A radar processing method and system for moving target phase-coherent accumulation and doppler domain detection

Abstract

The application provides a radar processing method and system for moving target coherent accumulation and Doppler domain detection, and relates to the technical field of radar signal processing. Raw baseband echo data streams containing multiple pulse repetition period digital sampling complex signals collected by a radar radio frequency front end are received, and pulse dimension distance processing is performed to generate a distance pulse two-dimensional data matrix. Frame division operation is performed on the distance pulse two-dimensional data matrix to obtain multiple coherent processing frame units and assign frame period labels. Doppler spectrum splitting and sub-band branch filter processing are performed on each frame to generate a slow time dimension Doppler sub-band decomposition data cube. Frame-to-frame multi-dimensional phase alignment operation is performed on each sub-band channel by using a Doppler sub-band cross-period phase compensation function to obtain a cross-frame coherent accumulation data set. Finally, two-dimensional Doppler domain detection processing based on a constant false alarm criterion is implemented to generate a point track detection output result. The application improves the moving target detection performance through sub-band level coherent accumulation and two-dimensional joint detection.

Description

Technical Field

[0001] This invention relates to the field of radar signal processing and moving target detection technology, and more specifically, to a radar processing method and system for moving target coherent accumulation and Doppler domain detection. Background Technology

[0002] Moving target detection is one of the core functions of radar systems to effectively sense and accurately locate moving targets. In the radar signal processing flow, coherent accumulation technology enhances target signal energy and suppresses clutter and noise interference by coherently superimposing echo signals within multiple pulse repetition periods, utilizing the Doppler frequency shift characteristics generated by moving targets in the slow time dimension. It is a key technical link in realizing long-range weak target detection. Traditional moving target coherent accumulation methods usually perform a complete Doppler Fast Fourier Transform on the slow time dimension based on the range pulse two-dimensional data matrix to obtain the Doppler spectrum and perform target detection in the frequency domain. However, in actual radar application scenarios, the motion state of moving targets often exhibits significant time-varying characteristics and non-stationarity. For example, when a target accelerates, decelerates, turns, or undergoes maneuvering, its Doppler frequency will rapidly drift and broaden within the continuous observation time. This causes the traditional coherent accumulation method based on full-frame unified processing to suffer a sharp decline in detection performance when facing large maneuvering targets or long accumulation time scenarios due to the severe loss of coherent accumulation gain.

[0003] In existing technologies, common strategies for addressing the Doppler frequency broadening problem of moving targets include incoherent splicing after segmented short-time coherent accumulation, or characterizing non-stationary signals using time-frequency analysis methods such as key transforms. However, while segmented short-time coherent accumulation can mitigate the impact of frequency broadening to some extent, it essentially comes at the cost of sacrificing coherent accumulation gain, and lacks an effective mechanism for maintaining phase correlation between segments, making it difficult to achieve stable detection capabilities over long observation timescales. While time-frequency analysis methods based on key transforms possess some ability to characterize non-stationary signals, they typically treat signal processing as a single-dimensional time-frequency mapping problem, failing to systematically decompose and reorganize the multidimensional structural characteristics of radar data. Therefore, they cannot achieve refined decomposition and sub-band-level processing of the Doppler spectrum while maintaining coherence. Furthermore, existing technologies often employ a single-dimensional amplitude threshold decision method in the Doppler domain detection stage, lacking a two-dimensional joint detection mechanism based on the constant false alarm rate criterion. This can easily lead to uncontrolled false alarm rates or insufficient detection probability in complex clutter backgrounds.

[0004] Furthermore, while some existing technologies attempt to perform phase compensation between frames to maintain cross-frame coherence, their compensation models are usually based on a single uniform motion assumption or a simple linear phase model, which is difficult to adapt to the complex motion state changes experienced by the moving target between multiple frames. Moreover, they do not consider the strategy of parallel processing of continuous pulse sequences according to Doppler subbands before implementing inter-frame alignment, thus failing to achieve fine-grained cross-frame phase compensation and efficient accumulation of coherent energy at the subband level. Summary of the Invention

[0005] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a radar processing method for moving target coherent accumulation and Doppler domain detection, the method comprising: The system receives raw baseband echo data streams acquired by the radar radio frequency front-end, wherein the raw baseband echo data streams contain digitally sampled complex signals with multiple pulse repetition cycles. The original baseband echo data stream is processed by pulse dimension distance processing to generate a two-dimensional distance pulse data matrix containing fast time dimension distance cells and slow time dimension pulse numbers; A frame partitioning operation is performed on the distance pulse two-dimensional data matrix to obtain multi-frame coherent processing frame units with continuous pulse sequence relationships, and a frame period label is assigned to each frame coherent processing frame unit. Each frame of coherent processing frame unit is subjected to Doppler spectrum fragmentation and subband splitting filtering to generate a slow-time dimension Doppler subband decomposition data cube. The slow-time dimension Doppler subband decomposition data cube divides continuous pulse segments into multiple parallel subband channels in the slow-time dimension. Using a preset Doppler subband cross-cycle phase compensation function, an inter-frame multi-dimensional phase alignment operation is performed on each subband channel of the slow-time dimension Doppler subband decomposition data cube to obtain an inter-frame aligned cross-frame coherent accumulation data set. The cross-frame coherent accumulation data set is subjected to two-dimensional Doppler domain detection processing based on the constant false alarm rate criterion to generate a point detection output result containing the distance coordinates and Doppler frequency coordinates of the moving target point.

[0006] Furthermore, embodiments of the present invention also provide a radar processing system for moving target coherent accumulation and Doppler domain detection, characterized in that it includes: A processor; a machine-readable storage medium for storing machine-executable instructions of the processor; wherein the processor is configured to perform the aforementioned radar processing method for moving target coherent accumulation and Doppler domain detection by executing the machine-executable instructions.

[0007] In another aspect, embodiments of the present invention also provide a computer program product, the computer program product including machine-executable instructions, the machine-executable instructions being stored in a computer-readable storage medium, a processor of a computer device reading the machine-executable instructions from the computer-readable storage medium, the processor executing the machine-executable instructions, causing the computer device to execute the aforementioned radar processing method for moving target coherent accumulation and Doppler domain detection.

[0008] Based on the above, by performing frame partitioning on the two-dimensional data matrix of distance pulses and assigning a frame period label to each frame, the continuous pulse sequence relationship is structured into multi-frame coherent processing frame units with clear time assignments. On this basis, Doppler spectrum fragmentation and subband splitting filtering are performed on each frame coherent processing frame unit, generating a Doppler subband decomposition data cube that divides continuous pulse segments into multiple parallel subband channels in the slow time dimension. This decomposes the originally coupled broadband Doppler spectrum into multiple narrowband subband channels for parallel processing. Furthermore, a preset Doppler subband cross-period phase compensation function is used to perform frame processing on each subband channel. Multi-dimensional phase alignment is performed to finely compensate for cross-frame phase drift at sub-band level accuracy, ensuring that high-quality coherent accumulation conditions are maintained in each sub-band channel between frames. This results in coherent gain far exceeding that of traditional short-time accumulation over long observation timescales. Finally, two-dimensional Doppler domain detection processing based on the constant false alarm rate criterion is implemented on the cross-frame coherent accumulation data set. The range dimension coordinate and the Doppler frequency dimension coordinate are jointly incorporated into the detection decision framework. While effectively controlling the false alarm rate, the two-dimensional information is fully utilized to improve detection sensitivity and positioning accuracy, achieving maximum preservation of coherent accumulation gain and significant improvement in detection performance under non-stationary motion conditions of moving targets. Attached Figure Description

[0009] Figure 1 This is a schematic diagram of the execution flow of the radar processing method for moving target coherent accumulation and Doppler domain detection provided in the embodiments of the present invention.

[0010] Figure 2 This is a schematic diagram of exemplary hardware and software components of a radar processing system for moving target coherent accumulation and Doppler domain detection provided in an embodiment of the present invention. Detailed Implementation

[0011] Figure 1This is a flowchart illustrating a radar processing method for coherent accumulation and Doppler domain detection of moving targets, provided in one embodiment of the present invention. This method operates on a computing device coupled to the radar receiver signal processing link. This embodiment uses an airborne pulse-Doppler radar detection scenario of a slowly moving ground vehicle target as an example to provide a unified explanation of each step. In this application scenario, the airborne radar platform flies along a straight track at cruising speed, with the radar antenna beam pointing towards the area in front of the platform's direction of motion, transmitting a coherent pulse train signal to the ground. The slowly moving ground vehicle target moves within the radar beam's illumination range at a radial velocity component much lower than the platform's speed. The radar receiver receives a mixed echo signal containing ground clutter, vehicle target reflected echoes, and receiver thermal noise. After orthogonal down-conversion and analog-to-digital conversion, the raw baseband echo data stream is formed. Data acquisition authorization has been obtained before acquiring the radar echo data, and legal and compliant requirements have been followed.

[0012] Step S110: Receive the raw baseband echo data stream acquired by the radar radio frequency front end, wherein the raw baseband echo data stream contains digitally sampled complex signals with multiple pulse repetition cycles.

[0013] The radar RF front-end performs quadrature down-conversion processing on the echo signal in each pulse repetition cycle to obtain in-phase and quadrature components. The analog-to-digital converter samples these component pairs using a sampling clock, forming a digital sampled complex signal. In the p-th pulse repetition cycle, a discrete complex sequence of length L sampling points is obtained, where the complex sample value of the l-th sampling point is represented as x(p, l) = I(p, l) + j × Q(p, l), where j is the imaginary unit. All discrete complex sequences from the P pulse repetition cycles are arranged sequentially according to the pulse transmission time sequence, forming the original baseband echo data stream D = {x(p, l) | p = 0, 1, ..., P-1; l = 0, 1, ..., L-1}. This original baseband echo data stream is transmitted sequentially to the signal processing link in the form of a one-dimensional data stream. In the airborne radar ground slow-speed vehicle detection scenario of this embodiment, the number of pulse repetition cycles P is set to the order of several thousand, and the number of sampling points L in each pulse repetition cycle is determined by the radar transmission signal bandwidth and the range observation range. Ground stationary clutter exhibits strong correlation during adjacent pulse repetition cycles, while the echo phase of slowly moving vehicle targets shows a gradual change during adjacent pulse repetition cycles due to the presence of radial velocity components.

[0014] Step S120: Perform pulse-dimensional distance processing on the original baseband echo data stream to generate a two-dimensional distance pulse data matrix containing fast time-dimensional distance units and slow time-dimensional pulse numbers.

[0015] Step S1201: Extract the discrete sampling sequence within a single pulse period of the original baseband echo data stream, and perform pulse compression matched filtering on the discrete sampling sequence to obtain a high-resolution range-oriented complex amplitude sequence for a single pulse period.

[0016] Extract the discrete sampling sequence xp=[x(p, 0), x(p, 1), ..., x(p, L-1)]T corresponding to the p-th pulse repetition period from the original baseband echo data stream D. Perform pulse compression matched filtering on this discrete sampling sequence xp. The pulse compression matched filtering adopts the frequency domain matched filtering method. Transform the discrete sampling sequence xp to the frequency domain through Fast Fourier Transform to obtain the frequency domain complex sequence Xp=FFT(xp). Multiply the conjugate frequency domain copy Hmatch of the radar transmitted signal with Xp point by point to obtain the matched filtered frequency domain sequence Yp(f)=Xp(f)×conj(Hmatch(f)), where conj represents the complex conjugate operation. Perform inverse Fast Fourier Transform on Yp to obtain the high-resolution range-axis complex amplitude sequence yp=IFFT(Yp)=[yp(0), yp(1), ..., yp(L-1)]T for this pulse repetition period. The l-th element yp(l) in the high-resolution range-oriented complex amplitude sequence yp represents the target reflectivity complex amplitude value at the l-th fast time range unit. Its magnitude |yp(l)| reflects the target scattering intensity at that range unit, and its argument arg(yp(l)) reflects the target echo phase at that range unit.

[0017] Step S1202: Arrange the high-resolution range-reverse amplitude sequence of all pulse cycles according to the pulse triggering time sequence, and construct an initial two-dimensional range pulse arrangement array with the fast time dimension sampling index as the row and the slow time dimension pulse number as the column.

[0018] Perform the pulse compression matched filtering process described above on all P pulse repetition cycles to obtain P high-resolution range-axis complex amplitude sequences {yp|p=0,1,...,P-1}. Arrange these high-resolution range-axis complex amplitude sequences according to the pulse emission sequence to construct an initial two-dimensional range pulse array M0. M0 is an L-row, P-column complex matrix. The element in the l-th row and p-th column of M0 is M0(l,p)=yp(l). The rows of matrix M0 correspond to the fast time dimension, and each row represents the complex amplitude value of a fixed range unit in different pulse repetition cycles, called the slow time sequence of that range unit. The columns of matrix M0 correspond to the slow time dimension, and each column represents the complex amplitude values ​​of all range units within a fixed pulse repetition cycle, called the range profile of that pulse repetition cycle.

[0019] Step S1203: Perform pulse repetition period time-varying deviation correction on the initial range pulse two-dimensional array in the slow time dimension to eliminate the slow time dimension sampling position offset accumulated due to radar transmission clock jitter, and generate a slow time dimension uniform array.

[0020] Radar clock jitter causes a small random offset between the actual transmission time of each pulse repetition cycle and the ideal uniform time grid. This random offset accumulates in the slow time dimension, resulting in a sampling position offset. A pulse repetition cycle time-varying deviation correction is applied to the initial range pulse two-dimensional array M0. The actual transmission timestamp tp, p=0,1,...,P-1, for each pulse repetition cycle recorded by the radar transmitter is obtained. Using the transmission time t0 of the first pulse repetition cycle as the reference time origin, the ideal transmission time for each pulse repetition cycle is calculated as tpideal=p×TPRI, where TPRI is the nominal pulse repetition interval. The time deviation between the actual transmission time and the ideal transmission time is Δtp=tp-tpideal. For the l-th range cell, its slow time sequence is M0(l,p), p=0,1,...,P-1. The corresponding time sampling point for this slow time sequence is tp=t0+p×TPRI+Δtp. The sinc interpolation kernel function is used to reconstruct this non-uniform sampling sequence onto a uniform time grid. The uniform time grid is defined as tquni = t0 + q × TPRI, where q = 0, 1, ..., P-1. The sinc interpolation reconstruction is calculated as follows: for each uniform time grid point q, the reconstructed value Muni(l, q) = ∑(p = 0 to P-1)M0(l, p) × sinc((tquni - tp) / TPRI), where sinc(x) = sin(π × x) / (π × x). This interpolation reconstruction process is performed on all L distance cells one by one to generate a slow-time dimension uniformly arranged array Muni. Muni has a size of L rows and P columns, and the slow-time dimension sampling positions have been corrected to the uniform time grid.

[0021] Step S1204: Extract complex sequences from each fast time dimension distance cell in the slow time dimension uniform arrangement array along the slow time dimension direction, and apply window function weighting operation to suppress slow time dimension sidelobes to obtain a slow time dimension windowed complex sequence.

[0022] For each row of the slow-time dimension uniformly arranged array Muni, i.e., the slow-time sequence Muni(l, q), q=0,1,...,P-1 of the l-th distance unit, a window function weighting operation is applied. The window function chosen is either a Chebyshev window or a Taylor window. The discrete-time expression of the Chebyshev window is derived from the Chebyshev polynomial, which exhibits equal sidelobe levels under a given main lobe width constraint. The window function sequence is denoted as w(q), q=0,1,...,P-1. The slow-time dimension windowed complex sequence is Mwin(l, q)=Muni(l, q)×w(q), q=0,1,...,P-1. The purpose of the window function weighting operation is to suppress the slow-time dimension frequency sidelobes generated by the finite number of pulses during subsequent Doppler processing, and to avoid the Doppler sidelobes of strong stationary clutter masking the Doppler main lobe response of weakly moving targets.

[0023] Step S1205: Recombine the slow-time dimension complex sequence after window function weighting with the original distance-direction complex amplitude sequence to generate a two-dimensional distance pulse data matrix. Each element of the fast-time dimension of the two-dimensional distance pulse data matrix corresponds to a distance unit, and each element of the slow-time dimension corresponds to a pulse number.

[0024] The slow-time windowed complex sequences Mwin(l, q) of all L distance units are arranged in ascending order of row index l and column index q, and then recombined into a two-dimensional range pulse data matrix M. M is an L-row, P-column complex matrix, M(l, q) = Mwin(l, q). The row index l of M corresponds to the fast-time dimension distance unit number, ranging from 0 to L-1. The column index q of M corresponds to the slow-time dimension pulse number, ranging from 0 to P-1. Each element in the two-dimensional range pulse data matrix M retains amplitude and phase information.

[0025] Step S1206: Calculate the clutter profile baseline of the slow time dimension data based on the energy parameters of each range pulse unit in the range pulse two-dimensional data matrix. Estimate the clutter complex response characteristics of each range unit in the slow time dimension based on the clutter profile baseline. Perform adaptive clutter cancellation operation on each range unit along the slow time dimension using the clutter complex response characteristics to generate a range pulse two-dimensional data matrix after clutter profile suppression.

[0026] Calculate the energy parameter E(l, q) = |M(l, q)|² for each element in the two-dimensional range pulse data matrix M. For each range cell l, calculate the median or truncated mean of the energy parameters for all P pulse cycles within that range cell along the slow time dimension, which serves as the clutter profile baseline Bclutter(l) = median{E(l, q)|q = 0, 1, ..., P-1}, where median represents the median operation. Based on the clutter profile baseline, estimate the clutter complex response characteristics of each range cell in the slow time dimension. For the q-th pulse repetition cycle of the l-th range cell, calculate the instantaneous clutter amplitude estimate Aclutter(l, q) = √(α × E(l, q) + (1 - α) × Bclutter(l)), where α is a smoothing coefficient between 0 and 1. The clutter complex response characteristics are constructed as Cclutter(l, q) = Aclutter(l, q) × exp(j × arg(M(l, q))). An adaptive clutter cancellation operation is performed on each range cell along the slow time dimension using the clutter complex response characteristics. The cancellation operation is Mclean(l,q) = M(l,q) - Cclutter(l,q). This adaptive clutter cancellation operation is performed on all L range cells and all P pulse repetition cycles to generate a clutter-contoured range pulse two-dimensional data matrix Mclean with dimensions of L rows and P columns. In the scenario of airborne radar detecting slow-moving vehicles on the ground, the above processing effectively suppresses the strong zero Doppler response of ground stationary clutter in the slow time dimension, allowing the Doppler components of slow-moving vehicle targets that were originally submerged in the clutter main lobe to become visible.

[0027] Step S130: Perform a frame partitioning operation on the distance pulse two-dimensional data matrix to obtain multi-frame coherent processing frame units with continuous pulse sequence relationship, and assign a frame period label to each frame coherent processing frame unit.

[0028] Step S1301: Calculate the frame partition boundary pulse index set between the total number of slow time dimension pulses in the distance pulse two-dimensional data matrix and the ideal intra-frame pulse number corresponding to the maximum coherent accumulation time of the target. The frame partition boundary pulse index set includes multiple non-overlapping frame start pulse indices and frame end pulse indices.

[0029] Obtain the total number of pulses P in the slow-time dimension of the two-dimensional data matrix Mclean of the distance pulses. Calculate the ideal intra-frame pulse count Nframe = Tcoh × PRF based on the target maximum coherence accumulation time Tcoh and the pulse repetition frequency PRF. Divide the frame boundaries in a non-overlapping manner in the slow-time dimension. The frame boundary pulse index set contains the frame start pulse index set {pstart(m) = m × Nframe | m = 0, 1, ..., Mframe-1} and the corresponding frame end pulse index set {pend(m) = min((m+1) × Nframe-1, P-1) | m = 0, 1, ..., Mframe-1}, where Mframe = ceil(P / Nframe) is the total number of frames, ceil represents the floor function, and min represents the minimum value operation. The frame boundary pulse index set ensures that there is no pulse overlap between frames.

[0030] Step S1302: Using the pulse sliding pulse step size between adjacent frames, advance bit by bit along the slow time dimension in the frame division boundary pulse index set to generate a continuous coherent processing frame unit sequence. The pulse overlap between adjacent frame units is determined by the difference between the pulse sliding pulse step size and the number of pulses within the frame.

[0031] Based on the set of pulse indices at the frame division boundaries, an inter-frame sliding step size Nslide is introduced. Nslide is less than Nframe to generate pulse overlap between adjacent frames. The starting pulse index of the m-th coherent processing frame unit is adjusted to pslide(m) = m × Nslide, and the ending pulse index is pslide(m) + Nframe - 1. The number of pulse overlaps between adjacent frame units is Nframe - Nslide. By progressively advancing the sliding step size, a continuous sequence of coherent processing frame units is generated along the slow time dimension. Each coherent processing frame unit contains two-dimensional range pulse data within Nframe consecutive pulse repetition periods. In airborne radar scenarios, inter-frame sliding processing increases the temporal coverage density of moving target coherent accumulation, improving the probability of acquiring the Doppler frequency change trajectory of moving targets.

[0032] Step S1303: Extract complex data of all pulses in each coherent processing frame unit arranged along the slow time dimension to form a three-dimensional data sub-block. The dimensions of the three-dimensional data sub-block include the fast time distance unit dimension, the slow time intra-frame pulse dimension, and the inter-frame sliding dimension.

[0033] For the m-th coherent processing frame unit, its frame contains N consecutive pulses from the pulse index pslide(m) to pslide(m) + Nframe-1. Data for the corresponding column range is extracted from the clutter profile-suppressed range pulse 2D data matrix Mclean to form the m-th frame's 2D complex matrix Fm, with dimensions L rows and N frames. Fm(l, n) = Mclean(l, pslide(m) + n), n = 0, 1, ..., Nframe-1. The 2D complex matrices Fm of all M frames of coherent processing frame units are stacked along the newly added inter-frame sliding dimension to form a 3D data sub-block T, with dimensions L × Nframe × Mframe. The three dimensions of the 3D data sub-block T correspond to the fast-time range unit dimension, the slow-time intra-frame pulse dimension, and the inter-frame sliding dimension, respectively.

[0034] Step S1304: Label each frame of three-dimensional data sub-blocks with sequential frame period labels along the inter-frame sliding dimension. The frame period label is associated with the absolute pulse number index value of the frame's starting pulse in the original distance pulse two-dimensional data matrix and the corresponding slow time sampling time.

[0035] For the 3D data sub-block T along the m-th layer of the inter-frame sliding dimension, i.e., the m-th coherent processing frame unit, a frame period label Lm is assigned. The frame period label Lm contains two associated fields: the absolute pulse number index value pidx(m) = pslide(m) of the frame start pulse in the distance pulse 2D data matrix Mclean, and the slow time sampling time tframe(m) = t0 + pslide(m) × TPRI corresponding to the frame start pulse. The frame period label Lm is stored in tuple form, Lm = (pidx(m), tframe(m)). All Mframe frame period labels constitute the frame period label sequence {L0, L1, ..., L(Mframe-1)}.

[0036] Step S1305: Based on the frame period tag and the carrier start reference phase of the radar transmitted waveform, compensate the frame-level initial carrier phase offset for the slow time dimension complex data of each frame three-dimensional data sub-block to obtain the coherent processing frame unit after initial phase alignment.

[0037] The carrier start reference phase of the radar transmitted waveform is For the m-th coherent processing frame unit, the slow-time sampling moment tframe(m) contained in its frame period label Lm corresponds to a frame-level initial carrier phase. init(m) = 2π × fc × tframe(m) + ref, where fc is the radar carrier frequency. For the m-th layer of the two-dimensional complex matrix Fm of the three-dimensional data sub-block T, the initial carrier phase offset at the frame level is compensated along all L range cells in the n-th column of the slow-time intra-frame pulse dimension. The phase compensation method is Fcalib(m, n, l) = Fm(l, n) × exp(-j × init(m)). Perform the above frame-level initial carrier phase compensation on all Mframe coherent processing frame units to obtain the initial phase-aligned sequence of coherent processing frame units. Initial phase alignment eliminates the fixed carrier phase difference between different frames caused by different frame start times.

[0038] Step S1306: Perform frame sequence merging along the inter-frame sliding dimension on all coherent processing frame units after initial phase alignment to form a multi-frame continuous phase correlation dataset for subsequent Doppler processing.

[0039] The two-dimensional complex matrix Fcalib,m of all M frames of coherently processed frame units after initial phase alignment is merged along the inter-frame sliding dimension. The merged dataset forms a multi-frame continuous phase correlation dataset Dmulti, which is a three-dimensional complex array with dimensions L×Nframe×Mframe. Dmulti(l,n,m)=Fcalib,m(l,n). The three dimensions of Dmulti correspond to the fast-time distance unit dimension, the slow-time intra-frame pulse dimension, and the inter-frame sliding dimension, respectively. In the multi-frame continuous phase correlation dataset, the data along the inter-frame sliding dimension maintains the continuous phase relationship across frames, providing a phase-coherent data foundation for subsequent cross-frame coherent accumulation. In the scenario of airborne radar detecting slow-speed vehicles on the ground, this multi-frame continuous phase correlation dataset contains the slow-time echo information of the vehicle target within multiple coherently processed frames. The Doppler frequency of the vehicle target changes slowly between frames, and the phase relationship is maintained.

[0040] Step S140: Perform Doppler spectrum fragmentation and subband splitting filtering on each frame coherent processing frame unit to generate a slow-time dimension Doppler subband decomposition data cube.

[0041] Step S1401: Extract the slow-time dimension complex pulse sequence of the coherent processing frame unit, perform initial Doppler spectrum analysis based on fast Fourier transform on the complex pulse sequence in the slow-time dimension, and generate the initial Doppler spectral distribution of the corresponding frame.

[0042] For the m-th frame of the multi-frame continuous phase-correlated dataset Dmulti, i.e., the m-th coherent processing frame unit's two-dimensional complex matrix, take the slow-time dimension complex pulse sequence s(l, m, n) = Dmulti(l, n, m), n = 0, 1, ..., Nframe-1 of its l-th distance unit. Perform initial Doppler spectrum analysis based on Fast Fourier Transform on this complex pulse sequence. After zero-padding s(l, m, n) to a length of Nfft, perform Fast Fourier Transform to obtain the initial Doppler spectral distribution Sspec(l, m, k) = ∑(n=0 to Nfft-1)s(l, m, n) × exp(-j × 2π × k × n / Nfft), k = 0, 1, ..., Nfft-1. Perform the above operation on all L distance units to generate the initial Doppler spectral distribution matrix Sspec, m of size L × Nfft for the m-th frame.

[0043] Step S1402: Based on the possible range of Doppler frequency distribution and the width of clutter Doppler spread of the moving target, the Doppler frequency axis covered by the initial Doppler spectral distribution is non-uniformly divided into several Doppler sub-band frequency intervals.

[0044] The potential range of Doppler frequency distribution for moving targets is determined by the target's velocity range and the radar's operating wavelength. The clutter Doppler spread width is determined by the platform's velocity, antenna beamwidth, and radar's operating wavelength. The normalized Doppler frequency axis covered by the initial Doppler spectral distribution is non-uniformly divided into Q Doppler sub-band frequency intervals from -0.5 to 0.5. The sub-band division strategy is as follows: in the main lobe clutter region, i.e., the interval where the normalized Doppler frequency is near zero, the sub-band width is set narrower, resulting in higher frequency resolution, to accurately capture the Doppler components of slow-moving targets in a strong clutter background; in the sidelobe clutter region and the clutter-free region, i.e., the interval where the normalized Doppler frequency deviates significantly from zero, the sub-band width is set wider. The center frequency of each Doppler sub-band frequency interval is denoted as {fsub, q | q = 0, 1, ..., Q-1}, and the upper and lower boundary frequencies of the interval are denoted as {fsub, q, low, fsub, q, high}, respectively.

[0045] Step S1403: Construct a group of finite impulse response bandpass filters that match the center frequency of each Doppler sub-band frequency interval. The passband range of each bandpass filter covers the upper and lower boundary frequencies of the corresponding Doppler sub-band frequency interval.

[0046] For the q-th Doppler sub-band frequency interval, a finite impulse response (FIR) bandpass filter is constructed. The ideal frequency response of this bandpass filter is 1 within [fsub, q, low, fsub, q, high] and 0 outside the interval. The ideal frequency response is converted into FIR filter coefficients using the window function method. An inverse discrete-time Fourier transform is performed on the ideal frequency response to obtain an infinite impulse response, which is then truncated using a Kaiser window to obtain a finite impulse response filter coefficient vector hq = [hq(0), hq(1), ..., hq(Nfir-1)]T of length Nfir. For each of the Q Doppler sub-band frequency intervals, corresponding bandpass filters are constructed to form a finite impulse response bandpass filter bank Hfbank = {h0, h1, ..., h(Q-1)}.

[0047] Step S1404: Simultaneously feed the slow-time dimension complex pulse sequence of each frame into each bandpass filter of the finite impulse response bandpass filter bank, and perform convolution filtering operation pulse by pulse in the slow-time dimension to obtain the sub-band slow-time complex sequence corresponding to each sub-band channel.

[0048] The slow-time complex pulse sequence s(l, m, n) of the l-th range unit in the m-th frame is simultaneously fed into Q bandpass filters of a finite impulse response bandpass filter bank. For the q-th bandpass filter, the convolution filtering operation is performed pulse-by-pulse in the slow-time dimension. The sub-band slow-time complex sequence output by the q-th sub-band channel is ysub(l, m, q, n) = ∑(τ=0 to Nfir-1)hq(τ) × s(l, m, n-τ), n = Nfir-1, Nfir, ..., Nframe-1. To ensure consistent sequence length after convolution filtering, s(l, m, n) is extended using a symmetric boundary extension method before n=0. The above convolution filtering operation is performed on all L range units, all M frames, and all Q sub-band channels.

[0049] Step S1405: Stack the slow-time complex sequences of all sub-band channels within the same frame along the newly added sub-band dimension to form a three-dimensional data format of fast-time dimension distance unit, slow-time dimension pulse number and sub-band dimension channel identifier, and generate a slow-time dimension Doppler sub-band decomposition data cube.

[0050] For the m-th frame, the slow-time complex sequences of the Q sub-band channels are stacked along the newly added sub-band dimension. Specifically, a three-dimensional array Csub,m of the slow-time Doppler sub-band decomposition data for the m-th frame is constructed. Its first dimension is the fast-time dimension distance cell index l, with length L; its second dimension is the slow-time dimension pulse number n, with length Nframes; and its third dimension is the sub-band dimension channel identifier q, with length Q. Csub,m(l,n,q) = ysub(l,m,q,n). The three-dimensional arrays Csub,m of the slow-time Doppler sub-band decomposition data for all M frames are stacked along the inter-frame sliding dimension to form a complete slow-time Doppler sub-band decomposition data cube Ctotal, with dimensions L × Nframes × Q × Mframes. Ctotal(l,n,q,m) = Csub,m(l,n,q).

[0051] Step S1406: Perform inter-channel amplitude equalization processing on the slow-time-dimensional Doppler subband decomposition data cube along the subband dimension, so that the clutter residual energy distribution in each subband channel tends to be flat, and output the equalized slow-time-dimensional Doppler subband decomposition data cube.

[0052] Inter-channel amplitude equalization is performed along the sub-band dimension of the slow-time Doppler sub-band decomposition data cube Ctotal. For the q-th sub-band channel, the statistical mean of clutter residual energy for all range cells and all frames within that channel is calculated as Esub(q) = (1 / (L×Mframe))×∑(l=0toL-1)∑(m=0toMframe-1)|Ctotal(l, 0, q, m)|2, where the amplitude value at the first slow-time pulse index n=0 of each frame represents the clutter residual level. The global mean of clutter residual energy for all Q sub-band channels is calculated as Eglobal = (1 / Q)×∑(q=0toQ-1)Esub(q). An equalization coefficient geq(q) = √(Eglobal / Esub(q)) is applied to the q-th sub-band channel. The equalization process is Ceq(l, n, q, m) = Ctotal(l, n, q, m)×geq(q). The output equalized slow-time Vidoppler subband decomposition data cube Ceq has dimensions of L×Nframe×Q×Mframe. In airborne radar scenarios, inter-channel amplitude equalization ensures that the output energy of different subband channels is comparable, avoiding uneven detection performance between subbands due to differences in filter amplitude response.

[0053] Step S150: Using a preset Doppler subband cross-cycle phase compensation function, perform inter-frame multi-dimensional phase alignment operation on each subband channel of the slow-time dimension Doppler subband decomposition data cube to obtain the cross-frame coherent accumulation data set after inter-frame alignment.

[0054] Step S1501: Extract the cross-frame two-dimensional complex matrix corresponding to each sub-band channel in the slow-time dimensional Doppler sub-band decomposition data cube. The horizontal axis of the cross-frame two-dimensional complex matrix is ​​the intra-frame pulse index, and the vertical axis is the frame period label sequence index.

[0055] For the equalized slow-time Vidoppler subband decomposition data cube Ceq, fix the l-th distance unit and the q-th subband channel, and extract the corresponding cross-frame two-dimensional complex matrix Z(l, q). The horizontal axis of Z(l, q) is the intra-frame pulse index n, n = 0, 1, ..., Nframe-1. The vertical axis of Z(l, q) is the frame period label sequence index m, m = 0, 1, ..., Mframe-1. The element in the m-th row and n-th column of Z(l, q) is Z(l, q)(m, n) = Ceq(l, n, q, m). This cross-frame two-dimensional complex matrix Z(l, q) contains the complex data of all pulses in all frames within the q-th subband channel of the l-th distance unit. The horizontal axis represents the intra-frame slow-time variation, and the vertical axis represents the cross-frame slow-time variation.

[0056] Step S1502: For the cross-frame two-dimensional complex matrix corresponding to each sub-band channel, calculate the complex conjugate cross-correlation value at the corresponding slow time pulse position between adjacent frames along the frame period label sequential index dimension to obtain the inter-frame cross-correlation phase difference sequence.

[0057] For a two-dimensional complex matrix Z(l, q) spanning multiple frames, the complex conjugate cross-correlation values ​​at the corresponding slow-time pulse positions between adjacent frames are calculated along the frame period label index dimension m. For the m-th frame and the (m+1)-th frame, the formula for calculating the inter-frame cross-correlation value is R(l, q)(m) = ∑(n=0 to Nframe-1)Z(l, q)(m, n) × conj(Z(l, q)(m+1, n)). The calculation is performed sequentially for m = 0, 1, ..., Mframe-2, resulting in a complex cross-correlation sequence of length Mframe-1. The m-th element in the inter-frame cross-correlation phase difference sequence is Δ. (l, q)(m) = arg(R(l, q)(m)), where arg represents the argument operation. Δ (l,q)(m) represents the cumulative Doppler phase difference caused by the target motion and radar platform motion between the m-th frame and the (m+1)-th frame.

[0058] Step S1503: Using the inter-frame cross-correlation phase difference sequence and the preset Doppler sub-band cross-cycle phase compensation function, generate a set of inter-frame phase compensation factors for each frame cycle label and the corresponding slow time sampling moment of each intra-frame pulse. The preset Doppler sub-band cross-cycle phase compensation function takes the sub-band center frequency, inter-frame pulse repetition period and frame cycle interval as input parameters.

[0059] The preset Doppler subband cross-cycle phase compensation function is Φcomp(fsub, q, tframe, TPRI) = 2π × fsub, q × tframe + δ(fsub, q, TPRI), where fsub, q are the center frequencies of the q-th subband, tframe is the slow-time sampling moment corresponding to the frame period tag, TPRI is the pulse repetition period, and δ is the system phase deviation correction term related to the subband center frequency and pulse repetition period. For the slow-time sampling moment t(m, n) = tframe(m) + n × TPRI corresponding to the m-th frame period tag and the n-th intra-frame pulse, the inter-frame phase compensation factor is Pcomp(l, q)(m, n) = exp(-j × Φcomp(fsub, q, t(m, n) ), TPRI)), where exp represents an exponential function with the natural constant e as the base. The inter-frame cross-correlation phase difference sequence Δ (l, q)(m) is used as the correction input of the phase compensation function to generate the enhanced inter-frame phase compensation factor Pcomp(l, q)(m, n) = exp(-j × (Φcomp(fsub, q, t(m, n) , TPRI) + ∑(k=0 tom-1)Δ (l, q)(k))). Generate a set of inter-frame phase compensation factors {Pcomp(l, q)(m, n)} for all range cells, sub-band channels, frame periodic labels, and intra-frame pulse indices.

[0060] Step S1504: Multiply the generated set of inter-frame phase compensation factors element by element with the cross-frame two-dimensional complex matrix of the corresponding sub-band channel according to the two-dimensional arrangement of frame period label and intra-frame pulse index, to obtain the sub-band cross-frame two-dimensional complex matrix after inter-frame phase alignment.

[0061] The inter-frame phase compensation factor Pcomp(l, q)(m, n) is multiplied element-wise with the corresponding elements of the cross-frame two-dimensional complex matrix Z(l, q) to obtain the inter-frame phase-aligned sub-band cross-frame two-dimensional complex matrix Zalign(l, q)(m, n) = Z(l, q)(m, n) × Pcomp(l, q)(m, n). The element-wise multiplication operation is performed on all distance cells, sub-band channels, frame period tags, and intra-frame pulse indices. The inter-frame phase alignment operation eliminates cross-frame phase mismatch, enabling continuous phase alignment of complex data between different frames within the same sub-band channel.

[0062] Step S1505: After aligning the inter-frame phase of all sub-band channels, perform inter-frame coherent superposition operation along the sub-band dimension to ensure that the complex data of sub-bands that span different frame period labels but belong to the same slow time dimension pulse index position are superimposed in phase to form the initial cross-frame coherent accumulation data block.

[0063] After inter-frame phase alignment of all Q sub-band channels, the two-dimensional complex matrix Zalign(l, q) for cross-frame sub-band is used for cross-frame coherent superposition along the frame period label dimension. For the l-th distance unit, the q-th sub-band channel, and the n-th intra-frame impulse index, the cross-frame coherent superposition is calculated as Asum(l, q, n) = ∑(m=0 to Mframe-1)Zalign(l, q)(m, n). After cross-frame coherent superposition, an initial cross-frame coherent accumulation data block Ainit is formed, with a size of L×Q×Nframe. Ainit(l, q, n) = Asum(l, q, n). Cross-frame coherent superposition accumulates and enhances the in-phase components within multiple coherently processed frames, with a signal amplitude increase factor of Mframe and a random noise component amplitude increase factor of √Mframe, achieving a cross-frame coherent accumulation signal-to-noise ratio improvement.

[0064] Step S1506: Reorganize the initial cross-frame coherent accumulation data block along the sub-band dimension to generate a cross-frame coherent accumulation data set after frame alignment. Each layer of the cross-frame coherent accumulation data set after frame alignment corresponds to a cross-frame coherent accumulation complex matrix of a sub-band channel.

[0065] The initial cross-frame coherent accumulation data block Ainit is reorganized along the sub-band dimension. For the q-th sub-band channel, the corresponding cross-frame coherent accumulation complex matrix is ​​Aq = Ainit(:, q,:), which means extracting all data of Ainit at index q in the second dimension to obtain an L-row N-frame-column two-dimensional complex matrix. The cross-frame coherent accumulation complex matrices of all Q sub-band channels are stacked along the sub-band dimension to form the cross-frame coherent accumulation data set Aalign = {Aq|q = 0, 1, ..., Q-1} after frame alignment. Aalign is a three-dimensional complex array with size L × Nframe × Q, where the three dimensions correspond to the fast temporal distance unit dimension, the slow temporal intra-frame pulse dimension, and the sub-band dimension, respectively.

[0066] Step S160: Perform two-dimensional Doppler domain detection processing based on the constant false alarm rate criterion on the cross-frame coherent accumulation data set to generate a point detection output result containing the distance coordinates and Doppler frequency coordinates of the moving target point.

[0067] Step S1601: Perform modulus squaring operation on the cross-frame coherent accumulation data set along the distance dimension and Doppler dimension to obtain a two-dimensional range-Doppler energy distribution map. The horizontal axis of the two-dimensional range-Doppler energy distribution map corresponds to the Doppler frequency dimension, and the vertical axis corresponds to the fast time distance unit dimension.

[0068] For the cross-frame coherent accumulation complex matrix Aq of each sub-band channel in the cross-frame coherent accumulation dataset Aalign, a Doppler transform is performed along the slow-time intra-frame pulse dimension. For the l-th row of the slow-time sequence Aq(l, n), n = 0, 1, ..., Nframe-1, a Fast Fourier Transform (FFT) is performed with zero padding to a length Ndop, yielding the Doppler spectrum Dq(l, k) = ∑(n=0 to Ndop-1)Aq(l, n) × exp(-j × 2π × k × n / Ndop), k = 0, 1, ..., Ndop-1. The Doppler spectra of the Q sub-band channels are incoherently accumulated along the sub-band dimension to obtain the fused Doppler spectrum Dfusion(l, k) = ∑(q=0 to Q-1)|Dq(l, k)|. The modulus squared operation is performed on the fused Doppler spectrum to obtain the distance-Doppler two-dimensional energy distribution map E(l, k) = |Dfusion(l, k)|². E(l,k) is an L-row, Ndop-column real matrix, where the horizontal index k corresponds to the Doppler frequency dimension and the vertical index l corresponds to the fast time distance dimension.

[0069] Step S1602: Perform one-dimensional cell average constant false alarm detection on the distance-Doppler two-dimensional energy distribution map along the Doppler direction in the Doppler frequency dimension, and use the average energy of the measured cell in the adjacent reference cell in the Doppler dimension as the basis for calculating the detection threshold to obtain the Doppler-dimensional threshold mark.

[0070] One-dimensional cell-averaged constant false alarm rate (CFAR) detection is performed cell-by-cell along the Doppler dimension of the two-dimensional energy distribution map E(l,k) of the range-Doppler array. For the measured Doppler cell kc of the l-th range cell, Nguard cells and Nref cells are taken on both sides of its Doppler dimension. The reference cells do not include the guard cells or the measured cell itself. The energy mean of the reference cells is calculated as μref(l,kc) = (1 / (2×Nref)) × (∑(k=kc-Nguard-Nreftokc-Nguard-1)E(l,k) + ∑(k=kc+Nguard+1tokc+Nguard+Nref)E(l,k)). The detection threshold Tca(l,kc) = μref(l,kc) × βca, where βca is the threshold multiplier coefficient set according to the expected false alarm probability. Compare E(l, kc) with Tca(l, kc). If E(l, kc) is greater than Tca(l, kc), then mark the Doppler cell as a Doppler-dimensional threshold mark, denoted as Mca, l(kc) = 1; otherwise, Mca, l(kc) = 0. Perform the above detection on all range cells and all Doppler cells to obtain a two-dimensional binary matrix Mdop of size L×Ndop for marking Doppler-dimensional threshold marks.

[0071] Step S1603: Perform one-dimensional cell average constant false alarm detection on the two-dimensional energy distribution map of the distance Doppler along the distance dimension in the fast time distance cell dimension, and use the average energy of the measured cell in the adjacent reference cell in the distance dimension as the basis for calculating the detection threshold to obtain the distance dimension threshold mark.

[0072] One-dimensional cell-averaged constant false alarm rate (CFAR) detection is performed on each Doppler cell along the range dimension of the two-dimensional energy distribution map E(l,k) of the range Doppler array. For the measured range cell lc of the k-th Doppler cell, Nguardr guard cells and Nrefr reference cells are taken on both sides of its range dimension. The energy mean of the reference cells is calculated as μrefr(lc,k) = (1 / (2×Nrefr)) × (∑(l=lc-Nguardr-Nrefrtolc-Nguardr-1)E(l,k) + ∑(l=lc+Nguardr+1tolc+Nguardr+Nrefr)E(l,k)). The detection threshold Tcar(lc,k) = μrefr(lc,k) × βcar. Compare E(lc, k) with Tcar(lc, k). If E(lc, k) is greater than Tcar(lc, k), then mark the distance cell as a distance-dimensional threshold point, denoted as Mcar, k(lc) = 1; otherwise, Mcar, k(lc) = 0. This yields a two-dimensional binary matrix Mrng with size L × Ndop, representing the distance-dimensional threshold point markers.

[0073] Step S1604: Perform a two-dimensional joint logic AND operation on the Doppler-dimensional threshold point mark and the distance-dimensional threshold point mark to obtain an initial two-dimensional point mark detection binary image. The points marked as valid in the initial two-dimensional point mark detection binary image correspond to the cell positions that simultaneously pass the two-dimensional constant false alarm detection.

[0074] A two-dimensional joint logical AND operation is performed on the Doppler-dimensional threshold marker matrix Mdop and the range-dimensional threshold marker matrix Mrng. Cell-wise, Mjoint(l,k) = Mdop(l,k) ∧ Mrng(l,k) is calculated, where ∧ represents the logical AND operation. Cells with a value of 1 in Mjoint represent candidate target marker positions that have passed both the Doppler and range-dimensional constant false alarm detections. Mjoint is the initial two-dimensional marker detection binary map with a size of L × Ndop.

[0075] Step S1605: Perform connected component analysis on the valid points marked in the initial two-dimensional point detection binary image to identify point clusters composed of mutually continuous detection points in the distance dimension and Doppler dimension space, and calculate the centroid coordinates of each point cluster.

[0076] Connectivity analysis is performed on the valid points marked in the initial two-dimensional point detection binary graph Mjoint. The connectivity analysis adopts the 8-connectivity criterion, meaning that valid points that are adjacent or diagonally adjacent in both the distance and Doppler dimensions are considered to belong to the same connected component. A region growing algorithm is used to label all valid points in Mjoint, resulting in a connected component label matrix Llabel, where each connected component has a unique label, and positions not belonging to any connected component are labeled 0. For the g-th connected component, the distance dimension index set {lg1, lg2, ...} and the Doppler dimension index set {kg1, kg2, ...} of all valid points within this connected component are extracted. The centroid coordinates of this connected component are calculated: distance dimension centroid coordinate lcent,g = (∑lg) / Npoint,g, and Doppler dimension centroid coordinate kcent,g = (∑kg) / Npoint,g, where Npoint,g is the number of valid points contained in this connected component.

[0077] Step S1606: Extract the centroid coordinates of each point cluster, map the centroid index in the distance dimension to the distance coordinates of the moving target point, map the centroid index in the Doppler dimension to the Doppler frequency coordinates of the moving target point, and generate a point detection output result containing multiple sets of correspondences between distance coordinates and Doppler frequency coordinates.

[0078] The centroid coordinates of each trace cluster are parameterized. The distance dimension centroid coordinate lcent of the g-th trace cluster is mapped to the distance coordinate Rg = lcent, g × ΔR of the moving target trace, where ΔR = c / (2 × Fs) is the distance between range cells, c is the electromagnetic wave propagation speed, and Fs is the range sampling rate. The Doppler dimension centroid coordinate kcent, g is mapped to the Doppler frequency coordinate fdop of the moving target trace, g = (kcent, g / Ndop - 0.5) × PRF. The distance coordinates, Doppler frequency coordinates, and corresponding trace energy information of all trace clusters are summarized to generate the trace detection output Plots = {(Rg, fdop, g, Epeak, g) | g = 1, 2, ..., G}, where G is the total number of detected trace clusters, and Epeak, g is the peak energy of the cluster. In the scenario of airborne radar detecting slow-moving vehicles on the ground, each entry in the Plots output corresponds to a detected slow-moving vehicle target on the ground, including the slant range information and radial Doppler frequency information of the vehicle relative to the radar platform.

[0079] Step S210: For each moving target point contained in the point detection output result, extract the fast time dimension distance cell index and slow time dimension Doppler frequency index corresponding to the moving target point, and backtrack the cross-frame complex amplitude sequence of the moving target point in each sub-band channel from the cross-frame coherent accumulation data set after frame alignment.

[0080] For the g-th moving target track in the track detection output Plots, its corresponding fast time dimension distance cell index is lg=round(Rg / ΔR), and its Doppler frequency index is kg=round((fdop, g / PRF+0.5)×Ndop), where round represents rounding to the nearest integer. From the cross-frame coherent accumulation data set Aalign after inter-frame alignment, the cross-frame complex amplitude sequence of this track in each sub-band channel is traced back. For the q-th sub-band channel, the lg-th row vector of the cross-frame coherent accumulation complex matrix Aq is taken at the distance cell lg, and then a discrete Fourier transform is performed on this row vector to extract the complex value Dq(lg, kg) at the Doppler frequency index kg. The complex amplitude values ​​of all Q sub-band channels are summarized to form the sub-band complex amplitude vector Vg=[D0(lg, kg), D1(lg, kg), ..., D(Q-1)(lg, kg)]T of this track. Simultaneously, the cross-frame complex amplitude sequence of each sub-band channel in multi-frame data is traced back. For the q-th sub-band channel, the complex sequences of distance cell lg, pulse index 0, sub-band channel q, and all frames m are extracted from the equalized slow-time Doppler sub-band decomposition data cube Ceq. This constitutes the cross-frame complex amplitude sequence Ztrace(g, q, m) = Ceq(lg, 0, q, m), where m = 0, 1, ..., Mframe-1.

[0081] Step S220: Perform Discrete Fourier Transform analysis on the extracted cross-frame complex amplitude sequence along the frame period label order to obtain the fine Doppler spectrum of the moving target point trace along the inter-frame slow time, and extract the inter-frame Doppler frequency offset corresponding to the peak position from the fine Doppler spectrum.

[0082] For the complex amplitude sequence Ztrace(g, q, m) of the g-th moving target trace in the q-th sub-band channel, a discrete Fourier transform analysis is performed after zero-padding to a length Mfft along the frame period label sequence m. The discrete Fourier transform is calculated as Sfine(g, q, u) = ∑(m=0 to Mfft-1)Ztrace(g, q, m) × exp(-j × 2π × u × m / Mfft), u = 0, 1, ..., Mfft-1. This yields the fine Doppler spectrum Sfine(g, q, u) of the trace along the inter-frame slow time. The frequency resolution of the fine Doppler spectrum is PRF / (Mfft × Nframe), which is higher than the intra-frame slow time Doppler resolution. Search for the amplitude peak position in the fine Doppler spectrum and extract the inter-frame Doppler frequency offset Δffine(g,q)=(upeak / Mfft-0.5)×(PRF / Nframe) corresponding to the peak position, where upeak is the frequency index corresponding to the maximum amplitude of the spectrum.

[0083] Step S230: Calculate the fine phase compensation value by combining the inter-frame Doppler frequency offset with the corresponding cross-frame accumulation time interval. Use the fine phase compensation value to perform inter-frame phase fine compensation again on the cross-frame complex amplitude sequence of each sub-band channel. Perform incoherent accumulation on the complex amplitude sequence of each sub-band channel after fine compensation to obtain the multi-sub-band incoherent accumulation energy value of each point.

[0084] Based on the inter-frame Doppler frequency offset Δffine(g, q) and the cross-frame accumulation time interval Tframe = Nframe × TPRI, the fine phase compensation value sequence Φfine(g, q, m) = 2π × Δffine(g, q) × m × Tframe, m = 0, 1, ..., Mframe-1, is calculated. This fine phase compensation value is then used to perform inter-frame fine phase compensation on the cross-frame complex amplitude sequence Ztrace(g, q, m), resulting in the finely compensated sequence Zfine(g, q, m) = Ztrace(g, q, m) × exp(-j × Φfine(g, q, m)). The energy of each sub-band channel's complex amplitude sequence after fine compensation is accumulated along the frame period label order dimension, and the energy value Efine(g, q) = |∑(m=0 to Mframe-1)Zfine(g, q, m)|² is calculated. The energy values ​​of all Q sub-band channels are incoherently accumulated to obtain the multi-sub-band incoherent accumulated energy value of the point trace, Emulti(g)=∑(q=0toQ-1)Efine(g,q).

[0085] Step S240: Sort all point detection output results by point energy according to the multi-subband incoherent cumulative energy value, filter out weak points at the end of the energy sorting with a preset proportion, and retain the points with higher confidence to form a reliable moving target point set.

[0086] The multi-subband incoherent cumulative energy values ​​{Emulti(g)|g=1, 2, ..., G} of all G detection results are sorted in descending order. A preset filtering ratio η is used to filter out the last η points in the energy ranking, i.e., removing the floor(η×G) points with the lowest energy values, where floor represents rounding down. The remaining points with higher energy rankings form a reliable moving target point set Plotsreliable={g retained}. The points in the reliable moving target point set have higher multi-subband incoherent cumulative energy.

[0087] Step S250: For each point in the set of reliable moving target points, a local two-dimensional complex region is extracted from the cross-frame coherent accumulation data set according to its range coordinates and Doppler frequency coordinates, and two-dimensional parameterized spectrum estimation processing is performed to obtain the range dimension super-resolution offset and Doppler dimension super-resolution offset of the moving target point.

[0088] For the g-th point in the set of reliable moving target points, its range coordinates are Rg and its Doppler frequency coordinates are fdop,g. Within the first sub-band channel A0 of the cross-frame coherent accumulation data set Aalign after inter-frame alignment, a local two-dimensional complex region is truncated centered on the point's location. The truncated range is [lg-Wr, lg+Wr] in the range dimension and [kg-Wd, kg+Wd] in the Doppler dimension, where Wr and Wd are the widths of the truncated half-window. Two-dimensional parametric spectrum estimation is performed within this local region. The two-dimensional parametric spectrum estimation uses the two-dimensional MUSIC algorithm. The two-dimensional complex data matrix within the local region is vectorized, its two-dimensional autocorrelation matrix is ​​estimated, and eigenvalue decomposition is performed on the two-dimensional autocorrelation matrix to separate the signal subspace and noise subspace. A two-dimensional MUSIC pseudospectrum is constructed using the orthogonality of the signal and noise subspaces. Peak positions are searched in the pseudospectrum to obtain the range dimension super-resolution offset δRg and the Doppler dimension super-resolution offset δfg. The dimension of δRg is meter, and the dimension of δfg is Hertz.

[0089] Step S260: Use the range dimension super-resolution offset and the Doppler dimension super-resolution offset to fine-tune and update the range coordinates and Doppler frequency coordinates of the original trace, generate the super-resolution refined moving target trace position parameters, and use the updated trace position parameters as the final trace detection output result.

[0090] The original trace's range coordinates Rg and Doppler frequency coordinates fdop,g are finely updated. The updated range coordinates are Rfine,g = Rg + δRg, and the updated Doppler frequency coordinates are fine,g = fdop,g + δfg. The updated trace position parameters (Rfine,g, fine,g) are used as the final trace detection output, Plotsfinal. Based on the original detection results, the final trace detection output achieves sub-resolution cell refinement of the range and Doppler positions using super-resolution spectral estimation techniques.

[0091] Step S310: Obtain the radar platform motion parameters corresponding to the original baseband echo data stream. The radar platform motion parameters include the platform velocity component along the track direction and the velocity component perpendicular to the track direction.

[0092] Platform motion parameters, synchronized with the raw baseband echo data stream, are obtained from the radar platform's inertial navigation system. These parameters include the platform's velocity component *va* along the track direction and its velocity component *vc* perpendicular to the track direction. The spatial geometric relationship between the platform's velocity component *va* along the track direction and the radar antenna beam pointing determines the Doppler center frequency shift introduced by the platform motion. The platform's velocity component *vc* perpendicular to the track direction affects the slant range change rate between the radar and ground targets.

[0093] Step S320: Based on the platform's velocity component along the track direction and the spatial geometric relationship between the platform and the beam pointing direction, establish the mapping relationship between the Doppler center offset caused by the platform motion and the slant range change rate.

[0094] The spatial geometric relationship between the platform and the beam pointing is determined by the antenna beam azimuth pointing angle θaz and elevation pointing angle θel. For the ground target corresponding to the l-th range cell, its slant range relative to the radar platform is R(l) = l × ΔR + Rmin, where Rmin is the minimum observation range. The angle between the beam line-of-sight direction corresponding to this range cell and the platform velocity vector is... (l) Calculated from geometric relationships, the calculation method is cos (l) = (va × cosθaz × cosθel + vc × sinθel) / √(va² + vc²), where cos represents the cosine function, sin represents the sine function, and √ represents the square root operation. The Doppler center offset fdop caused by the platform motion is plat(l) = 2 × va × cosθel. (l) / λ, where λ is the radar operating wavelength. Slant range variation rate dRdt(l) = vc × sin (l). Establish the mapping relationship between the Doppler center offset fdop, plat and the slant range dRdt.

[0095] Step S330: Using the mapping relationship and the slow-time sampling time of the pulse transmission, calculate the slow-time dimension phase error sequence corresponding to each fast-time distance unit, and use the phase error sequence to perform platform motion phase error compensation on the distance pulse two-dimensional data matrix after clutter profile suppression along the slow-time dimension to obtain the platform motion compensated distance pulse two-dimensional data matrix.

[0096] For the l-th range cell, its slow-time sampling time in the p-th pulse repetition cycle is tp = p × TPRI. The cumulative slant range change due to platform motion at this time is ΔRplat(l, p) = ∫(0 to tp)(dRdt(l))dt. The slow-time phase error sequence caused by platform motion is Φplat(l, p) = 4π × ΔRplat(l, p) / λ. A platform motion phase error compensation factor Pplat(l, p) = exp(-j × Φplat(l, p)) is constructed. Platform motion phase error compensation is performed along the slow-time dimension on the clutter profile-suppressed range pulse two-dimensional data matrix Mclean. The compensation method is element-wise multiplication: Mcomp(l, p) = Mclean(l, p) × Pplat(l, p). The platform motion-compensated range pulse two-dimensional data matrix Mcomp is obtained, with dimensions L × P.

[0097] Step S340: Replace the two-dimensional distance pulse data matrix with the motion-compensated distance pulse data matrix of the platform, and perform subsequent frame division operations.

[0098] The range pulse two-dimensional data matrix Mclean in step S130 is replaced with the range pulse two-dimensional data matrix Mcomp after platform motion compensation, and used as the input data for the frame partitioning operation. Subsequent steps are executed according to the aforementioned frame partitioning operation process until a multi-frame continuous phase correlation dataset is generated.

[0099] Step S350: Obtain the azimuth beam pointing angle sequence of the radar antenna, and calculate the antenna pattern amplitude weighting coefficient for each pulse period based on the azimuth beam pointing angle sequence.

[0100] The antenna azimuth beam pointing angle sequence {θaz(p)|p=0,1,...,P-1}, synchronized with each pulse repetition period, is obtained from the radar antenna servo control system. The antenna pattern amplitude weighting coefficient for each pulse period is calculated based on the antenna pattern model. The antenna pattern model adopts a sinc-type pattern. The antenna pattern amplitude weighting coefficient for the p-th pulse period is Gant(p)=sinc(Dant / λ×(sin(θaz(p))-sin(θtarget))), where Dant is the antenna azimuth aperture size, θtarget is the target azimuth angle, and sinc(x)=sin(π×x) / (π×x). Gant(p) reflects the change of antenna gain in the target direction over time during antenna beam scanning.

[0101] Step S360: Multiply the antenna pattern amplitude weighting coefficient along the slow time dimension with the pulse train complex data corresponding to the range pulse two-dimensional data matrix after platform motion compensation pulse pulse one pulse at a time to compensate for the influence of antenna beam scanning on the amplitude modulation of moving target, and obtain the range pulse two-dimensional data matrix after antenna pattern compensation.

[0102] For the motion-compensated range pulse two-dimensional data matrix Mcomp, antenna pattern amplitude-weighted compensation is applied pulse-by-pulse along the slow time dimension. The compensation coefficient for the p-th pulse repetition period is Gcomp(p) = Gnom / Gant(p), where Gnom is the nominal antenna gain. The compensation method is Mant(l, p) = Mcomp(l, p) × Gcomp(p), performed on all L range cells and all P pulse repetition periods. The antenna pattern-compensated range pulse two-dimensional data matrix Mant is obtained, with a size of L × P.

[0103] Step S370: Use the range pulse two-dimensional data matrix after antenna pattern compensation as the input data for subsequent multi-frame coherent processing frame units.

[0104] The range pulse 2D data matrix `Mant`, after antenna pattern compensation, is used as the final range pulse 2D data matrix and input into all subsequent processing steps. In the airborne radar scenario, after platform motion compensation and antenna pattern compensation, the amplitude modulation and phase error of moving target echoes introduced by platform motion and antenna scanning are effectively eliminated.

[0105] Step S410: Extract the complex sequence of the fast time dimension distance unit of all frame units in the multi-frame continuous phase correlation dataset, and calculate the complex time sequence at the same distance unit along the inter-frame sliding dimension.

[0106] For a multi-frame continuous phase-correlated dataset Dmulti, with the l-th distance unit fixed, complex data for this distance unit at all frames and all intra-frame pulse positions are extracted. Sliding along the inter-frame dimension, for the l-th distance unit at the n-th intra-frame pulse position, a cross-frame complex time series Xinter(l, n, m) = Dmulti(l, n, m), where m = 0, 1, ..., Mframe-1, is extracted. This complex time series reflects the variation of data with the frame period label at fixed distance units and fixed intra-frame pulse positions.

[0107] Step S420: Perform time-domain differential interferometry on the complex time series, and multiply the complex sequences corresponding to adjacent frame period labels by conjugate to obtain the inter-frame differential interferometry phase sequence.

[0108] Perform temporal differential interferometry on the complex time series Xinter(l, n, m). Calculate the conjugate multiplication value Idiff(l, n, m) = Xinter(l, n, m) × conj(Xinter(l, n, m+1)), m = 0, 1, ..., Mframe-2, for the complex sequences corresponding to adjacent frame period tags. The m-th element in the inter-frame differential interferometry phase sequence is Φdiff(l, n, m) = arg(Idiff(l, n, m)). The inter-frame differential interferometry phase sequence reflects the phase change at the pulse position within the distance unit between adjacent frames.

[0109] Step S430: Detect the phase abrupt change position along the inter-frame sliding dimension using the inter-frame differential interference phase sequence. When the differential phase change at the phase abrupt change position exceeds a preset phase step threshold, mark the phase abrupt change position as an inter-frame phase discontinuity boundary.

[0110] For the inter-frame differential interferometric phase sequence Φdiff(l, n, m), calculate the change in adjacent differential phases ΔΦdiff(l, n, m) = Φdiff(l, n, m+1) - Φdiff(l, n, m), where m = 0, 1, ..., Mframe-3, along the inter-frame sliding dimension m. A preset phase step threshold Thjump is used. When |ΔΦdiff(l, n, m)| is greater than Thjump, the frame period label index m+1 is marked as the inter-frame phase discontinuity boundary position. By traversing all range cells and intra-frame pulse positions, the set of inter-frame phase discontinuity boundaries Bdiscont = {mbound, 1, mbound, 2, ...} is comprehensively determined.

[0111] Step S440: Divide the multi-frame continuous phase association dataset into several phase continuous data segments according to the inter-frame phase discontinuity boundary, and the inter-frame phase within each phase continuous data segment maintains the characteristic of continuous change.

[0112] Using the inter-frame phase discontinuity boundary set Bdiscont, the multi-frame continuous phase correlation dataset Dmulti is divided into several phase-continuous data segments along the inter-frame sliding dimension. Each phase-continuous data segment contains data corresponding to all frame period labels from the previous discontinuity boundary to the current discontinuity boundary. The s-th phase-continuous data segment Dseg,s contains data with frame period label indices from mbound,s-1 to mbound,s-1. The inter-frame phase within each phase-continuous data segment maintains a continuous change characteristic.

[0113] Step S450: Perform subsequent Doppler spectrum fragmentation and subband splitting filtering independently on each phase-continuous data segment, and perform inter-frame multidimensional phase alignment operation within the segment. Each segment independently generates a sub-segment cross-frame coherent accumulation data block.

[0114] For the s-th phase-continuous data segment Dseg,s, the Doppler spectrum fragmentation and subband splitting filtering process described in step S140 is performed independently to generate the slow-time dimension Doppler subband decomposed data cube Cseg,s for this segment. Then, the inter-frame multi-dimensional phase alignment operation described in step S150 is performed within the segment to generate the cross-frame coherent accumulation data block Aseg,s for this segment. Each segment independently generates a sub-segment cross-frame coherent accumulation data block.

[0115] Step S460: Concatenate the sub-segment cross-frame coherent accumulation data blocks corresponding to all phase continuous data segments along the inter-frame sliding dimension to form a cross-frame coherent accumulation data set after segment merging.

[0116] All S phase-continuous data segments corresponding to the corresponding sub-segments of cross-frame coherent accumulation data blocks Aseg are concatenated end-to-end along the inter-frame sliding dimension. The concatenated cross-segment merged cross-frame coherent accumulation data set Amerg has its inter-frame sliding dimension length restored to the sum of the lengths of each segment. Amerg replaces the original inter-frame aligned cross-frame coherent accumulation data set Aalign and participates in subsequent two-dimensional Doppler domain detection processing.

[0117] Step S510: Obtain the radar transmission waveform parameters in the original baseband echo data stream, wherein the radar transmission waveform parameters include the transmission pulse duration and the linear frequency modulation bandwidth of the transmission signal.

[0118] Obtain the radar transmit waveform parameters corresponding to the raw baseband echo data stream from the radar transmit waveform parameter configuration interface. The radar transmit waveform parameters include the transmit pulse duration Tpulse and the transmit signal linear frequency modulation bandwidth BLFM.

[0119] Step S520: Determine the range sampling rate based on the transmission pulse duration, the linear frequency modulation bandwidth of the transmission signal, and the system sampling rate. Based on the range sampling rate, perform a resampling operation on the fast time range unit dimension of the slow time dimension Doppler subband decomposition data cube to obtain a Doppler subband decomposition data cube with uniform range dimension scale.

[0120] The range sampling rate is determined based on the transmit pulse duration Tpulse, the transmit signal linear frequency modulation bandwidth BLFM, and the system sampling rate Fsys. The resampled range sampling rate Fsres satisfies the Nyquist sampling theorem with BLFM. Resampling is performed on the fast-time range cell dimension of the slow-time Doppler subband decomposition data cube Ctotal. The resampling operation uses polyphase filtering or Lagrange interpolation to unify the range cell spacing to ΔRres = c / (2 × Fsres). This yields the Doppler subband decomposition data cube Cres with a unified range dimension scale, and its range dimension length is adjusted to Lres.

[0121] Step S530: Using the Doppler subband decomposition data cube with uniform distance dimension scale, perform autocorrelation matrix estimation on the pulse complex sequence of each distance unit along the slow time dimension in each subband channel to generate the slow time autocorrelation matrix of each distance unit in each subband channel.

[0122] For the Doppler subband decomposition data cube Cres with a uniform range dimension scale, within the q-th subband channel, for the l-th range cell, its slow-time dimension pulse complex sequence sq(l, n) = Cres(l, n, q, 0), where n = 0, 1, ..., Nframe-1, is taken as representative data from frame 0. The autocorrelation matrix of this slow-time dimension pulse complex sequence is estimated using a forward and backward spatial smoothing method. Construct an autocorrelation matrix Rsq(l) of size Nsmooth×Nsmooth, where the element in the i-th row and j-th column is Rsq(l)(i,j)=(1 / (2×(Nframe-Nsmooth)))×∑(n=0toNframe-Nsmooth-1)[sq(l,n+i)×conj(sq(l,n+j))+sq(l,Nframe-1-ni)×conj(sq(l,Nframe-1-nj))]. Generate the slow-time autocorrelation matrix of each distance cell in each sub-band channel.

[0123] Step S540: Perform eigenvalue decomposition on each slow-time autocorrelation matrix and extract the eigenvector corresponding to the largest eigenvalue as the dominant slow-time mode vector of that distance unit in that sub-band channel.

[0124] Eigenvalue decomposition is performed on the slow-time autocorrelation matrix Rsq(l) of the l-th distance unit in the q-th sub-band channel. The eigenvalue decomposition yields a set of eigenvalues ​​{λ1, λ2, ..., λNsmooth} and corresponding eigenvectors {v1, v2, ..., vNsmooth}, where the eigenvalues ​​are arranged in descending order. The eigenvector vmax corresponding to the largest eigenvalue λmax is extracted as the dominant slow-time mode vector mdominant(l, q) = vmax for that distance unit in that sub-band channel. The dominant slow-time mode vector reflects the main variation pattern of the slow-time signal in that sub-band channel of that distance unit.

[0125] Step S550: Perform inter-subband cross-correlation matching on the dominant slow-time pattern vectors of different sub-band channels in the same distance cell, and calculate the inter-subband cross-correlation matching coefficient. The inter-subband cross-correlation matching coefficient reflects the degree of similarity of the slow-time patterns of the moving target in different sub-bands.

[0126] For the l-th distance cell, take the dominant slow-time mode vectors mdominant(l, q1) and mdominant(l, q2) of two different sub-band channels q1 and q2, and calculate the cross-correlation matching coefficient between these two vectors. The cross-correlation matching coefficient is calculated as ρ(l, q1, q2) = |mdominant(l, q1)T × conj(mdominant(l, q2))| / (||mdominant(l, q1)|| × ||mdominant(l, q2)||), where T represents the transpose operation and ||·|| represents the vector norm. The cross-correlation matching coefficient is calculated for all sub-band channel pairs.

[0127] Step S560: Construct a cross-subband slow-time mode consistency weighted network using the cross-correlation matching coefficients between subbands, and perform cross-subband consistency weighted enhancement on the slow-time complex sequences of each subband channel in each distance unit to obtain the cross-subband slow-time mode enhanced Doppler subband decomposition data cube.

[0128] A cross-subband slow-time mode consistency weighted network is constructed using subband channels as nodes and inter-subband cross-correlation matching coefficients as edge weights. For the l-th distance cell, the consistency enhancement weight wconsis(l, q) = ∑(q'=0, q'≠qtoQ-1)ρ(l, q, q') for the q-th subband channel is calculated. The consistency enhancement weight is applied to the slow-time complex sequence of this subband channel, and the enhancement process is Cennh(l, n, q, m) = Cres(l, n, q, m) × wconsis(l, q). The cross-subband slow-time mode enhanced Doppler subband decomposition data cube Cennh is obtained.

[0129] Step S570: Replace the slow-time dimension Doppler subband decomposition data cube with the cross-subband slow-time mode enhanced Doppler subband decomposition data cube for subsequent inter-frame multi-dimensional phase alignment operations.

[0130] The original slow-time dimension Doppler subband decomposition data cube Ctotal is replaced with the cross-subband slow-time mode enhanced Doppler subband decomposition data cube Cenh, and the subsequent inter-frame multidimensional phase alignment operation is performed in step S150.

[0131] Based on the same inventive concept, please refer to Figure 2 The diagram shows a schematic block diagram of a radar processing system 100 for moving target coherent accumulation and Doppler domain detection provided in an embodiment of this application. The radar processing system 100 for moving target coherent accumulation and Doppler domain detection may include a communication unit 110, a machine-readable storage medium 120, and a processor 130.

[0132] In this embodiment, the machine-readable storage medium 120 can also be integrated into the processor 130 and can communicate and interact with external systems through the communication unit 110. The machine-readable storage medium 120 stores machine-executable instructions for executing the scheme of this application, and the processor 130 executes the machine-executable instructions stored in the machine-readable storage medium 120 to implement the radar processing method for moving target coherent accumulation and Doppler domain detection provided in the aforementioned method embodiments.

[0133] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A radar processing method for coherent accumulation and Doppler domain detection of moving targets, characterized in that, The method includes: The system receives raw baseband echo data streams acquired by the radar radio frequency front-end, wherein the raw baseband echo data streams contain digitally sampled complex signals with multiple pulse repetition cycles. The original baseband echo data stream is processed by pulse dimension distance processing to generate a two-dimensional distance pulse data matrix containing fast time dimension distance cells and slow time dimension pulse numbers; A frame partitioning operation is performed on the distance pulse two-dimensional data matrix to obtain multi-frame coherent processing frame units with continuous pulse sequence relationships, and a frame period label is assigned to each frame coherent processing frame unit. Each frame of coherent processing frame unit is subjected to Doppler spectrum fragmentation and subband splitting filtering to generate a slow-time dimension Doppler subband decomposition data cube. The slow-time dimension Doppler subband decomposition data cube divides continuous pulse segments into multiple parallel subband channels in the slow-time dimension. Using a preset Doppler subband cross-cycle phase compensation function, an inter-frame multi-dimensional phase alignment operation is performed on each subband channel of the slow-time dimension Doppler subband decomposition data cube to obtain an inter-frame aligned cross-frame coherent accumulation data set. The cross-frame coherent accumulation data set is subjected to two-dimensional Doppler domain detection processing based on the constant false alarm rate criterion to generate a point detection output result containing the distance coordinates and Doppler frequency coordinates of the moving target point.

2. The radar processing method for moving target coherent accumulation and Doppler domain detection according to claim 1, characterized in that, The step of performing pulse-dimensional distance processing on the original baseband echo data stream to generate a two-dimensional distance pulse data matrix containing fast-time-dimensional distance cells and slow-time-dimensional pulse numbers includes: The discrete sampling sequence within a single pulse period of the original baseband echo data stream is extracted, and the discrete sampling sequence is subjected to pulse compression matched filtering to obtain a high-resolution range complex amplitude sequence for a single pulse period. Arrange the high-resolution range-reverse amplitude sequence of all pulse cycles according to the pulse triggering time sequence, and construct an initial two-dimensional range pulse arrangement array with the fast time dimension sampling index as the row and the slow time dimension pulse number as the column; In the slow time dimension, pulse repetition period time-varying deviation correction is performed on the initial range pulse two-dimensional arrangement array to eliminate the slow time dimension sampling position offset accumulated due to radar transmission clock jitter and generate a slow time dimension uniform arrangement array. For each fast time dimension distance cell in the slow time dimension uniform arrangement array, extract the complex sequence along the slow time dimension direction, apply window function weighting operation to suppress the slow time dimension sidelobes, and obtain the slow time dimension windowed complex sequence. The slow-time dimension complex sequence, which is weighted by a window function, is recombined with the original distance-direction complex amplitude sequence to generate a two-dimensional distance pulse data matrix. Each element of the fast-time dimension of the distance pulse two-dimensional data matrix corresponds to a distance unit, and each element of the slow-time dimension corresponds to a pulse number. Based on the energy parameters of each range pulse unit in the range pulse two-dimensional data matrix, the clutter profile baseline of the slow time dimension data is calculated. The clutter complex response characteristics of each range unit in the slow time dimension are estimated according to the clutter profile baseline. The clutter complex response characteristics are used to perform adaptive clutter cancellation operation on each range unit along the slow time dimension to generate a range pulse two-dimensional data matrix with clutter profile suppression.

3. The radar processing method for moving target coherent accumulation and Doppler domain detection according to claim 1, characterized in that, The step of performing a frame partitioning operation on the two-dimensional distance pulse data matrix to obtain multi-frame coherent processing frame units with continuous pulse sequence relationships, and assigning a frame period label to each frame coherent processing frame unit, includes: Calculate the frame partition boundary pulse index set between the total number of slow-time dimension pulses in the distance pulse two-dimensional data matrix and the ideal intra-frame pulse number corresponding to the maximum coherent accumulation time of the target. The frame partition boundary pulse index set contains multiple non-overlapping frame start pulse indices and frame end pulse indices. Using the pulse sliding pulse step size between adjacent frames, the pulse index set of the frame division boundary is advanced bit by bit along the slow time dimension to generate a continuous coherent processing frame unit sequence. The pulse overlap between adjacent frame units is determined by the difference between the pulse sliding pulse step size and the number of pulses within the frame. For each coherent processing frame unit, extract all complex data of pulses within its frame arranged along the slow time dimension to form a three-dimensional data sub-block. The dimensions of the three-dimensional data sub-block include the fast time distance unit dimension, the slow time intra-frame pulse dimension, and the inter-frame sliding dimension. Each frame of three-dimensional data sub-blocks is labeled with sequential frame period labels along the inter-frame sliding dimension. The frame period label is associated with the absolute pulse number index value of the frame's starting pulse in the original distance pulse two-dimensional data matrix and the corresponding slow time sampling time. Based on the frame period tag and the carrier start reference phase of the radar transmitted waveform, the frame-level initial carrier phase offset is compensated for the slow time dimension complex data of each frame three-dimensional data sub-block to obtain the coherent processing frame unit after initial phase alignment. After initial phase alignment, all coherent processing frame units are merged along the inter-frame sliding dimension to form a multi-frame continuous phase-correlated dataset for subsequent Doppler processing.

4. The radar processing method for moving target coherent accumulation and Doppler domain detection according to claim 1, characterized in that, The step of performing Doppler spectrum fragmentation and subband splitting filtering on each frame coherently processed frame unit to generate a slow-time dimension Doppler subband decomposed data cube includes: Extract the slow-time dimension complex pulse sequence of the coherent processing frame unit, and perform initial Doppler spectrum analysis based on fast Fourier transform on the complex pulse sequence in the slow-time dimension to generate the initial Doppler spectral distribution of the corresponding frame. Based on the possible range of Doppler frequency distribution and the width of clutter Doppler spread of the moving target, the Doppler frequency axis covered by the initial Doppler spectral distribution is non-uniformly divided into several Doppler sub-band frequency intervals. Construct a finite impulse response bandpass filter bank that matches the center frequency of each Doppler sub-band frequency interval. The passband range of each bandpass filter covers the upper and lower boundary frequencies of the corresponding Doppler sub-band frequency interval. Each frame of slow-time dimension complex pulse sequence is simultaneously fed into each bandpass filter of the finite impulse response bandpass filter bank, and convolution filtering is performed pulse-by-pulse in the slow-time dimension to obtain the sub-band slow-time complex sequence corresponding to each sub-band channel. The slow-time complex sequences of all sub-band channels within the same frame are stacked along the newly added sub-band dimension to form a three-dimensional data format of fast-time dimension distance unit, slow-time dimension pulse number and sub-band dimension channel identifier, generating a slow-time dimension Doppler sub-band decomposition data cube; The slow-time-dimensional Doppler subband decomposition data cube is subjected to inter-channel amplitude equalization along the subband dimension, so that the clutter residual energy distribution in each subband channel tends to be flat, and the equalized slow-time-dimensional Doppler subband decomposition data cube is output.

5. The radar processing method for moving target coherent accumulation and Doppler domain detection according to claim 1, characterized in that, The method utilizes a preset Doppler subband cross-cycle phase compensation function to perform inter-frame multi-dimensional phase alignment operations on each subband channel of the slow-time-dimensional Doppler subband decomposition data cube, resulting in an inter-frame aligned cross-frame coherent accumulation data set, including: Extract the cross-frame two-dimensional complex matrix corresponding to each sub-band channel in the slow-time dimensional Doppler sub-band decomposition data cube. The horizontal axis of the cross-frame two-dimensional complex matrix is ​​the intra-frame pulse index, and the vertical axis is the frame period label order index. For each sub-band channel, the complex conjugate cross-correlation value at the corresponding slow time pulse position between adjacent frames is calculated along the frame period label index dimension to obtain the inter-frame cross-correlation phase difference sequence. Using the inter-frame cross-correlation phase difference sequence and the preset Doppler sub-band cross-cycle phase compensation function, a set of inter-frame phase compensation factors is generated for each frame cycle label and the slow time sampling moment corresponding to each intra-frame pulse. The preset Doppler sub-band cross-cycle phase compensation function takes the sub-band center frequency, the inter-frame pulse repetition period, and the frame cycle interval as input parameters. The generated set of inter-frame phase compensation factors is arranged in a two-dimensional manner according to the frame period label and the intra-frame pulse index, and multiplied element by element with the cross-frame two-dimensional complex matrix of the corresponding sub-band channel to obtain the sub-band cross-frame two-dimensional complex matrix after inter-frame phase alignment. After aligning the inter-frame phase of all sub-band channels, the two-dimensional complex matrix of the sub-band cross-frame is subjected to inter-frame coherent superposition operation along the sub-band dimension, so that the complex data of the sub-bands that cross different frame period labels but belong to the same slow time dimension pulse index position are superimposed in phase to form the initial cross-frame coherent accumulation data block. The initial cross-frame coherent accumulation data block is reorganized along the sub-band dimension to generate an inter-frame aligned cross-frame coherent accumulation data set. Each layer of the inter-frame aligned cross-frame coherent accumulation data set corresponds to a cross-frame coherent accumulation complex matrix of a sub-band channel.

6. The radar processing method for moving target coherent accumulation and Doppler domain detection according to claim 1, characterized in that, The step of performing two-dimensional Doppler domain detection processing based on the constant false alarm rate criterion on the cross-frame coherent accumulated data set generates a point detection output result containing the distance coordinates and Doppler frequency coordinates of the moving target point, including: The cross-frame coherent accumulated data set is subjected to modulus squaring along the distance dimension and Doppler dimension to obtain a two-dimensional range-Doppler energy distribution map. The horizontal axis of the two-dimensional range-Doppler energy distribution map corresponds to the Doppler frequency dimension, and the vertical axis corresponds to the fast time distance unit dimension. One-dimensional cell-average constant false alarm detection is performed on the two-dimensional energy distribution map of the distance-Doppler in the Doppler frequency dimension along the Doppler direction cell by cell. The average energy of the measured cell in the Doppler dimension adjacent reference cell is used as the basis for calculating the detection threshold to obtain the Doppler dimension threshold mark. In the fast time distance cell dimension, one-dimensional cell average constant false alarm detection is performed on the two-dimensional energy distribution map of the distance Doppler along the distance dimension direction. The average energy of the measured cell in the distance dimension adjacent to the reference cell is used as the basis for calculating the detection threshold, and the distance dimension threshold mark is obtained. A two-dimensional joint logical AND operation is performed on the Doppler-dimensional threshold point markers and the distance-dimensional threshold point markers to obtain an initial two-dimensional point detection binary image. The points marked as valid in the initial two-dimensional point detection binary image correspond to the unit positions that simultaneously pass the two-dimensional constant false alarm detection. Connectivity analysis is performed on the valid points marked in the initial two-dimensional point trace detection binary image to identify point trace clusters composed of mutually continuous detection points in the distance dimension and Doppler dimension space, and the centroid coordinates of each point trace cluster are calculated. Extract the centroid coordinates of each point cluster, map the centroid index in the distance dimension to the distance coordinates of the moving target point, and map the centroid index in the Doppler dimension to the Doppler frequency coordinates of the moving target point, generating a point detection output containing multiple sets of correspondences between distance coordinates and Doppler frequency coordinates.

7. The radar processing method for moving target coherent accumulation and Doppler domain detection according to claim 2, characterized in that, The method further includes: For each moving target point in the point detection output, extract the fast time dimension distance cell index and slow time dimension Doppler frequency index corresponding to the moving target point, and backtrack the cross-frame complex amplitude sequence of the moving target point in each sub-band channel from the cross-frame coherent accumulation data set after frame alignment. The extracted cross-frame complex amplitude sequence is subjected to discrete Fourier transform analysis along the frame period label order to obtain the fine Doppler spectrum of the moving target point trace along the inter-frame slow time. The inter-frame Doppler frequency offset corresponding to the peak position is extracted from the fine Doppler spectrum. By using the inter-frame Doppler frequency offset and the corresponding cross-frame accumulation time interval, a fine phase compensation value is calculated. The fine phase compensation value is then used to perform inter-frame phase fine compensation again on the cross-frame complex amplitude sequence of each sub-band channel. The complex amplitude sequence of each sub-band channel after fine compensation is then incoherently accumulated to obtain the multi-sub-band incoherent accumulation energy value of each point. Based on the multi-subband incoherent cumulative energy value, all point detection output results are sorted by point energy. Weak points at the end of the energy sorting are filtered out, and points with higher confidence are retained to form a reliable moving target point set. For each point in the set of reliable moving target points, a local two-dimensional complex region is extracted from the cross-frame coherent accumulation data set based on its range coordinates and Doppler frequency coordinates, and two-dimensional parameterized spectrum estimation is performed to obtain the range dimension super-resolution offset and Doppler dimension super-resolution offset of the moving target point. The distance coordinates and Doppler frequency coordinates of the original trace are finely adjusted and updated using the range dimension super-resolution offset and the Doppler dimension super-resolution offset to generate the position parameters of the moving target trace after super-resolution refinement, and the updated trace position parameters are used as the final trace detection output result.

8. The radar processing method for moving target coherent accumulation and Doppler domain detection according to claim 2, characterized in that, The method further includes: Obtain the radar platform motion parameters corresponding to the original baseband echo data stream. The radar platform motion parameters include the platform velocity component along the track direction and the velocity component perpendicular to the track direction. Based on the platform's velocity components along the track direction and the spatial geometric relationship between the platform and the beam pointing direction, a mapping relationship between the Doppler center offset caused by the platform's motion and the slant range change rate is established. Using the mapping relationship and the slow-time sampling time of the pulse transmission, the slow-time dimension phase error sequence corresponding to each fast-time distance unit is calculated. The phase error sequence is used to compensate for the platform motion phase error of the range pulse two-dimensional data matrix after clutter profile suppression along the slow-time dimension, so as to obtain the platform motion compensated range pulse two-dimensional data matrix. The distance pulse two-dimensional data matrix is ​​replaced by the distance pulse two-dimensional data matrix after platform motion compensation, and the subsequent frame division operation is performed. Obtain the azimuth beam pointing angle sequence of the radar antenna, and calculate the antenna pattern amplitude weighting coefficient for each pulse period based on the azimuth beam pointing angle sequence; The antenna pattern amplitude weighting coefficient is multiplied pulse-by-pulse along the slow time dimension with the complex data of the pulse train corresponding to the range pulse two-dimensional data matrix after platform motion compensation, in order to compensate for the influence of antenna beam scanning on the amplitude modulation of moving targets and obtain the range pulse two-dimensional data matrix after antenna pattern compensation. The range pulse two-dimensional data matrix after antenna pattern compensation is used as the input data for subsequent multi-frame coherent processing frame units to complete the moving target coherent accumulation processing for joint compensation of platform motion and antenna scanning.

9. The radar processing method for moving target coherent accumulation and Doppler domain detection according to claim 4, characterized in that, The method further includes: Obtain radar transmit waveform parameters from the original baseband echo data stream, wherein the radar transmit waveform parameters include transmit pulse duration and transmit signal linear frequency modulation bandwidth; The range sampling rate is determined based on the transmission pulse duration, the linear frequency modulation bandwidth of the transmission signal, and the system sampling rate. Based on the range sampling rate, the fast time range unit dimension of the slow time dimension Doppler subband decomposition data cube is resampled to obtain a Doppler subband decomposition data cube with uniform range dimension scale. Using the Doppler subband decomposition data cube with uniform distance dimension scale, the autocorrelation matrix of the pulse complex sequence of each distance unit is estimated along the slow time dimension in each subband channel, generating the slow time autocorrelation matrix of each distance unit in each subband channel; Eigenvalue decomposition is performed on each slow-time autocorrelation matrix, and the eigenvector corresponding to the largest eigenvalue is extracted as the dominant slow-time mode vector of that distance unit in that sub-band channel. The dominant slow-time pattern vectors of different sub-band channels in the same distance unit are cross-correlation matched between sub-bands, and the cross-correlation matching coefficient between sub-bands is calculated. The cross-correlation matching coefficient between sub-bands reflects the degree of similarity of the slow-time patterns of the moving target in different sub-bands. A cross-subband slow-time mode consistency weighting network is constructed using the cross-subband cross-correlation matching coefficients. Cross-subband consistency weighting enhancement is performed on the slow-time complex sequences of each subband channel in each distance unit to obtain the cross-subband slow-time mode enhanced Doppler subband decomposition data cube. The Doppler subband decomposition data cube enhanced with the cross-subband slow-time mode is used to replace the slow-time dimension Doppler subband decomposition data cube for subsequent inter-frame multidimensional phase alignment operations.

10. A radar processing system for coherent accumulation and Doppler domain detection of moving targets, characterized in that, include: processor; A machine-readable storage medium for storing machine-executable instructions of the processor; The processor is configured to execute the radar processing method for moving target coherent accumulation and Doppler domain detection as described in any one of claims 1 to 9 by executing the machine-executable instructions.

Images