Terahertz imaging method based on CEEMD noise reduction
By combining CEEMD and RMA algorithms, the problems of modal aliasing and high computational complexity in terahertz imaging are solved, and efficient sparse array imaging is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- THE 41ST INST OF CHINA ELECTRONICS TECH GRP
- Filing Date
- 2025-12-24
- Publication Date
- 2026-05-08
AI Technical Summary
In existing terahertz imaging methods, empirical mode decomposition suffers from mode aliasing, integrated empirical mode decomposition leaves residual white noise, and the back projection imaging algorithm for sparse non-uniform arrays has high computational complexity.
CEEMD is used to denoise the signal, decomposing only the first few IMF components. An equally spaced equivalent phase center sequence is found in the sparse array, and the imaging process is combined with the RMA algorithm to reduce computational complexity.
It effectively reduces computational complexity, improves computational efficiency, and achieves efficient terahertz imaging.
Smart Images
Figure CN121995376A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of terahertz imaging technology and relates to a terahertz imaging method based on CEEMD noise reduction. Background Technology
[0002] Terahertz waves (THz) lie between microwaves and infrared in the electromagnetic spectrum, possessing both optical and electromagnetic properties. Their frequency range spans from 0.1 THz to 10 THz, and they exhibit strong penetrating power and low photon energy. Consequently, terahertz imaging has been extensively studied in academia and industry, showing broad development and application prospects in fields such as non-destructive testing and security inspection. The combination of terahertz waves and MIMO-SAR (Synthetic Aperture Radar) technology has become a key technology in the field of terahertz imaging. On the one hand, it can fully utilize spatial degrees of freedom, obtaining a far greater number of equivalent phase centers than the actual number of antennas through multiple antennas for transmission and reception; on the other hand, it can achieve high-resolution imaging capabilities through synthetic aperture processing.
[0003] Existing methods use Empirical Mode Decomposition (EMD) to denoise signals, but they suffer from mode aliasing. Ensemble EMD (EEMD) introduces white noise into the signal to be decomposed, which can improve the mode aliasing of EMD decomposition, but EEMD leaves some white noise in each intrinsic mode function (IMF) component. Complementary EMD (CEEMD) applies the same white noise with opposite signs to the signal, which can cancel the residual noise in the IMF components. However, CEEMD is computationally expensive during decomposition. Furthermore, for sparse and non-uniform arrays, the computational complexity is too high if a back projection imaging algorithm is used. Summary of the Invention
[0004] To address the problems existing in the prior art, this invention provides a terahertz imaging method based on CEEMD noise reduction. This method uses CEEMD to reduce the noise of the acquired signal, which can reduce computational complexity.
[0005] The technical solution provided by this invention is as follows: A terahertz imaging method based on CEEMD noise reduction, comprising the following steps: (1) The digital signal received in the fast time dimension is decomposed using CEEMD to calculate the first P IMF components; the original signal is then subtracted from the first P IMF components. (2) Decompose the three-dimensional data processed in step (1) into four groups of data that are uniformly sampled on the XY plane; (3) Using the wavenumber domain-based RMA method, the four sets of data were processed to form images; (4) Coherently accumulate the four images formed by the four sets of data to form the final image.
[0006] Preferably, for a non-uniformly distributed sparse array, four equally spaced equivalent phase center sequences are found and processed using an RMA algorithm based on the wavenumber domain.
[0007] Preferably, in step (4), during accumulation, the same position of the target in the image is identified, and accumulation is performed with that position as the alignment center.
[0008] Preferably, in step (3), when performing RMA algorithm processing, the echo data is processed using Fast Fourier Transform.
[0009] Preferably, this method is applicable to sparse MIMO arrays.
[0010] Compared with the prior art, the beneficial effects of the present invention are: (1) When using CEEMD for modal decomposition, only the first few IMF terms are decomposed to reduce computational complexity; (2) For a non-uniformly distributed sparse array, four sets of equally spaced equivalent phase center sequences can be found, so that the RMA algorithm can be used for fast calculation, thereby improving the computational efficiency. Attached Figure Description
[0011] Figure 1 This is a schematic diagram of a sparse MIMO array layout; Figure 2 This is a schematic diagram of the equivalent phase center distribution. Detailed Implementation
[0012] To more clearly understand the technical content of this invention, its solution will now be described in detail with reference to the accompanying drawings and specific embodiments. The accompanying drawings show a preferred embodiment of this invention, but its scope of protection is not limited thereto, and can be flexibly adjusted according to specific usage requirements in practical applications. This embodiment aims to help those skilled in the art to fully and deeply understand the inventive concept of this solution.
[0013] The sparse MIMO array of this invention consists of 32 transmit and 12 receive antennas. The coordinates of the receive antennas are (-0.055+0.01m, -0.017, 0), where m = 0, 2…11. The coordinates of the transmit antennas are (Tn, 0.017, 0), where n = 1,…32. The transmit antennas consist of four groups of antennas, each group consisting of 8 array elements. The coordinates (in meters) of the 32 transmit antennas on the X-axis are: (1).
[0014] like Figure 1 As shown, the sparse MIMO array is arranged in the XY plane. The target under test is placed below and moves uniformly along the Y-axis. It is assumed that the target moves j=1,…J azimuths at equal intervals along the Y-axis in slow time. Simultaneously, the terahertz sparse MIMO array uses a time-division multiplexing transmission method, with the transmitting antennas transmitting alternately and the receiving antennas receiving simultaneously. 32 transmitting antennas and 12 receiving antennas form 384 equivalent phase centers, meaning spatial sampling is performed at 384 azimuths along the X-axis. Therefore, 384xJ spatial samples are performed in the X and Y planes. The terahertz signal transmitted by each equivalent phase center is scattered by the transmitting antenna, the target under test, and the receiving antenna. After a series of processing steps by the imaging system and digital sampling in fast time (the number of uniform digital sampling points is K), a digital signal is formed. The final three-dimensional digital signal is Data[k][i][j], where k=1,…K; i=1,2,…384; j=1,2…J.
[0015] To reduce noise in the received data, this invention employs CEEMD for noise reduction of the digital signal in the fast time dimension. Since CEEMD performs L lumped-mean decompositions, the lth addition of noise requires adding the same white noise with opposite signs. While this can largely cancel out the noise, it increases computational complexity and time. Since noise is generally concentrated in the first few IMF components, this invention, to reduce computational complexity, only calculates the first few IMF components during CEEMD decomposition and then subtracts them from the original signal. This reduces computational complexity while simultaneously reducing noise.
[0016] In the slow time dimension, the 384 equivalent phase centers formed along the X-axis are not evenly spaced, meaning the spatial sampling intervals along the X-axis are non-uniform. Based on the layout of formula (1), this invention finds four sets of evenly spaced equivalent phase center sequences from the 384 equivalent phase centers, labeled (A), (B), (C), and (D), as follows: Figure 2 As shown, the intervals of the four sequences on the X-axis are 0.005 meters, 0.005 meters, 0.005 meters, and 0.0025 meters, respectively.
[0017] The coordinates of the four sets of equivalent phase center sequences on the X-axis are as follows: (A) Sequence: Tx(n) = -0.2637 + n * 0.005, n = 0, ... 95; (B) Sequence: Tx (n) = -0.2360 + n * 0.005, n = 0, ... 95; (2) (C) Sequence: Tx(n) = -0.2225 + n * 0.005, n = 0, ... 95; (D) Sequence: Tx (n) = -0.2487 + n * 0.0025, n = 0, ... 185.
[0018] Although these four sets of sequences have overlapping equivalent phase centers, they make full use of the 384 equivalent phase centers. The spatial sampling along the Y-axis is uniform, so 96xJ, 96xJ, 96xJ, and 186xJ spatially uniform sampling data can be formed in the XY plane, ultimately forming four sets of three-dimensional data: Adata[k][i][j] (k=1,….K;i=1,2,…96;j=1,2..J), Bdata[k][i][j] (k=1,….K;i=1,2,…96;j=1,2..J), Cdata[k][i][j] (k=1,….K;i=1,2,…96;j=1,2..J) and Ddata[k][i][j] (k=1,….K;i=1,2,…186;j=1,2..J).
[0019] This invention employs the RMA algorithm for imaging the four sets of data mentioned above, which can improve imaging efficiency and reduce computational complexity by utilizing the Fast Fourier Transform.
[0020] The terahertz imaging method based on CEEMD noise reduction proposed in this invention has the following specific steps: (1) Perform CEEMD noise reduction on the sampled data in the fast time dimension.
[0021] Taking the fast-time data Data[k][1][1] (where k=1,….K) at j=1 and k=1 as an example, first calculate the mean m_11 of these K data points, and let: Data[k][1][1]= Data[k][1][1]-m_11 (3); Perform CEEMD decomposition on Data[k][1][1], setting the number of decompositions to P (P is generally no greater than 3), and obtain P IMF components. Then, stop the decomposition and set: Data[k][1][1] = Data[k][1][1] - (4).
[0022] Let i = 1, 2, ... 384; j = 1, 2.. J, and perform noise reduction on the 384 x K fast time dimension data according to formulas (3) and (4).
[0023] (2) According to formula (2), the three-dimensional data Data after step 1 is decomposed into four groups of data Adata, Bdata, Cdata and Ddata that are uniformly sampled on the XY plane.
[0024] (3) Using the RMA method, images were formed by imaging the four sets of data respectively. Since the equivalent phase center interval of the first three sets of data Adata, Bdata, and Cdata on the X-axis is 0.005 meters, the second dimension of these three sets of data (i.e., the dimension i=1,2,…96) was interpolated by a factor of 2 to make the interval 0.0025m. In this way, the equivalent phase center interval of the four sets of data on the X-axis is the same.
[0025] (4) Perform coherent accumulation on the four images formed by the four sets of data. During accumulation, it is necessary to identify the same position of the target in the image and use that position as the alignment center for accumulation.
[0026] The method of this invention can reduce computational complexity and improve computational efficiency.
Claims
1. A terahertz imaging method based on CEEMD noise reduction, characterized in that, Includes the following steps: (1) The sampled data in the fast time dimension is decomposed using CEEMD to calculate the first P IMF components; the original signal is subtracted from the first P IMF components; (2) Decompose the three-dimensional data processed in step (1) into four groups of data that are uniformly sampled on the XY plane; (3) Using the wavenumber domain-based RMA algorithm, the four sets of data were processed to form images; (4) Coherently accumulate the four images formed by the four sets of data to form the final image.
2. The terahertz imaging method based on CEEMD noise reduction according to claim 1, characterized in that, For a non-uniformly distributed sparse array, four equally spaced equivalent phase center sequences are found and processed using an RMA algorithm based on the wavenumber domain.
3. The terahertz imaging method based on CEEMD noise reduction according to claim 1 or 2, characterized in that, In step (3), when performing RMA algorithm processing, the echo data is processed using Fast Fourier Transform.
4. The terahertz imaging method based on CEEMD noise reduction according to claim 1 or 2, characterized in that, In step (4), during accumulation, the same position of the target in the image is identified, and accumulation is performed with that position as the alignment center.
5. The method according to claim 1 or 2, characterized in that: This method is applicable to sparse MIMO arrays.