsequent distance-time domain, while simultaneously extracting electromagnetic interference features. The purified signal after electromagnetic interference removal and the electromagnetic interference features are both used as input signals for weighted fusion of multi-domain features.

[0244] Optionally, the energy feature extraction and dust scattering anomaly identification submodule is specifically used for:

[0245] First, distance-time energy features are extracted to analyze the energy distribution of the signal in the distance and time dimensions, and then applied to the purified signal after electromagnetic interference removal. Performing a distance-dimensional Fourier transform, the energy matrix is ​​expressed as follows:

[0246]

[0247] In the above formula, r is the distance variable. This represents the Fourier transform along the distance dimension;

[0248] Next, wavelets are used to decompose the signal into low-frequency background. With high frequency components The low-frequency background This refers to the energy scattered by dust, specifically the high-frequency components. This is the target signal, and its expression is as follows:

[0249]

[0250] Next, anomaly identification of dust scattering is performed, and the statistical energy change law along the distance direction is used to fit the dust attenuation model as follows:

[0251]

[0252] In the above formula, Let I0 be the average energy at a distance r, I0 be the initial energy, and α be the dust attenuation coefficient. This indicates an abnormal dust concentration or attenuation.

[0253] Simultaneously define the dust energy ratio DER To measure the proportion of dust scattering energy in the total signal:

[0254]

[0255] The dust energy ratio (DER) is used to identify and quantify the masking and attenuation effects of dust on signal energy.

[0256] Then, dust energy compensation and characteristic output are performed to execute low-frequency energy subtraction and normalization compensation, thereby obtaining the energy compensation matrix. Its expression is as follows:

[0257]

[0258] In the above formula, To prevent division by zero of small constants;

[0259] The contrast enhancement rate was calculated after final compensation. CIR The expression is as follows:

[0260]

[0261] Obtain the energy compensation matrix To enhance target echo contrast, the energy compensation matrix serves as the input signal in the Doppler time domain, used to detect dynamic disturbances and simultaneously extract dust features. It serves as the input signal for weighted fusion of multi-domain features.

[0262] Optionally, the dynamic feature extraction and dust disturbance anomaly identification submodule is specifically used for:

[0263] First, Doppler time dynamic features are extracted to further identify the time-varying characteristics of target motion and dust disturbance. Then, time-Doppler analysis is performed on the signal after dust energy compensation. This process uses short-time Fourier transform to achieve local spectral expansion in the time domain, thereby obtaining the joint distribution of signal energy in time and Doppler frequency shift, expressed as follows:

[0264]

[0265] In the above formula, This indicates performing a local Fourier transform on the signal within a time window, where t is time. The Doppler frequency shift reflects the radial velocity change of the scattering body;

[0266] The target radial velocity is calculated using the Doppler frequency shift, and its expression is as follows:

[0267]

[0268] In the above formula, f 0 is the center frequency. cAssuming the speed of light, through this transformation, the signal is mapped onto the time-Doppler plane. If a continuously moving target presents as a time-continuous, frequency-narrow band of energy ridges, it indicates smooth velocity changes and stable reflection; dust disturbances, on the other hand, form short-duration, broadband, and randomly distributed energy patches, indicating that the reflector particles are small in size and move randomly.

[0269] Next, dust disturbance anomaly identification is performed to quantitatively distinguish whether there is dust disturbance anomaly. Three types of dynamic criteria are defined: trajectory continuity index, local bandwidth index, and energy stability index.

[0270] Among them, the trajectory continuity index for:

[0271]

[0272] In the above formula, L c The continuous length of the Doppler ridge line. T For analysis of time windows. When R c →1, the target trajectory is continuous, indicating stable motion; when R c 1 indicates energy breakage, signifying the presence of dust disturbance;

[0273] The local bandwidth index for:

[0274]

[0275] In the above formula, and The 95% and 5% frequency positions of the cumulative energy distribution at that moment, when the target's motion energy is concentrated. Very small; energy diffuses when dust is disturbed. Significantly increased;

[0276] The energy stability index for:

[0277]

[0278] In the above formula, and Let represent the time variance and mean, respectively. When the signal energy fluctuates drastically over time... An increase indicates frequent short-term disturbances;

[0279] Combining the trajectory continuity index, the local bandwidth index, and the energy stability index, the disturbance criterion is established as follows: if or or This area is then identified as a dust disturbance zone. β and γ are empirical thresholds, calibrated from measured samples. Continuous targets have high continuity, narrow bandwidth, and stable energy characteristics; dust disturbances have low continuity, wide bandwidth, and random energy characteristics.

[0280] Then, dynamic perturbation suppression and feature output are performed to adaptively suppress the identified dust perturbation area. The soft mask function is constructed as follows:

[0281]

[0282] In the above formula, σ( () is the Sigmoid function. a 1, a 2, a 3 is an adjustment coefficient used to balance the effects of continuity and diffusion;

[0283] The mask function is applied for spectral suppression, and its expression is as follows:

[0284]

[0285] when The target trajectory is preserved during the process; when Energy to suppress dust;

[0286] To eliminate short-term jumps in the target velocity, Kalman filtering is applied to the frequency shift of the main ridge line, and its expression is as follows:

[0287]

[0288] In the above formula, This is the current estimated frequency; For predicting frequencies; For observation frequency; Kalman gain; H The observation matrix;

[0289] The smoothed velocity reflects the true motion of the target without being affected by random disturbances, and is expressed as:

[0290]

[0291] To quantify the inhibition effect, a dynamic residual energy ratio is introduced:

[0292]

[0293] In the above formula, For the sum of the absolute values ​​of matrices, when A lower value indicates that the dust disturbance energy has been effectively weakened;

[0294] The final output features are dynamic features. The transplanted spectrum is the input signal used for weighted fusion of multi-domain features. The signal is transmitted to the distance Doppler domain to further enable multipath reflection anomaly identification.

[0295] Optionally, the feature trajectory feature extraction and multipath reflection anomaly identification submodule is specifically used for:

[0296] If the echo signal has a multipath reflection secondary trajectory, in order to distinguish the real target main trajectory from the multipath secondary trajectory, feature trajectory extraction and multipath reflection anomaly identification are performed in the range Doppler domain.

[0297] First, distance Doppler trajectory features are extracted.

[0298] To simultaneously reflect the spatial distance distribution and velocity characteristics of the target, the input spectrum... Performing a two-dimensional fast Fourier transform yields the range-Doppler energy spectrum, which is expressed as follows:

[0299]

[0300] In the above formula, FFTr( ) performs a one-dimensional FFT along the distance dimension to obtain the distance gate energy distribution; FFTt( This involves performing a one-dimensional FFT along the time dimension to extract the Doppler frequency shift. r Displacement distance variable; f D The Doppler frequency shift is related to the target radial velocity v as follows: c The speed of light; f 0 represents the radar carrier frequency; A spectrum reflecting the energy distribution of a target in the range-velocity space;

[0301] On the plane of the distance-Doppler domain, the energy distribution of a real target typically forms a single main ridge; if there is reflection from a wall or device, a secondary trajectory parallel to or offset from the main ridge will appear. The energy distribution spectrum visually reveals the spatial motion coupling characteristics of the target echo.

[0302] Next, multipath reflection anomaly identification is performed to identify reflection signals caused by walls, ground, or equipment by quantifying the offset of the primary and secondary trajectories in terms of distance and Doppler values. Specifically, this includes:

[0303] First, extract the peak coordinates of the main trajectory and the secondary trajectory, and calculate their distance and frequency offset:

[0304]

[0305] In the above formula, , The peak position corresponding to the main trajectory; , If the peak position corresponding to the trajectory, or The spatial or velocity characteristics of the secondary trajectory are significantly different from those of the primary trajectory, indicating multipath reflection interference. and These are the range and the Doppler threshold, respectively. These values ​​are determined by the system's geometric layout and empirical data. Multipath secondary trajectories usually originate from the secondary reflection of radar waves through walls or mechanical structures. Their propagation paths are relatively long, leading to increased range. At the same time, phase delay causes a pseudo-velocity component. Therefore, multipath identification is achieved through joint offset detection.

[0306] Next, multipath separation and feature output are performed to separate the real target from the multipath reflection components in the mixed signal, so as to preserve the main trajectory and achieve robust target representation. The observed range Doppler signal is regarded as the superposition of multiple scattering sources through the multipath mixing model, and its expression is as follows:

[0307]

[0308] In the above formula, X The observed signal matrix is ​​obtained by the radar receiver on the range-velocity plane; , where is the source signal vector, corresponding to the main target echo and the multipath reflection echo, respectively; A This is a hybrid matrix that describes the propagation attenuation and phase shift along different paths; N Noise term;

[0309] The main signal was isolated by independent component analysis. and secondary signals :

[0310]

[0311] In the above formula, These represent the separated main trajectory signal and sub-trajectory signal, respectively. The main trajectory signal... Used for trajectory smoothing and feature extraction, sub-trajectory signal Used for multipath quantization;

[0312] To smooth the main trajectory state and constrain continuity, a recursive Kalman filter is used for temporal smoothing to address velocity abrupt changes caused by multipath propagation. The expression for this filter is as follows:

[0313]

[0314] In the above formula, r k For distance, v kFor speed;

[0315] The state and observation equations are defined as follows:

[0316]

[0317] The following equation is obtained by recursively updating using Kalman filtering:

[0318]

[0319] In the above formula, F Represented as a state transition matrix, it is determined by the sampling period and the motion model; H The observation matrix; It is by Extracted observation vectors; The Kalman gain is adaptively updated based on the error covariance.

[0320] After filtering, a smooth trajectory sequence is obtained. And calculate the continuity index, the expression of which is:

[0321]

[0322] In the above formula, e To prevent small constants with a denominator of zero, when At that time, the trajectory continuity was good, indicating that the main trajectory was stable and there was no multipath disturbance;

[0323] The Kalman filter is used for temporal smoothing of the local main trajectory, while a global Kalman consistency constraint is introduced at the fusion level.

[0324] The final output is the distance-Doppler domain feature vector: The feature result serves as the input signal for the multi-domain feature weighted fusion module, and together with the outputs of the power spectral density domain, the distance-time domain, and the Doppler time domain, it constructs a unified representation space to achieve highly robust identification of targets in the alleyway.

[0325] Optionally, the unified module 230 is specifically used for:

[0326] The features from each domain are weighted and fused to construct a multi-domain fusion deep recognition framework, forming a unified feature vector for target recognition and interference separation. Specifically, this includes:

[0327] First, multi-domain features are summarized, and the outputs from each domain are combined to obtain four types of feature vectors, whose expressions are as follows:

[0328]

[0329] The four types of feature vectors correspond to electromagnetic interference spectrum, dust energy distribution, dynamic disturbance and multipath trajectory characteristics, respectively, and together reflect the main interference and motion elements of millimeter-wave radar signals in the complex environment of coal mine roadways.

[0330] Next, feature standardization and representation space mapping are performed.

[0331] To eliminate the differences in characteristic units and dynamic ranges between different domains, it is necessary to use the standardization operator Φ( Standardize the scale:

[0332]

[0333] In the above formula, m i , s i The first i The mean and standard deviation of the domain features. These are the normalized feature vectors;

[0334] Subsequently, canonical correlation analysis was used to project the multi-domain features onto a unified representation space:

[0335]

[0336] In the above formula, W i For the projection matrices of each domain, maximize cross-domain correlation. Learned; To maintain cross-domain comparability and physical consistency in the mapped feature representation;

[0337] Next, information gain weights are calculated to quantify the contribution of each domain feature to target recognition, and the information gain is calculated as follows:

[0338]

[0339] In the above formula, Represents the target label entropy; Let the target entropy be given the features of the i-th domain. Information gain for the feature domain;

[0340] The weights of each domain are then determined based on the information gain as follows:

[0341]

[0342] Next, multi-domain feature weighted fusion is performed to weight and fuse the projected features from each domain according to weight coefficients, forming a unified representation feature vector:

[0343]

[0344] After fusion, the total characteristic energy index is calculated as follows:

[0345]

[0346] In the above formula, Energy intensity and fusion confidence level used to reflect overall characteristics;

[0347] Fusion feature vectors As the output of multi-domain fusion recognition, it constitutes the input features for subsequent physical consistency constraints;

[0348] Finally, key physical features are extracted and observation variables are generated to extract features with clear physical meaning from the multi-domain representation matrix. The main peak position, energy centroid, or main trajectory center of each domain matrix corresponds to the main features of distance, velocity, and energy, respectively.

[0349]

[0350] In the above formula, Mi( ) is the first i The energy matrix of the domain;

[0351] Multi-domain consistent representation is achieved by normalizing eigenvalues ​​and weighting them according to information gain, and its expression is as follows:

[0352]

[0353] In the above formula, , The normalized distance and velocity characteristics, The energy of the main peak in each domain is used; after weighting, three physical characteristics are obtained: fusion distance, fusion speed, and fusion energy.

[0354] The final fusion result is defined as the observed input:

[0355]

[0356] The observed input is used for physical consistency constraints and trajectory recognition optimization.

[0357] Optionally, the multiple physical consistency between the motion law and the spatial structure includes: velocity consistency verification, spatial redundancy suppression, temporal smoothing, and continuity; the constraint recognition module 240 is specifically used to perform constraints and output the final recognition result, including:

[0358] The speed consistency check includes:

[0359] The instantaneous Doppler frequency components of the target's main trajectory are extracted from the spectrum of the range-Doppler domain. fD,k The peak position of the energy distribution in the Doppler domain spectrum is taken from the distance, and its calculation method is as follows:

[0360]

[0361] In the above formula, The distance-Doppler spectrum obtained by Fourier transform; r k The distance position corresponds to the main trajectory of the current frame; the instantaneous Doppler main peak frequency reflects the instantaneous velocity of the target;

[0362] For sequence f D,k A smoothing correction is performed to avoid pseudo-frequency fluctuations caused by noise spikes; its expression is as follows:

[0363]

[0364] In the above formula, T f The frequency variation threshold is set by the radar system bandwidth and sampling rate, and is corrected accordingly. Used for subsequent speed consistency calculations;

[0365] According to the radar Doppler effect, instantaneous radial velocity v k Calculated from the frequency components of the main trajectory:

[0366]

[0367] In the above formula, To observe radial velocity; c The speed of light;

[0368] The rate of change of velocity between adjacent frames is defined as follows:

[0369]

[0370] If Δ v k > T v If the velocity change does not conform to the law of continuous motion, a weighted correction is applied as follows:

[0371]

[0372] In the above formula For smoothing coefficients, T v The speed jump threshold;

[0373] Revised This serves as a spatial redundancy and temporal smoothing and continuity process following the observation velocity input, ensuring that the main trajectory velocity sequence is continuous and stable in the time dimension;

[0374] The spatial redundancy suppression includes:

[0375] Clustering and redundancy reduction are performed on the observation set after the velocity consistency test, which specifically includes:

[0376] Let the set of detection points be:

[0377]

[0378] In the above formula, For fusion observation distance (m); This is the observation rate after consistency correction; Corresponding energy characteristics;

[0379] Redundancy detection is performed using a density clustering method based on spatial velocity similarity.

[0380]

[0381] In the above formula, Δ r 0, Δ v 0 represents the normalized scaling parameter; This is the clustering threshold used to determine whether objects belong to the same physical target cluster.

[0382] The energy-weighted merging of intra-cluster clicks is as follows:

[0383]

[0384] Obtain a unique target representation To achieve redundant detection elimination and feature aggregation;

[0385] The temporal smoothing and continuity include:

[0386] Observation sequences after spatial redundancy suppression Global smoothing is performed using a recursive Kalman filter, and the global state vector is defined as follows:

[0387]

[0388] In the above formula, The target distance is estimated globally. The target velocity is estimated globally; This is the global state vector;

[0389] The state and observation model is as follows:

[0390]

[0391] In the above formula, This is the global state transition matrix. T s The sampling period; The observation matrix; This refers to the process noise term; For the observed noise term;

[0392] The update equation for global smoothing using Kalman filtering recursively is:

[0393]

[0394] In the above formula, The Kalman gain matrix is ​​adaptively updated based on the error covariance. This represents the smoothed global trajectory state.

[0395] The trajectory continuity index is defined as:

[0396]

[0397] In the above formula, To prevent small constants with a denominator of zero, when At that time, it indicates that the global trajectory smoothness and consistency are good;

[0398] The final recognition result output includes:

[0399] After being constrained by the multiple physical consistency rules, the final identification status of the target is as follows:

[0400]

[0401] In the above formula, This is the final estimated target distance; The final estimated target speed; To integrate energy characteristics; As an indicator of trajectory continuity;

[0402] The final identification result comprehensively represents the target's spatial location, motion state, energy intensity, and temporal stability, and can be directly used for target classification, trajectory monitoring, and stability assessment.

[0403] In summary, the multi-domain interference feature extraction and identification method for millimeter-wave radar in coal mine roadways proposed in this application can effectively address multiple interference sources in the complex environment of coal mine roadways. The coal mine roadway environment is complex, with common interferences including coal dust, moisture, gas, and metal reflection. Traditional signal processing methods cannot effectively extract target features in such an environment. This application, by combining multiple signal processing methods such as power spectral density (PSD) analysis, Doppler-time domain (DT) analysis, range-time domain (RT) analysis, and range-Doppler domain (RD) analysis, can effectively distinguish between target signals and interference signals, effectively address the complex interference environment in coal mine roadways, significantly improve the target identification capability and robustness of millimeter-wave radar systems in complex environments, and provide a new solution for safety monitoring in coal mine roadways, with broad application prospects and high practicality.

[0404] Although preferred embodiments of the present application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the embodiments of the present application.

[0405] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device 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 terminal device. 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 terminal device that includes said element.

[0406] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims. All of these forms are within the protection scope of this application.

Claims

1. A method for multi-domain interference identification using millimeter-wave radar in complex environments of coal mine roadways, characterized in that, include: The echo signal data of the tunnel is obtained by using a continuous wave millimeter wave radar deployed in the underground roadway to collect the echo signal. Based on the echo signal data, feature extraction and interference suppression are carried out sequentially in the power spectral density domain, range-time domain, Doppler time domain, and range-Doppler domain to construct a multi-domain link feature consisting of spectrum cleaning, energy compensation, dynamic suppression, and trajectory separation. The feature matrices corresponding to the multi-domain link features are uniformly represented, and the features of different domains are mapped to a unified representation space, so that the feature matrices are comparable and fusionable in terms of scale, dimension and physical meaning. Then, the features of each domain are weighted and fused to form a unified feature vector after fusion, which is used for target recognition and interference separation. Based on the unified feature vector, multiple physical consistency constraints based on motion laws and spatial structures are introduced to output the final recognition result, so that the recognition result has physical rationality and temporal continuity.

2. The millimeter-wave radar multi-domain interference identification method for complex environments in coal mine roadways according to claim 1, characterized in that, By using continuous wave millimeter-wave radar deployed in underground roadways to collect echo signals, the echo signal data of the roadways is obtained, including: The continuous-wave millimeter-wave radar is deployed in the underground tunnels to transmit continuous frequency-modulated signals for target detection and to receive echo signals reflected back from the targets. The frequency range of the frequency-modulated signals is [missing information]. f 0, the frequency modulation slope is Δ f The duration is T ; The echo signal Echo ( t ) is represented as: In the above formula, A is the amplitude of the echo signal. f 0 is the center frequency, Δ f For frequency modulation slope, T For duration, t For time variables, j The imaginary unit; The echo signal is preprocessed, which includes removing background noise and DC component to improve the signal-to-noise ratio of the echo signal, and the echo signal is filtered and denoised to obtain the echo signal data. The echo signal data is stored in time series to provide a data foundation for subsequent data analysis and feature extraction.

3. The millimeter-wave radar multi-domain interference identification method for complex environments in coal mine roadways according to claim 1, characterized in that, Based on the echo signal data, feature extraction and interference suppression are performed sequentially in the power spectral density domain, range-time domain, Doppler time domain, and range-Doppler domain, including: The echo signal data is subjected to spectral feature extraction and electromagnetic interference anomaly identification in the power spectral density domain in order to identify and suppress spectral anomalies of electromagnetic interference. Energy feature extraction and dust scattering anomaly identification are performed on the echo signal data processed in the power spectral density domain in the distance-time domain, so as to extract the energy distribution law of dust scattering and realize energy compensation. Dynamic feature extraction and dust disturbance anomaly identification are performed on the echo signal data processed in the distance-time domain in the Doppler time domain, so as to identify and suppress the dynamic fluctuation characteristics of dust disturbance. In the range-Doppler domain, feature trajectory feature extraction and multipath reflection anomaly identification are performed on the echo signal data after Doppler time-domain processing to separate the sub-trajectory structure of multipath reflection and enhance the target response.

4. The millimeter-wave radar multi-domain interference identification method for complex environments in coal mine roadways according to claim 3, characterized in that, The spectral feature extraction and electromagnetic interference anomaly identification of the echo signal data in the power spectral density domain include: First, power spectrum feature extraction is performed, and then a Fast Fourier Transform is applied to the echo signal data to calculate the echo signal at various frequencies. f The power spectral density PSD(f) is expressed as follows: In the above formula, The received echo signal, The Fourier transform operator is used to describe the distribution of electromagnetic energy in the frequency dimension. Next, electromagnetic interference anomaly identification is performed to identify abnormal frequency bands and calculate the local mean of the power spectrum. with standard deviation And set the adaptive threshold Tf as follows: In the above formula, k For adjustment coefficients, when When, determine the frequency f Anomalies are identified by their frequency, forming anomaly sets. ; Calculate anomalous energy And count the percentage of abnormal frequencies. This describes the intensity and range of electromagnetic interference, identifies the concentrated frequency bands of electromagnetic interference through threshold detection, and provides a location basis for subsequent filtering and clearing. Then perform spectrum compensation and feature output. To remove abnormal frequency bands, a band-stop filter is used to perform band-stop filtering on the abnormal frequency points. The expression is as follows: In the above formula, F s Sampling frequency, r This is the stopband sharpness factor. For abnormal frequencies, the purified signal after filtering to remove electromagnetic interference is expressed as follows: Based on the expression of the purified signal after electromagnetic interference removal and after normalizing the whitening spectrum, the following formula is obtained: The purified signal after electromagnetic interference removal suppresses the electromagnetic narrowband peak, which serves as the input signal in the subsequent distance-time domain, while simultaneously extracting electromagnetic interference features. The purified signal after electromagnetic interference removal and the electromagnetic interference features are both used as input signals for weighted fusion of multi-domain features.

5. The millimeter-wave radar multi-domain interference identification method for complex environments in coal mine roadways according to claim 4, characterized in that, The energy feature extraction and dust scattering anomaly identification of the echo signal data processed in the power spectral density domain in the distance-time domain include: First, distance-time energy features are extracted to analyze the energy distribution of the signal in the distance and time dimensions, and then applied to the purified signal after electromagnetic interference removal. Performing a distance-dimensional Fourier transform, the energy matrix is ​​expressed as follows: In the above formula, r is the distance variable. This represents the Fourier transform along the distance dimension; Next, wavelets are used to decompose the signal into low-frequency background. With high frequency components The low-frequency background This refers to the energy scattered by dust, specifically the high-frequency components. This is the target signal, and its expression is as follows: Next, anomaly identification of dust scattering is performed, and the statistical energy change law along the distance direction is used to fit the dust attenuation model as follows: In the above formula, Let I0 be the average energy at a distance r, I0 be the initial energy, and α be the dust attenuation coefficient. This indicates an abnormal dust concentration or attenuation. Simultaneously define the dust energy ratio DER To measure the proportion of dust scattering energy in the total signal: The dust energy ratio (DER) is used to identify and quantify the masking and attenuation effects of dust on signal energy. Then, dust energy compensation and characteristic output are performed to execute low-frequency energy subtraction and normalization compensation, thereby obtaining the energy compensation matrix. Its expression is as follows: In the above formula, To prevent division by zero of small constants; The contrast enhancement rate was calculated after final compensation. CIR The expression is as follows: Obtain the energy compensation matrix To enhance target echo contrast, the energy compensation matrix serves as the input signal in the Doppler time domain, used to detect dynamic disturbances and simultaneously extract dust features. It serves as the input signal for weighted fusion of multi-domain features.

6. The millimeter-wave radar multi-domain interference identification method for complex environments in coal mine roadways according to claim 5, characterized in that, Dynamic feature extraction and dust disturbance anomaly identification are performed on the echo signal data processed in the distance-time domain in the Doppler time domain, including: First, Doppler time dynamic features are extracted to further identify the time-varying characteristics of target motion and dust disturbance. Then, time-Doppler analysis is performed on the signal after dust energy compensation. This process uses short-time Fourier transform to achieve local spectral expansion in the time domain, thereby obtaining the joint distribution of signal energy in time and Doppler frequency shift, expressed as follows: In the above formula, This indicates performing a local Fourier transform on the signal within a time window, where t is time. The Doppler frequency shift reflects the radial velocity change of the scattering body; The target radial velocity is calculated using the Doppler frequency shift, and its expression is as follows: In the above formula, f 0 is the center frequency. c Assuming the speed of light, through this transformation, the signal is mapped onto the time-Doppler plane. If a continuously moving target presents as a time-continuous, frequency-narrow band of energy ridges, it indicates smooth velocity changes and stable reflection; dust disturbances, on the other hand, form short-duration, broadband, and randomly distributed energy patches, indicating that the reflector particles are small in size and move randomly. Next, dust disturbance anomaly identification is performed to quantitatively distinguish whether there is dust disturbance anomaly. Three types of dynamic criteria are defined: trajectory continuity index, local bandwidth index, and energy stability index. Among them, the trajectory continuity index for: In the above formula, L c The continuous length of the Doppler ridge line. T For analysis of time windows; when R c →1, the target trajectory is continuous, indicating stable motion; when R c 1 indicates energy breakage, signifying the presence of dust disturbance; The local bandwidth index for: In the above formula, and The positions of the 95% and 5% frequencies of the cumulative energy distribution at time t represent the points where the target's motion energy is concentrated. Very small; energy diffuses when dust is disturbed. Significantly increased; The energy stability index for: In the above formula, and Let represent the time variance and mean, respectively. When the signal energy fluctuates drastically over time... An increase indicates frequent short-term disturbances; Combining the trajectory continuity index, the local bandwidth index, and the energy stability index, the disturbance criterion is established as follows: if or or This area is then identified as a dust disturbance zone. β and γ are empirical thresholds, calibrated from measured samples. Continuous targets have high continuity, narrow bandwidth, and stable energy characteristics; dust disturbances have low continuity, wide bandwidth, and random energy characteristics. Then, dynamic perturbation suppression and feature output are performed to adaptively suppress the identified dust perturbation area. The soft mask function is constructed as follows: In the above formula, σ( () is the Sigmoid function. a 1, a 2, a 3 is an adjustment coefficient used to balance the effects of continuity and diffusion; The mask function is applied for spectral suppression, and its expression is as follows: when The target trajectory is preserved during the process; when Energy to suppress dust; To eliminate short-term jumps in the target velocity, Kalman filtering is applied to the frequency shift of the main ridge line, and its expression is as follows: In the above formula, This is the current estimated frequency; For predicting frequencies; For observation frequency; Kalman gain; H The observation matrix; The smoothed velocity reflects the true motion of the target without being affected by random disturbances, and is expressed as: To quantify the inhibition effect, a dynamic residual energy ratio is introduced: In the above formula, For the sum of the absolute values ​​of matrices, when A lower value indicates that the dust disturbance energy has been effectively weakened; The final output features are dynamic features. The transplanted spectrum is the input signal used for weighted fusion of multi-domain features. The signal is transmitted to the distance Doppler domain to further enable multipath reflection anomaly identification.

7. The millimeter-wave radar multi-domain interference identification method for complex environments in coal mine roadways according to claim 6, characterized in that, The process of extracting feature trajectories and identifying multipath reflection anomalies from the Doppler-time-domain processed echo signal data in the range-Doppler domain includes: If the echo signal has a multipath reflection secondary trajectory, in order to distinguish the real target main trajectory from the multipath secondary trajectory, feature trajectory extraction and multipath reflection anomaly identification are performed in the range Doppler domain. First, distance Doppler trajectory features are extracted. To simultaneously reflect the spatial distance distribution and velocity characteristics of the target, the input spectrum... Performing a two-dimensional fast Fourier transform yields the range-Doppler energy spectrum, which is expressed as follows: In the above formula, FFTr( ) performs a one-dimensional FFT along the distance dimension to obtain the distance gate energy distribution; FFTt( This involves performing a one-dimensional FFT along the time dimension to extract the Doppler frequency shift. r Displacement distance variable; f D The Doppler frequency shift is related to the target radial velocity v as follows: c The speed of light; f 0 represents the radar carrier frequency; A spectrum reflecting the energy distribution of a target in the range-velocity space; On the plane of the distance-Doppler domain, the energy distribution of a real target typically forms a single main ridge; if there is reflection from a wall or device, a secondary trajectory parallel to or offset from the main ridge will appear. The energy distribution spectrum visually reveals the spatial motion coupling characteristics of the target echo. Next, multipath reflection anomaly identification is performed to identify reflection signals caused by walls, ground, or equipment by quantifying the offset of the primary and secondary trajectories in terms of distance and Doppler values. Specifically, this includes: First, extract the peak coordinates of the main trajectory and the secondary trajectory, and calculate their distance and frequency offset: In the above formula, , The peak position corresponding to the main trajectory; , If the peak position corresponding to the trajectory, or The spatial or velocity characteristics of the secondary trajectory are significantly different from those of the primary trajectory, indicating multipath reflection interference. and These are the range and the Doppler threshold, respectively. These values ​​are determined by the system's geometric layout and empirical data. Multipath secondary trajectories usually originate from the secondary reflection of radar waves through walls or mechanical structures. Their propagation paths are relatively long, leading to increased range. At the same time, phase delay causes a pseudo-velocity component. Therefore, multipath identification is achieved through joint offset detection. Next, multipath separation and feature output are performed to separate the real target from the multipath reflection components in the mixed signal, so as to preserve the main trajectory and achieve robust target representation. The observed range Doppler signal is regarded as the superposition of multiple scattering sources through the multipath mixing model, and its expression is as follows: In the above formula, X The observed signal matrix is ​​obtained by the radar receiver on the range-velocity plane; , where is the source signal vector, corresponding to the main target echo and the multipath reflection echo, respectively; A This is a hybrid matrix that describes the propagation attenuation and phase shift along different paths; N Noise term; The main signal was isolated by independent component analysis. and secondary signals : In the above formula, These represent the separated main trajectory signal and sub-trajectory signal, respectively. The main trajectory signal... Used for trajectory smoothing and feature extraction, sub-trajectory signal Used for multipath quantization; To smooth the main trajectory state and constrain continuity, a recursive Kalman filter is used for temporal smoothing to address velocity abrupt changes caused by multipath propagation. The expression for this filter is as follows: In the above formula, r k For distance, v k For speed; The state and observation equations are defined as follows: The following equation is obtained by recursively updating using Kalman filtering: In the above formula, F Represented as a state transition matrix, it is determined by the sampling period and the motion model; H The observation matrix; It is by Extracted observation vectors; The Kalman gain is adaptively updated based on the error covariance. After filtering, a smooth trajectory sequence is obtained. And calculate the continuity index, the expression of which is: In the above formula, ε To prevent small constants with zero denominators, when At that time, the trajectory continuity was good, indicating that the main trajectory was stable and there was no multipath disturbance; The Kalman filter is used for temporal smoothing of the local main trajectory, while a global Kalman consistency constraint is introduced at the fusion level. The final output is the distance-Doppler domain feature vector: The range-Doppler domain feature vector serves as the input signal for the multi-domain feature weighted fusion module. Together with the power spectral density domain, the range-time domain, and the Doppler time domain output, they construct a unified representation space, thereby achieving highly robust identification of targets in the roadway.

8. The millimeter-wave radar multi-domain interference identification method for complex environments in coal mine roadways according to claim 7, characterized in that, Features extracted from different domains are uniformly projected onto a representation space. The feature matrices are comparable and fusionable in terms of scale, dimension, and physical meaning. Then, the features from each domain are weighted and fused to form a unified feature vector, including: The features from each domain are weighted and fused to construct a multi-domain fusion deep recognition framework, forming a unified feature vector for target recognition and interference separation. Specifically, this includes: First, multi-domain features are summarized, and the outputs from each domain are combined to obtain four types of feature vectors, whose expressions are as follows: The four types of feature vectors correspond to electromagnetic interference spectrum, dust energy distribution, dynamic disturbance and multipath trajectory characteristics, respectively, and together reflect the main interference and motion elements of millimeter-wave radar signals in the complex environment of coal mine roadways. Next, feature standardization and representation space mapping are performed. To eliminate the differences in characteristic units and dynamic ranges between different domains, it is necessary to use the standardization operator Φ( Standardize the scale: In the above formula, μ i , σ i The first i The mean and standard deviation of the domain features. These are the normalized feature vectors; Subsequently, canonical correlation analysis was used to project the multi-domain features onto a unified representation space: In the above formula, W i For the projection matrices of each domain, maximize cross-domain correlation. Learned; To maintain cross-domain comparability and physical consistency in the mapped feature representation; Next, information gain weights are calculated to quantify the contribution of each domain feature to target recognition, and the information gain is calculated as follows: In the above formula, Represents the target label entropy; Let the target entropy be given the features of the i-th domain. Information gain for the feature domain; The weights of each domain are then determined based on the information gain as follows: Next, multi-domain feature weighted fusion is performed to weight and fuse the projected features from each domain according to weight coefficients, forming a unified representation feature vector: After fusion, the overall characteristic energy index is calculated as follows: In the above formula, Energy intensity and fusion confidence level used to reflect overall characteristics; Fusion feature vectors As the output of multi-domain fusion recognition, it constitutes the input features for subsequent physical consistency constraints; Finally, key physical features are extracted and observation variables are generated to extract features with clear physical meaning from the multi-domain representation matrix. The main peak position, energy centroid, or main trajectory center of each domain matrix corresponds to the main features of distance, velocity, and energy, respectively. In the above formula, Mi( ) is the first i The energy matrix of the domain; Multi-domain consistent representation is achieved by normalizing eigenvalues ​​and weighting them according to information gain, and its expression is as follows: In the above formula, , The normalized distance and velocity characteristics, The energy of the main peak in each domain is used; after weighting, three physical characteristics are obtained: fusion distance, fusion speed, and fusion energy. The final fusion result is defined as the observed input: The observed input is used for physical consistency constraints and trajectory recognition optimization.

9. The millimeter-wave radar multi-domain interference identification method for complex environments in coal mine roadways according to claim 1, characterized in that, The multiple physical consistency between the motion laws and spatial structure includes: velocity consistency verification, spatial redundancy suppression, temporal smoothing, and continuity; based on the unified feature vector, constraints based on the multiple physical consistency between the motion laws and spatial structure are introduced, including: The speed consistency check includes: Extract the instantaneous Doppler main peak frequency of the target's main trajectory from the spectrum of the distance Doppler domain. f D,k The instantaneous Doppler main peak frequency f D,k The peak position of the energy distribution in the Doppler domain spectrum is taken from the distance, and its calculation method is as follows: In the above formula, The distance-Doppler spectrum obtained by Fourier transform; r k The distance position corresponds to the main trajectory of the current frame; the instantaneous Doppler main peak frequency reflects the instantaneous velocity of the target; For sequence f D,k A smoothing correction is performed to avoid pseudo-frequency fluctuations caused by noise spikes; its expression is as follows: In the above formula, T f The frequency variation threshold is set by the radar system bandwidth and sampling rate, and is corrected accordingly. Used for subsequent speed consistency calculations; According to the radar Doppler effect, instantaneous radial velocity v k Calculated from the frequency components of the main trajectory: In the above formula, To observe radial velocity; c The speed of light; The rate of change of velocity between adjacent frames is defined as follows: If Δ v k > T v If the velocity change does not conform to the law of continuous motion, a weighted correction is applied as follows: In the above formula For smoothing coefficients, T v The speed jump threshold; Revised This serves as a spatial redundancy and temporal smoothing and continuity process following the observation velocity input, ensuring that the main trajectory velocity sequence is continuous and stable in the time dimension; The spatial redundancy suppression includes: Clustering and redundancy reduction are performed on the observation set after the velocity consistency test, which specifically includes: Let the set of detection points be: In the above formula, For fusion observation distance (m); This is the observation rate after consistency correction; Corresponding energy characteristics; Redundancy detection is performed using a density clustering method based on spatial velocity similarity. In the above formula, Δ r 0, Δ v 0 represents the normalized scaling parameter; This is the clustering threshold used to determine whether objects belong to the same physical target cluster. The energy-weighted merging of intra-cluster clicks is as follows: Obtain a unique target representation To achieve redundant detection elimination and feature aggregation; The temporal smoothing and continuity include: Observation sequences after spatial redundancy suppression Global smoothing is performed using a recursive Kalman filter, and the global state vector is defined as follows: In the above formula, The target distance is estimated globally. The target velocity is estimated globally; This is the global state vector; The state and observation model is as follows: In the above formula, This is the global state transition matrix. T s The sampling period; The observation matrix; This refers to the process noise term; For the observed noise term; The update equation for global smoothing using Kalman filtering recursively is: In the above formula, The Kalman gain matrix is ​​adaptively updated based on the error covariance. This represents the smoothed global trajectory state. The trajectory continuity index is defined as: In the above formula, To prevent small constants with zero denominators, when At that time, it indicates that the global trajectory smoothness and consistency are good; The final recognition result output includes: After being constrained by the multiple physical consistency rules, the final identification status of the target is as follows: In the above formula, This is the final estimated target distance; The final estimated target speed; To integrate energy characteristics; As an indicator of trajectory continuity; The final identification result comprehensively represents the target's spatial location, motion state, energy intensity, and temporal stability, and can be directly used for target classification, trajectory monitoring, and stability assessment.

10. A millimeter-wave radar multi-domain interference identification device for complex environments in coal mine roadways, characterized in that, include: The acquisition module is used to acquire echo signals using a continuous wave millimeter wave radar deployed in the underground roadway, and obtain echo signal data of the roadway. The extraction module is used to perform feature extraction and interference suppression in the power spectral density domain, range-time domain, Doppler time domain, and range-Doppler domain according to the echo signal data, and to construct a multi-domain link feature consisting of spectrum cleaning, energy compensation, dynamic suppression and trajectory separation. The unified module is used to perform unified representation processing on the feature matrix corresponding to the multi-domain link features, map the features of different domains to a unified representation space, so that the feature matrix is ​​comparable and fusionable in scale, dimension and physical meaning, and then perform weighted fusion of the features of each domain to form a fused unified feature vector for target recognition and interference separation. The constraint recognition module is used to introduce multiple physical consistency constraints based on motion laws and spatial structures based on the unified feature vector and output the final recognition result, so that the recognition result has physical rationality and temporal continuity.

Images