A MEMS-assisted handheld MIMO-SAR security inspection imaging method

Abstract

The present invention discloses a handheld security inspection imaging method based on MEMS assistance. The present invention is intended to meet the needs of rapid and precise security inspection of personnel in crowded security inspection sites, and combines MEMS sensors with MIMO radar boards to form a high-precision, low-cost, convenient handheld security inspection imaging method. First, the radar board transmits a linear frequency modulation signal and receives the echo signal reflected by the target. During this period, the MEMS sensor feeds back the motion trajectory and position information of the handheld security inspection system in real time, providing accurate posture and position information. Secondly, the position of the MIMO radar antenna trajectory information is calibrated through the data of the MEMS sensor, and the accelerometer zero drift is compensated based on the wavelet packet change to reduce the error caused by the system movement and position change. Finally, the residual phase error is estimated and eliminated based on the particle swarm algorithm. On this basis, the wavenumber domain imaging algorithm is combined to realize non-contact three-dimensional imaging of the target. Finally, deep learning, image recognition and other technologies are combined to realize dangerous goods inspection.

Description

Technical Field

[0001] The present invention belongs to the field of radar signal processing and relates to a MEMS-assisted handheld MIMO-SAR security inspection imaging method. Background Art

[0002] Millimeter-wave Multiple Input Multiple Output Synthetic Aperture Radar (MIMO-SAR) near-field imaging technology, with its high resolution and penetrating capabilities, has become a new approach for security inspection applications. However, existing millimeter-wave radar security inspection equipment is generally bulky and difficult to install and use, limiting its application and flexibility in certain environments.

[0003] The development of miniaturized, handheld security inspection equipment can greatly facilitate operations for security personnel, making the inspection process more flexible and efficient, thereby broadening the application prospects of millimeter-wave radar technology. On the one hand, handheld security inspection equipment can use the swing of the arm to form a synthetic aperture. Combining the real aperture formed by the MIMO array with the bandwidth of high-frequency radar signals such as millimeter-wave terahertz, it can obtain high-resolution three-dimensional images of the target to be inspected. On the other hand, handheld security inspection methods are more easily applied in crowded or confined spaces, such as large-scale events and public transportation systems, improving the convenience and practicality of security inspections.

[0004] However, in practical applications, handheld millimeter-wave radar devices face some technical difficulties, especially in terms of swing trajectory estimation and swing error compensation. To address these problems, this patent designs a radar imaging method based on the assistance of microelectromechanical systems (MEMS). This method uses MEMS sensors to obtain coarse information on device acceleration and position, compensates for zero drift on MEMS sensor data through efficient signal processing algorithms, and accurately estimates motion parameters from radar echoes based on the particle swarm optimization minimum entropy algorithm, further eliminating phase errors, thereby achieving high-precision three-dimensional imaging of the object to be detected. Based on this three-dimensional image, deep learning, image recognition and other methods can be used to detect dangerous goods. This new handheld imaging technology will provide broader application prospects for the application of millimeter-wave radar in security inspection and other fields. Summary of the Invention

[0005] The purpose of this invention is to improve the mobility and flexibility of the security inspection imaging system by using a handheld MEMS and MIMO-SAR radar system, combining MEMS signal and radar signal processing methods, enhancing the quality and accuracy of security inspection imaging, simplifying operations and improving user friendliness. Its application scenarios include Figure 1 It specifically includes the following steps:

[0006] The first step is to use a handheld radar system containing multiple transmitting antennas and receiving antennas for scanning. That is, the MIMO radar system transmits signals one by one, and the receiving antennas receive the signals in sequence.

[0007] The signal is sent from the mth transmitting antenna, reflected by the target, and received by the nth receiving antenna. The position of the target point from the transmitting and receiving antennas is expressed as:

[0008]

[0009]

[0010] Among them, (x, y, z) is the target position, and are the locations of the transmitting and receiving antennas.

[0011] The reflected signal received by the nth receiving antenna from the mth transmitting antenna is:

[0012]

[0013] in, is the instantaneous distance, σ p is the reflection coefficient of target P at position (x, y, z), and O is the set of all point targets in the imaging scene. The synthesizer generates a linear frequency modulated pulse, which is transmitted by the transmit antenna and received by the receive antenna. The chirp is then mixed with the received signal by the mixer to produce the processed signal.

[0014] Step 2: Swing the handheld MIMO-SAR radar device to form a three-dimensional imaging model, where the position of the antenna will change with the swing of the arm. The swing diagram is shown as follows Figure 2 This change can be described as follows:

[0015] Assume that the position of the antenna changes with time due to the swing of a person's arm. After the original antenna position changes, the arm swing will cause the azimuth and elevation angles to change. This angle changes with time t. This change leads to the phase φ in the signal model. nm (t) also changes with the change of antenna position. The real aperture of the MIMO antenna itself determines the system's resolution in one-dimensional space. Each antenna element can be regarded as an independent sensor that can capture signals from a specific direction. Through the movement of the arm, the MIMO system uses the principle of synthetic aperture radar to enhance imaging capabilities. The system's movement path allows it to collect data from different angles. Through the comprehensive processing of this data, it can construct a two-dimensional plane image of the target object, that is, to obtain the resolution of the antenna movement direction. Its azimuth resolution ρ az Expressed as:

[0016]

[0017] Among them, L SAR The bandwidth of the signal is a key factor in determining the range resolution of the radar system. In three-dimensional imaging, this allows the system to distinguish multiple targets that are very close together, thereby providing depth information to obtain range resolution. Range resolution ρ range It can be expressed as:

[0018]

[0019] By combining these three types of resolution, the MIMO antenna system can create a detailed three-dimensional image of the target area by fusing information from different dimensions.

[0020] Step 3: Use MEMS sensors to obtain the three-dimensional motion parameters of the security scanner, including acceleration, angular velocity, and roll axis parameters. Based on the preliminary three-dimensional parameters of the MEMS device, the antenna position is determined. Make corrections and thus update the signal model.

[0021] The corrected antenna position is used to update the signal model:

[0022]

[0023]

[0024] Based on the new antenna position, the signal reflection signal is recalculated and the estimated value is Constructing distance, imaging, and then obtaining a 3D radar image of the target based on the following imaging mechanism:

[0025]

[0026] The signal model is combined with wavenumber domain processing and wavenumber domain approximation (WDBA) is used. First, based on the wavenumber domain support domain approximation theory, an efficient 3D reconstruction model is developed. The wavenumber support domains of the scattering point and the reference point are approximately the same. Based on the wavenumber support domain, the wavenumber of the scattering point can be expressed as:

[0027]

[0028]

[0029]

[0030] in, and represent the 3D wavenumber support domains for emission and reception, respectively.

[0031] Taylor expansion is used to process the wavenumber domain approximate phase error (WDBAPE), analyze and correct the 3D distortion caused by the first-order terms, and establish a 3D geometric mapping from the WDBA imaging position to the actual position to correct these distortions.

[0032] Φ=(k' x x+k' y y+k' z z)+Δφ (12)

[0033] where Δφ is the phase error term, given by:

[0034]

[0035] After processing the distortion, an adaptive sub-scene segmentation and stitching strategy is employed to reduce computational overhead and handle large datasets. This involves effectively covering the imaging target area rather than the entire scene, thus reducing the number of necessary sub-scenes and optimizing the processing.

[0036] Each sub-scene is processed independently in the wavenumber domain, and a 3D inverse fast Fourier transform (3D-IFFT) is used to reconstruct a 3D image based on the wavenumber domain signal:

[0037]

[0038] Through the above detailed steps, we not only improved the accuracy of signal processing, but also significantly reduced the computational complexity of the algorithm, making it more suitable for real-time and large-scale security screening applications.

[0039] Step 4: Although the antenna position has been corrected, the new antenna position (x' a ,y' a ,z' a ) may still have errors, which will cause phase errors, thereby affecting the imaging quality and causing the image to be blurred, defocused, and other problems.

[0040] The motion error model is applied to the radar echo signal to calculate the imaging result before correction. The output signal data of the MEMS accelerometer, including the signal with zero drift error, can be collected and the accelerometer output signal is subjected to wavelet packet decomposition to decompose the signal into sub-signals in different frequency bands. The formula is as follows:

[0041]

[0042] in is the wavelet packet coefficient of the kth node in the jth layer, ψ j,k (t) is the wavelet packet basis function, and x(n) is the original signal. Features are then extracted from the sub-signals after wavelet packet decomposition. The features may include the statistical characteristics of the sub-signals. Using the features obtained from wavelet packet decomposition as input and the zero drift error as the target output, a discrete high-pass filter is designed to test its performance. The loss function can be the mean square error:

[0043]

[0044] Among them, y i is the actual zero drift error, is the zero-drift error after filtering. Perform wavelet packet decomposition on the new data, extract features, and input them into the compensation model to calculate the zero-drift error. Based on the results, zero-drift compensation is performed on the accelerometer output signal. Verify and evaluate the compensated signal to ensure effective compensation. Model performance can be evaluated by experimentally comparing signal stability and accuracy before and after compensation.

[0045] The accuracy of the antenna position information obtained by MEMS will seriously affect the accuracy of the imaging results. The phase error caused by the distance error will cause serious deformation and defocus in the imaging results, affecting the image quality. Therefore, phase error correction is necessary.

[0046] Step 5: During the imaging process, due to the dynamic changes in the antenna position, phase error becomes a key factor affecting the imaging quality. This phase error is mainly caused by the non-ideal movement of the antenna (such as the swing of the arm), resulting in a phase deviation between the received signal and the actual reflected signal. Use a small amount of data to perform minimum entropy and particle swarm algorithm to estimate the phase error of the antenna. Specifically, set the initial particle swarm parameters and calculate the image entropy as the cost function. The phase formula is:

[0047]

[0048] in

[0049]

[0050]

[0051] To accurately compensate for phase errors, we first obtain the real-time antenna position using a MEMS sensor and initialize the particle swarm parameters based on this position information. Each particle represents a set of potential antenna parameters (velocity and acceleration). The particle swarm is optimized to estimate the parameters that will achieve the best imaging quality. is the position of particle i at the beginning, is the velocity of particle i at the start. For each particle, the imaging result is calculated under the current parameters, and the image entropy is used as the cost function to evaluate the imaging quality. The image entropy formula is as follows. The particle state is adjusted using the following update rule: the particle position and velocity are updated, and the optimal solution is iterated.

[0052]

[0053] Among them, p k Is the probability that the pixel intensity in the imaging result is k. Update the position and velocity of the particle to find the optimal solution:

[0054]

[0055]

[0056] Among them, w is the inertia weight, which controls the influence of the previous speed, c1 and c2 are learning factors, r1 and r2 are random numbers in the interval [0,1], and p i is the optimal position currently found for particle i, and g is the global optimal position. Optimization stops if the maximum number of iterations is reached or the change in particle position is less than a certain threshold. Re-imaging the target using the optimal parameters can eliminate phase errors and produce higher-quality images.

[0057] Step 6: Compensate for phase errors in all data using the optimal parameters found by the particle swarm optimization algorithm, perform fine imaging on the compensated signals, and obtain high-quality 3D imaging results, ensuring that all motion and phase errors are corrected:

[0058]

[0059] Before performing a three-dimensional Fast Fourier Transform (FFT), the updated signal data is interpolated. Interpolation is used to estimate values ​​between discrete data points, making the data converted to the frequency domain smoother and more continuous.

[0060] If V(x,y,z) is the corrected data, the interpolated data V′(x′,y′,z′) can be expressed as:

[0061]

[0062] Perform a three-dimensional FFT on the interpolated data to convert the data from the time domain to the frequency domain, which is necessary for high-quality imaging. The mathematical expression of FFT is:

[0063]

[0064] Among them, (kx ,k y ,k z ) is the coordinate in the frequency domain, is the data in the frequency domain.

[0065] The final imaging result is constructed using the output of the 3D FFT. The complex frequency domain data is converted back to real time domain data, and the amplitude information is extracted as the imaging result. The imaging result is analyzed to confirm that all phase and position deviations caused by motion have been properly compensated. Based on the preliminary imaging results, interpolation and FFT parameters are further adjusted to optimize the imaging quality. Ensure that all steps work together to achieve optimal imaging performance.

[0066] This additional step ensures that the radar system can effectively process and compensate for any phase errors caused by motion when performing complex dynamic imaging. At the same time, it uses high interpolation and FFT processing to speed up imaging while ensuring high quality and high accuracy of imaging results.

[0067] Step 7: Use deep learning and other methods to detect dangerous goods. First, collect a large amount of MIMO-SAR imaging data, including image data of known categories of dangerous goods and non-dangerous goods. These data will be used to train deep learning models. During the deployment phase, the imaging data provided by the MIMO-SAR system is received in real time, and the CNN model quickly identifies and classifies each scanned object. The model will output the classification probability of each object. The system determines whether it is a dangerous good based on the probability threshold and feeds back the results to the operator in real time. In addition, in order to adapt to the ever-changing security inspection environment and emerging types of dangerous goods, the model needs to be regularly learned and updated online. By continuously monitoring the performance of the model and collecting new data samples for model fine-tuning, ensure that the system remains efficient and accurate in new security inspection scenarios. The system workflow diagram is as follows Figure 3 shown.

[0068] The present invention offers advantages not found in traditional imaging methods: First, it provides timely feedback and is suitable for dynamic environments (handheld devices can provide real-time or near-real-time imaging results, which is crucial for applications requiring immediate diagnosis and decision-making. Handheld devices are particularly well-suited for use in dynamic or changing environments because the operator can instantly adjust their position and angle to adapt to on-site conditions). Second, it offers high resolution and accuracy (by using linear frequency modulation signals, it can accurately capture the distance, velocity, and angle information of the target object. This technology provides high-resolution imaging and more accurate capture of details). Third, it offers strong anti-interference capabilities (combined with wavelet packet transform methods, it can effectively compensate for zero drift, thereby improving the system's resistance to external interference. Traditional methods often struggle to handle interference in complex environments). Third, it enables automatic calibration and error compensation (by automatically calibrating and compensating for the systematic errors of MEMS sensors, it ensures data accuracy and reliability during the imaging process, effectively reducing systematic errors). Fourth, it enables fast non-uniform imaging (with the help of a fast non-uniform imaging algorithm, it can efficiently process acquired data, accurately process non-uniform sampling points, and improve imaging speed and quality. This advantage gives this method significant advantages in real-time imaging and large-scale data processing).

[0069] After adopting the above-mentioned solution, the present invention can achieve the following beneficial effects:

[0070] 1. By acquiring information such as the distance, speed, and angle of an object through the radar system, it is possible to accurately locate and track the target. This multi-dimensional information can improve the accuracy and reliability of target tracking, especially in dynamic environments.

[0071] 2. Radar systems have strong anti-interference capabilities against ambient light changes and obstructions, and can operate stably in complex environments. In contrast, optical tracking systems may experience performance degradation in low light conditions or when there is severe reflection interference.

[0072] 3. Combined with the wavelet packet transform method of MEMS sensors, effective zero drift compensation and error calibration are performed, improving the overall accuracy of the system.

[0073] This is especially important in long-term monitoring and dynamic target detection.

[0074] 4. A fast non-uniform imaging algorithm is used to process the collected data, and the non-uniform sampling points are processed through the wavenumber domain interpolation method, which improves the data processing speed and imaging accuracy, enabling the system to respond to and process large amounts of data in real time.

[0075] 5. Using a particle swarm optimization algorithm to estimate antenna parameters and compensate for phase errors, the system can adapt to complex motion trajectories and ensure accurate imaging and tracking. This flexibility enables the system to perform better in diverse practical application scenarios.

[0076] 6. By establishing an echo signal model and analyzing scattering points, the radar system can simultaneously identify and monitor multiple targets, making it suitable for applications in multi-target environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] Figure 1 A schematic diagram of the application scenario.

[0078] Figure 2 Schematic diagram of swing

[0079] Figure 3 System workflow diagram DETAILED DESCRIPTION

[0080] Step 1: Deploy a MIMO radar system with multiple transmit and receive antennas. The transmit antennas transmit signals one by one, and the receive antennas receive signals reflected by the target one by one. The signal model is defined as: By receiving signals reflected from a target, radar can measure the target's distance, speed, and angle.

[0081] Step 2: As the arm swings, the antenna position changes. First, the target direction is determined using the antenna array's angular resolution. Then, synthetic aperture technology is used to enhance horizontal image details. Finally, high-bandwidth signals are used to achieve precise depth resolution.

[0082] Step 3: Process the data obtained after scanning, including the acceleration and angular velocity obtained by the sensor, correct the preliminary three-dimensional parameters of the antenna, and update the signal model. The reconstruction imaging algorithm is

[0083] Step 4: Correct the antenna position error and use the wavelet packet transform method to compensate for the system error of the MEMS sensor. The signal decomposition formula is: as well as h and g are the filter coefficients of the wavelet transform.

[0084] Step 5: Compensate for phase error, optimize the parameters using particle swarm optimization algorithm, evaluate the fitness of each particle and iteratively update the particle state until convergence. The update formula is: i =v i,0 + learning factor × (local optimal position - current position), x i =x i,0 +v i .

[0085] Step 6: Apply the optimal parameters found by the above particle swarm algorithm to compensate for the phase error, interpolate the compensated data, set the number of samples to N, and perform three-dimensional FFT on the calculated results. And perform high-precision imaging processing.

[0086] Step 7: Collect a large amount of dangerous goods image data with known classifications as a training set and use data augmentation techniques to improve the model's generalization and robustness. Using a transfer learning strategy, you can use pre-trained network structures and parameters as a starting point, accelerating the learning process and improving recognition accuracy.

[0087] The above description is only a specific embodiment of the present invention. Any feature disclosed in this specification, unless otherwise stated, can be replaced by other equivalent or alternative features with similar purposes; all disclosed features, or all steps in the methods or processes, except for mutually exclusive features and / or steps, can be combined in any way.

Claims

1. A MEMS-assisted handheld MIMO-SAR security inspection imaging method, comprising the following steps: Step 1: The handheld radar scans the target to be inspected. The MIMO radar transmits a linear frequency modulation signal, which is reflected by the target and received by the antenna. The reflected signal received by the nth receiving antenna from the mth transmitting antenna can be abbreviated as σ p is the reflection coefficient of the target P at the (x, y, z) position, O is the set of all point targets in the imaging scene, and f = f c ±B / 2 is the signal frequency domain, f c , B are the center frequency and bandwidth of the signal, t is time, c is the speed of light, and path The transmitting antenna position Receiving antenna position It will change over time, causing phase changes; The second step is to use the handheld MIMO radar to construct a MIMO-SAR three-dimensional imaging model through arm movement. This model first uses the real aperture of the antenna array to obtain resolution in the array direction. Second, the model uses synthetic aperture technology to obtain resolution in the direction of antenna motion. Finally, the high-bandwidth signal is used to obtain resolution in the MIMO array normal, or range direction. This imaging model provides the system with three-dimensional imaging capabilities. Step 3: Obtain the three-dimensional motion parameters of the security scanner based on the MEMS device, build a MIMO-SAR signal model, and use the estimated value of the antenna coordinates Establish instantaneous distance Based on the idea of ​​matched filter imaging algorithm, the uncorrected imaging results can be expressed as follows Then, based on the wavenumber domain approximation, an efficient three-dimensional model is established; Step 4: To address the systematic error issues of MEMS sensors, we combine the wavelet packet variation method to effectively compensate for zero drift. We perform wavelet packet decomposition on the accelerometer's output signal to extract the characteristics of each frequency band. We then design a discrete high-pass filter using the processed data as input and the zero drift error as the target output. Finally, we integrate the filter into the sensor's operating environment, continuously monitor its performance, and adjust the model based on new conditions or problems that arise in actual applications to predict and compensate for new data, thereby achieving zero drift error calibration. Step 5: For the remaining phase error, a compensation phase error estimation model can be established to convert the phase error problem into an optimization problem. The optimal solution of the problem is then found through the particle swarm algorithm. The static parameter vector to be estimated is initialized. Each particle represents a possible parameter set. The fitness of each particle is evaluated according to the degree of matching between the radar echo signal and the theoretical model. The parameters are optimized based on the optimal position of the particle itself and the optimal position of the group. and The position and velocity of particle i at the beginning are updated according to the optimal position of the particle itself and the optimal position of the group. The iterative steps are repeated until the error threshold is reached. The compensation phase is constructed based on the obtained optimal static parameters to compensate the phase of the echo signal to eliminate the remaining phase error. Step 6: After compensating for phase errors, a non-uniform imaging algorithm is used to achieve fast and precise imaging. Three-dimensional interpolation technology is used to fill in data gaps caused by scanning intervals to ensure imaging continuity and smoothness. Processing is then performed based on the distance from the reference point to the transmitting and receiving antennas, as well as the three-dimensional wavenumber support domain of the echo corresponding to the reference point. Step 7: Integrate deep learning technology into the MIMO-SAR security imaging system to achieve automatic identification and classification of dangerous goods. First, use the high-precision three-dimensional imaging data generated in the previous steps to extract the characteristics of the target object, including shape, size, and scattering intensity information. Then build a deep convolutional neural network model for training and identifying different types of dangerous goods.

Images