Pre-scan range selection and adaptive spiral scan real-time sar imaging method and system

Abstract

This invention discloses a pre-scan range selection and adaptive spiral scanning real-time SAR imaging method and system, belonging to the field of radar signal processing technology. The method includes: obtaining average pre-scan echo frame data through multi-line pre-scanning and element-wise averaging; automatically determining target range cells based on energy and variance characteristics; generating an adaptive center-expanding square spiral trajectory centered on a user-specified starting point for scanning; selecting effective frames in the uniform velocity segment based on timestamps in each motion segment and uniformly resampling them to fill a three-dimensional data cube; performing focused imaging using a wavenumber domain focusing operator; and triggering a refresh display after each complete scan. This invention achieves automatic selection of target range cells and priority imaging of the region of interest, ensuring data acquisition quality and improving real-time imaging efficiency.

Description

Technical Field

[0001] This invention relates to the field of Synthetic Aperture Radar (SAR) imaging and real-time signal processing technology, and in particular to a pre-scan range selection and adaptive spiral scanning real-time SAR imaging method and system. Background Technology

[0002] Synthetic Aperture Radar (SAR) utilizes platform motion to synthesize an equivalent long aperture in the azimuth direction, enabling the acquisition of high-resolution two-dimensional / three-dimensional imaging results under all-weather and all-day conditions. It has been widely used in remote sensing mapping, target detection and identification, industrial non-destructive testing, and security inspection. In recent years, with the improvement of millimeter-wave / centimeter-wave devices, miniaturized radar front-ends, motion control, and computing platform performance, near-range SAR imaging has gradually shifted from offline weather processing to real-time "scan-while-imaging" technology. The system typically uses a two-dimensional motion platform to drive the antenna or target to perform azimuth and elevation grid scanning, acquiring intermediate frequency (IF) echo data containing multi-point fast-time sampling at each azimuth sampling location, and performing range compression and focusing imaging on the host computer. This allows for continuous updating of the visual image during the scanning process to support rapid on-site judgment and interactive parameter adjustment.

[0003] However, several key limitations remain in engineering implementation: Near-field echoes are susceptible to antenna coupling leakage, static clutter, and installation location; target scattering may span multiple range cells, leading to frequent reliance on manual experience for range gate / range cell selection, which can easily result in errors. Simultaneously, platform acceleration / deceleration phases, serial port frame drops, and serpentine trajectory direction switching introduce inconsistencies in azimuth sampling. Furthermore, the computational load of processing links such as range compression, wavenumber domain focusing, and amplitude compression is substantial, making it difficult to balance real-time imaging quality and efficiency. In addition, existing systems generally employ a serpentine grid scanning trajectory, i.e., the pitch axis is stepped row by row, while the azimuth axis is scanned in a reciprocating straight line. This scanning method strictly follows the row sequence from bottom to top, preventing users from specifying regions of interest for priority imaging. In large-aperture scanning scenarios, the imaging results for the target area must be observed only after all scans are completed, resulting in insufficient real-time performance.

[0004] To address the range cell selection problem, existing methods largely rely on operators' manual interpretation of radar echo spectra or pre-setting fixed range cells based on prior knowledge. The former requires operators to possess professional signal processing knowledge, while the latter lacks adaptability in application scenarios where target distances are uncertain. Furthermore, the results of a single pre-scan are susceptible to random noise and occasional clutter interference, leading to unstable range cell selection results.

[0005] To address the scanning trajectory issue, spiral scanning, as a scanning strategy that expands outward from the center, can prioritize data coverage of the area of ​​interest, theoretically improving the response speed of real-time imaging. However, how to design a non-overlapping square spiral path on a rectangular grid, how to accurately select effective frames matching the grid step number from continuously acquired frames of variable-speed motion in each movement segment of the spiral scan, and how to achieve incremental real-time imaging refresh during the spiral expansion are all technical problems that urgently need to be solved.

[0006] Therefore, there is an urgent need for a SAR imaging method that can automatically determine the target distance cell through multi-line pre-scanning and averaging before imaging, expand the acquisition outward along a spiral trajectory centered on the user's area of ​​interest during the acquisition process, complete the selection of effective frames and uniform resampling in each motion segment, and achieve real-time focusing display by loop by loop by reusable focusing operators. Summary of the Invention

[0007] In view of this, the purpose of this invention is to provide a pre-scan range selection and adaptive spiral scanning real-time SAR imaging method and system. This method uses multi-line pre-scanning and energy-variance comprehensive scoring to automatically select target range cells, and designs a square spiral trajectory that expands outward from the center of the region of interest, thus solving the problems of reliance on manual labor, imaging lag in the region of interest, and uneven data.

[0008] To achieve the above objectives, the present invention provides the following technical solution: The pre-scan range selection and adaptive helical scanning real-time SAR imaging method provided by this invention includes the steps of controlling a two-dimensional motion platform to scan the scene under test and acquiring echo frame data, and also includes the following steps: S1: Control the two-dimensional motion platform to perform azimuth dimension linear pre-scan at multiple preset pitch positions in the scene to be tested, collect the pre-scan echo frame data of each row, and perform element-wise averaging on the multiple rows of pre-scan echo frame data to obtain the average pre-scan echo frame data. S2: Perform static clutter suppression and fast time-dimensional spectrum analysis on the average pre-scan echo frame data, calculate a comprehensive score based on the energy characteristics and variance characteristics of each range cell, and automatically determine the target range cell based on the comprehensive score; S3: Based on the target distance unit and the user-specified starting grid point, control the two-dimensional motion platform to perform formal scanning according to the adaptive center-outward square spiral trajectory. In each motion segment, real-time echo frame data is collected. Based on the motion parameters of each segment and the timestamp of the echo frame, select the effective frame of the uniform speed segment and uniformly resample it to the same number of frames as the grid step number of that segment. Fill the three-dimensional data cube point by point with the resampled echo frame data. S4: Perform Hilbert transform and distance compression on the three-dimensional data cube, extract two-dimensional data slices based on the target distance unit, and use wavenumber domain focusing operators to perform two-dimensional focusing to obtain a complex value focusing map; S5: Perform amplitude normalization and logarithmic compression on the complex-valued focusing image, and output the imaging result; During the execution of steps S3 to S5, after each complete spiral scan and update of the three-dimensional data cube, steps S4 and S5 are triggered to refresh the displayed imaging results.

[0009] Furthermore, in step S1, the plurality of preset pitch positions are at least two different pitch positions selected within the pitch range of the scene to be tested.

[0010] In step S1, multiple preset pitch positions are selected as representative rows; for each representative row, the azimuth axis is controlled to perform a full-length linear scan at a preset scanning speed and acquire echo frame data; after aligning the data matrix of each representative row with the number of frames, the arithmetic mean is taken element by element along the row dimension. Furthermore, in step S2, calculating the comprehensive score further includes: calculating the energy characteristics and variance characteristics of each distance cell, normalizing them, and then linearly combining them according to preset weights to obtain the comprehensive score. ; in, This represents the overall score of the k-th distance unit; This represents the normalized energy of the k-th distance cell; This represents the normalized variance of the k-th distance cell; Indicates the distance unit number; and The weighting coefficients and ; After setting the comprehensive score of a preset number of near-range units corresponding to the radar blind zone to zero, the range unit with the highest score is selected as the target range unit.

[0011] Furthermore, in step S3, the method for generating the adaptive center-expanding square spiral trajectory is as follows: Starting from the user-specified grid point Taking the center of the spiral as the minimum distance from the starting point to the four boundaries as the maximum spiral radius. ; Starting from the first rotation, expand outwards with each rotation until the maximum spiral radius is reached. The circle is executed in five segments: "positive azimuth, positive pitch, negative azimuth, negative pitch, and positive azimuth," with the number of steps for each segment being as follows: , , , , This ensures that each grid point is visited exactly once.

[0012] Furthermore, in step S3, selecting valid frames of the uniform speed segment for each motion segment and uniformly resampling further includes: The existence of a uniform speed interval is determined based on the acceleration, deceleration, and scanning speed parameters of the motion segment. If it exists, the frames within the uniform speed segment are selected as candidate sets using the timestamps of each frame. The number of frames with the same grid step number is extracted according to the equally spaced index. If the number of candidate frames is insufficient, the first and last frames are symmetrically padded to make up the difference.

[0013] Furthermore, in step S4, the wavenumber domain focusing operator is pre-calculated and cached, and reused each time imaging is triggered; the focusing operator is calculated according to the following formula: ; in, Indicates the focusing operator; This indicates the distance from the target to the center of the cell; Indicates the longitudinal wavenumber; The According to The following formula is used for calculation: ; in, For free space wavenumber, These are the space wavenumbers in the azimuth and elevation directions.

[0014] Furthermore, in step S5, the process of refreshing the displayed imaging results is controlled by a mutex lock mechanism and a minimum refresh interval.

[0015] Furthermore, it also includes the step of controlling the two-dimensional motion platform to move to the user-specified starting grid point position for physical position verification before the formal scan; and the step of asynchronously saving the original echo frame data of each motion segment to the disk during the scan.

[0016] The pre-scan range selection and adaptive spiral scanning real-time SAR imaging system provided by this invention includes a two-dimensional motion platform, a radar front-end, a motion controller, and a host computer, wherein the host computer includes: The pre-scan control and acquisition module is used to control the two-dimensional motion platform to perform azimuth linear pre-scan at multiple preset pitch positions in the scene to be tested, acquire the pre-scan echo frame data of each row, and perform element-wise averaging of the multiple rows of pre-scan echo frame data to obtain the average pre-scan echo frame data. The target distance cell automatic selection module is used to perform static clutter suppression and fast time-dimensional spectrum analysis on the average pre-scan echo frame data, calculate a comprehensive score based on the energy characteristics and variance characteristics of each distance cell, and automatically determine the target distance cell according to the comprehensive score. The spiral scanning and data organization module is used to generate an adaptive center-expanding square spiral trajectory based on the target distance unit and the user-specified starting grid point, control the two-dimensional motion platform to perform formal scanning according to the trajectory, collect echo frame data in real time in each motion segment, select effective frames of the uniform speed segment based on the motion parameters of each segment and the timestamp of the echo frame, and uniformly resample to the number of frames consistent with the number of grid steps of that segment, and fill the three-dimensional data cube point by point with the resampled echo frame data. The real-time imaging processing module is used to perform Hilbert transform and distance compression on the three-dimensional data cube, extract two-dimensional data slices based on the target distance unit, and perform two-dimensional focusing using wavenumber domain focusing operators to obtain complex value focusing images. The display and refresh module is used to normalize the amplitude and logarithmically compress the complex-valued focusing image, output the imaging result, and trigger the real-time imaging processing module to process and refresh the display after each complete spiral scan and update of the three-dimensional data cube.

[0017] Furthermore, the target distance unit automatic selection module is also used to display the distance-time amplitude spectrum heatmap, comprehensive scoring curve and energy curve of the pre-scan echo data on the human-computer interaction interface, and mark the automatically recommended target distance units, while also supporting users to manually select and confirm the target distance units by clicking. The spiral scanning and data organization module generates an adaptive center-expanding square spiral trajectory by using a user-specified starting grid point. Taking the center of the spiral as the minimum distance from the starting point to the four boundaries as the maximum spiral radius. Starting from the first turn, expand outwards with each turn until the maximum spiral radius is reached. The circle is executed in five segments: "positive azimuth, positive pitch, negative azimuth, negative pitch, and positive azimuth," with the number of steps for each segment being as follows: , , , , This ensures that each grid point is visited exactly once; The spiral scanning and data organization module selects valid frames of the uniform speed segment and resamples them uniformly for each motion segment as follows: it determines whether there is a uniform speed interval based on the acceleration, deceleration and scanning speed parameters of the motion segment. If there is, it uses the timestamp of each frame to filter the frames in the uniform speed segment as a candidate set; it extracts the number of frames consistent with the grid step number according to the equally spaced index. If the number of candidate frames is insufficient, it uses the method of symmetrical padding of the first and last frames to make up the number.

[0018] The beneficial effects of this invention are as follows: This invention provides a pre-scan range selection and adaptive spiral scanning real-time SAR imaging method and system, belonging to the field of radar signal processing technology. The method includes: controlling a two-dimensional motion platform to perform azimuth-dimensional linear pre-scans at multiple representative elevation positions; averaging multiple rows of echo data element-wise to suppress random noise; automatically determining target range cells through static clutter suppression and spectral analysis combined with energy and variance characteristics; generating an adaptive center-expanding square spiral scanning trajectory centered on a user-specified starting grid point; selecting effective frames in the uniform speed segment based on timestamps and uniformly resampling in each motion segment; filling a three-dimensional data cube point-by-point; subsequently extracting two-dimensional slices based on the target range cells; rapidly focusing using a pre-calculated wavenumber domain focusing operator; and instantly refreshing and displaying the imaging results after each spiral scan. This invention automates range cell selection and prioritizes imaging of the scanning area of ​​interest, providing efficient and stable incremental real-time SAR imaging capabilities that gradually sharpen from the center outwards, suitable for close-range high-resolution detection and identification scenarios. This modified method, based on pre-scan automatic range cell selection and adaptive center-expanding square spiral scanning real-time synthetic aperture radar imaging process, has the following beneficial effects: 1. By using multi-line pre-scanning and element-by-element averaging, random noise interference is suppressed, and the target distance cell is automatically selected by combining energy-variance comprehensive scoring, eliminating the need for manual interpretation; 2. The adaptive center-expanding square spiral trajectory enables the region of interest to obtain data coverage and be imaged first, significantly improving the real-time response speed in large-aperture scanning scenarios; 3. The time-stamp-based uniform segment frame filtering and uniform resampling mechanism ensures the data quality of each grid point, and the non-overlapping path design avoids the waste of time due to repeated collection; 4. The incremental wavenumber domain focusing imaging, combined with the focusing operator multiplexing mechanism, enables efficient real-time imaging refresh.

[0019] The above and other objects, advantages, and features of the present invention will be more fully set forth and demonstrated through the following detailed description of specific embodiments in conjunction with the accompanying drawings. Those skilled in the art, upon referring to the following detailed description and the accompanying drawings, will be able to better understand and realize the above advantages of the present invention. Other objects, features, and advantages of the present invention will become clearer after being described in detail in the detailed description section in conjunction with the accompanying drawings. Attached Figure Description

[0020] To make the objectives, technical solutions, and beneficial effects of this invention clearer, the following drawings are provided for illustration.

[0021] Figure 1 This is a schematic diagram of the overall process of the real-time SAR imaging method in this embodiment.

[0022] Figure 2 This is a schematic diagram of the system composition and signal control relationship in this embodiment.

[0023] Figure 3 This is a schematic diagram of the target distance unit automatic recommendation interface in this embodiment.

[0024] Figure 4 This is a schematic diagram of the adaptive center-expanding square spiral scanning trajectory in this embodiment.

[0025] Figure 5 This is a schematic diagram of the selection of effective frames and uniform resampling in the uniform speed segment during a single-segment motion in this embodiment.

[0026] Figure 6 This is a schematic diagram of the distance compression and wavenumber domain focusing processing link in this embodiment.

[0027] Figure 7 This is a schematic diagram of the host computer software interface and real-time imaging refresh per circle in this embodiment.

[0028] Figure 8 This is a real-time imaging result image after the data acquisition is completed. Detailed Implementation

[0029] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments described are not intended to limit the present invention. Example 1

[0030] like Figure 1As shown, the pre-scan range selection and adaptive spiral scanning real-time SAR imaging method provided in this embodiment achieves intelligent selection of target range cells through multi-line pre-scanning and element-wise averaging processing, and realizes data acquisition and real-time imaging by expanding outward from the user's area of ​​interest in circles through an adaptive center-outward square spiral scanning trajectory. The method includes the steps of controlling a two-dimensional motion platform to scan the scene under test and acquire echo frame data. Its characteristic feature is that it further includes the following steps: S1: Control the two-dimensional motion platform to perform azimuth dimension linear pre-scan at multiple preset pitch positions in the scene to be tested, collect the pre-scan echo frame data of each row, and perform element-wise averaging on the multiple rows of pre-scan echo frame data to obtain the average pre-scan echo frame data. S2: Perform static clutter suppression and fast time-dimensional spectrum analysis on the average pre-scan echo frame data, calculate a comprehensive score based on the energy characteristics and variance characteristics of each range cell, and automatically determine the target range cell based on the comprehensive score; In step S2 of this embodiment, static clutter suppression is performed by subtracting the mean vector from the pre-scanned data matrix in the slow time dimension; after windowing, a one-dimensional FFT is performed to obtain the first half-spectral amplitude value to obtain the distance-time amplitude spectrum; the energy and variance of each distance cell are calculated, and after normalization, a comprehensive score is obtained by linear combination according to a preset weight; after setting the near-range distance cell to zero, the one with the largest score is selected as the target distance cell. S3: Based on the target distance unit and the user-specified starting grid point, control the two-dimensional motion platform to perform formal scanning according to the adaptive center-outward square spiral trajectory. In each motion segment, real-time echo frame data is collected. Based on the motion parameters of each segment and the timestamp of the echo frame, select the effective frame of the uniform speed segment and uniformly resample it to the same number of frames as the grid step number of that segment. Fill the three-dimensional data cube point by point with the resampled echo frame data. S4: Perform Hilbert transform and distance compression on the three-dimensional data cube, extract two-dimensional data slices based on the target distance unit, and use wavenumber domain focusing operators to perform two-dimensional focusing to obtain a complex value focusing map; In step S4 of this embodiment, after performing Hilbert transform along the pitch dimension, a range image is obtained by IFFT along the fast time dimension. Two-dimensional slices of the target range unit are extracted, and after being aligned with the size of the focusing operator, a two-dimensional FFT is performed, followed by a two-dimensional IFFT to obtain the focusing image. The focusing operator is calculated only once and reused in the entire real-time imaging process. S5: Perform amplitude normalization and logarithmic compression on the complex-valued focusing image, and output the imaging result; During the execution of steps S3 to S5, after each complete spiral scan and update of the three-dimensional data cube, steps S4 and S5 are triggered to refresh the displayed imaging results.

[0031] In step S1 of this embodiment, the plurality of preset pitch positions are at least two different pitch positions selected within the pitch range of the scene to be tested.

[0032] In step S1, multiple preset pitch positions are selected as representative rows; for each representative row, the azimuth axis is controlled to perform a full-length linear scan at a preset scanning speed and acquire echo frame data; after aligning the data matrix of each representative row with the number of frames, the arithmetic mean is taken element by element along the row dimension. In step S2 of this embodiment, calculating the comprehensive score further includes: calculating the energy characteristics and variance characteristics of each distance cell, normalizing them, and then linearly combining them according to preset weights to obtain the comprehensive score. ; in, This represents the overall score of the k-th distance unit; This represents the normalized energy of the k-th distance cell; This represents the normalized variance of the k-th distance cell; Indicates the distance unit number; and The weighting coefficients and ; After setting the comprehensive score of a preset number of near-range units corresponding to the radar blind zone to zero, the range unit with the highest score is selected as the target range unit.

[0033] In step S3 of this embodiment, the method for generating the adaptive center-expanding square spiral trajectory is as follows: Starting from the user-specified grid point Taking the center of the spiral as the minimum distance from the starting point to the four boundaries as the maximum spiral radius. ; Starting from the first rotation, expand outwards with each rotation until the maximum spiral radius is reached. The circle is executed in five segments: "positive azimuth, positive pitch, negative azimuth, negative pitch, and positive azimuth," with the number of steps for each segment being as follows: , , , , This ensures that each grid point is visited exactly once. Indicates the number of spiral turns (incrementing from 1); In this embodiment, step S3, selecting effective frames of the uniform speed segment for each motion segment and uniformly resampling, further includes: The existence of a uniform speed interval is determined based on the acceleration, deceleration, and scanning speed parameters of the motion segment. If it exists, the frames within the uniform speed segment are selected as candidate sets using the timestamps of each frame. The number of frames with the same grid step number is extracted according to the equally spaced index. If the number of candidate frames is insufficient, the first and last frames are symmetrically padded to make up the difference.

[0034] In step S4 of this embodiment, the wavenumber domain focusing operator is pre-calculated and cached, and reused each time imaging is triggered; the focusing operator is calculated according to the following formula: ; in, Indicates the focusing operator; This indicates the distance from the target to the center of the cell; Indicates the longitudinal wavenumber; the Calculate using the following formula: ; in, For free space wavenumber, , These are the space wavenumbers in the azimuth and elevation directions.

[0035] In step S5 of this embodiment, the process of refreshing and displaying the imaging results is controlled by a mutex lock mechanism and a minimum refresh interval.

[0036] This embodiment also includes the steps of controlling the two-dimensional motion platform to move to the starting grid point position specified by the user for physical position verification before the formal scan; asynchronously saving the original echo frame data of each motion segment to the disk during the scan; displaying the distance-time amplitude spectrum heat map, comprehensive scoring curve and energy curve of the pre-scan echo data on the human-computer interaction interface, and marking the automatically recommended target distance unit, while also supporting the user to manually select and confirm the target distance unit by clicking.

[0037] This embodiment also provides a pre-scan range selection and adaptive spiral scanning real-time SAR imaging system. This system achieves real-time synthetic aperture radar imaging based on pre-scan automatic range cell selection and adaptive center-expanding square spiral scanning. It includes a two-dimensional motion platform, a radar front-end, a motion controller, and a host computer. The host computer comprises: The pre-scan control and acquisition module is used to control the two-dimensional motion platform to perform azimuth linear pre-scan at multiple preset pitch positions in the scene to be tested, acquire the pre-scan echo frame data of each row, and perform element-wise averaging of the multiple rows of pre-scan echo frame data to obtain the average pre-scan echo frame data. The target distance cell automatic selection module is used to perform static clutter suppression and fast time-dimensional spectrum analysis on the average pre-scan echo frame data, calculate a comprehensive score based on the energy characteristics and variance characteristics of each distance cell, and automatically determine the target distance cell according to the comprehensive score. The spiral scanning and data organization module is used to generate an adaptive center-expanding square spiral trajectory based on the target distance unit and the user-specified starting grid point, control the two-dimensional motion platform to perform formal scanning according to the trajectory, collect echo frame data in real time in each motion segment, select effective frames of the uniform speed segment based on the motion parameters of each segment and the timestamp of the echo frame, and uniformly resample to the number of frames consistent with the number of grid steps of that segment, and fill the three-dimensional data cube point by point with the resampled echo frame data. The real-time imaging processing module is used to perform Hilbert transform and distance compression on the three-dimensional data cube, extract two-dimensional data slices based on the target distance unit, and perform two-dimensional focusing using wavenumber domain focusing operators to obtain complex value focusing images. The display and refresh module is used to normalize the amplitude and logarithmically compress the complex-valued focusing image, output the imaging result, and trigger the real-time imaging processing module to process and refresh the display after each complete spiral scan and update of the three-dimensional data cube.

[0038] In this embodiment, the target distance unit automatic selection module is also used to display the distance-time amplitude spectrum heatmap, comprehensive scoring curve and energy curve of the pre-scanned echo data on the human-computer interaction interface, and mark the automatically recommended target distance units. At the same time, it supports users to manually select and confirm the target distance units by clicking.

[0039] In this embodiment, the spiral scanning and data organization module generates an adaptive center-expanding square spiral trajectory by using a user-specified starting grid point. Taking the center of the spiral as the minimum distance from the starting point to the four boundaries as the maximum spiral radius. Starting from the first turn, expand outwards with each turn until the maximum spiral radius is reached. The circle is executed in five segments: "positive azimuth, positive pitch, negative azimuth, negative pitch, and positive azimuth," with the number of steps for each segment being as follows: , , , , This ensures that each grid point is visited exactly once.

[0040] In this embodiment, the spiral scanning and data organization module selects valid frames of the uniform speed segment and resamples them uniformly for each motion segment as follows: it determines whether there is a uniform speed interval based on the acceleration, deceleration and scanning speed parameters of the motion segment. If there is, it uses the timestamp of each frame to filter the frames in the uniform speed segment as a candidate set; it extracts the number of frames consistent with the grid step number according to the equally spaced index. If the number of candidate frames is insufficient, it uses the method of symmetrical padding of the first and last frames to make up the number.

[0041] In this embodiment, the real-time imaging processing module pre-calculates and caches the wavenumber domain focusing operator, which is then reused each time imaging is triggered; the focusing operator is calculated according to the following formula: ; in, Indicates the focusing operator; This indicates the distance from the target to the center of the cell; Indicates the longitudinal wavenumber; the Calculate using the following formula: ; in, For free space wavenumber, , These are the space wavenumbers in the azimuth and elevation directions.

[0042] In this embodiment, the display and refresh module uses a mutex lock mechanism and a minimum refresh interval for throttling control.

[0043] In this embodiment, the spiral scanning and data organization module is also used to control the two-dimensional motion platform to move to the starting grid point position specified by the user for physical position verification before the formal scanning.

[0044] In this embodiment, the spiral scanning and data organization module is also used to asynchronously save the original echo frame data of each motion segment to the disk during the scanning process. Example 2

[0045] like Figure 1 As shown, Figure 1 This is a schematic diagram of the overall process of the real-time SAR imaging method in this embodiment. This embodiment describes in detail the pre-scan range selection and adaptive spiral scanning real-time SAR imaging method, including the following steps: Step 1, System Connection and Parameter Configuration Process: like Figure 2 As shown, Figure 2This diagram illustrates the system composition and signal control relationships in this embodiment. It shows the hierarchical structure of the interaction between the hardware layer and the host computer software layer. The hardware layer includes the target / scene under test, a radar transmitter / receiver module (FMCW millimeter-wave front-end + acquisition card), a two-dimensional motion platform (azimuth axis X + pitch axis Y), and a motion controller (FMC4030). The radar module interacts with the target via electromagnetic wave transmission and reception; the radar module is mounted on the two-dimensional motion platform; the motion controller connects to the host computer via Ethernet (TCP), receives motion commands and feeds back position information, and simultaneously outputs drive signals to the two-dimensional motion platform. The host computer processing and display software layer includes six core functional modules: Pre-scan control and acquisition module: responsible for multi-line pre-scan and element-by-element averaging; Target distance cell automatic selection module: responsible for clutter suppression, FFT, energy / variance calculation and comprehensive scoring; Helical scanning and data organization module: responsible for trajectory generation, segment-by-segment acquisition, frame selection, resampling, and cube filling; Real-time imaging processing module: responsible for Hilbert transform, IFFT, slicing, multiplication of 2D-FFT with focusing operator, and 2D-IFFT; Display and refresh module: responsible for normalization, logarithmic compression, pseudo-color rendering, and status panel display; Human-computer interaction interface module: responsible for interactive functions such as RTI heatmap, scoring curve, starting point input, and mode switching.

[0046] Therefore, it can be seen that the host computer establishes a communication connection with the radar transmitter and receiver module through a serial port and sends acquisition control commands to the radar; at the same time, it establishes a connection with the two-dimensional motion platform controller and sets the azimuth axis travel of 400mm, the pitch axis travel of 400mm, and the azimuth sampling step size. =1mm (corresponding to 400 azimuth sampling points), pitch step =1mm (corresponding to 401 scan lines), azimuth scan speed 40mm / s, azimuth axis acceleration / deceleration 5000mm / s 2 Pitch axis acceleration / deceleration 1000 mm / s 2 Motion parameters. Each frame of echo data contains 512 fast time-dimension sampling points, with a raw byte length of 1024 bytes. Radar operating carrier frequency f. o =(58.5+3.25)×10 9 The sampling rate is 10MHz, and the frequency modulation slope is K=1.1818×10¹⁴Hz / s. The final three-dimensional data cube has dimensions of 401×400×512 (pitch × azimuth × fast time).

[0047] Step 2, Multi-line pre-scan and target distance cell automatic selection process: Before the formal helical scan, the system performs a multi-line pre-scan to obtain stable echo feature data.

[0048] like Figure 3 As shown, Figure 3 This is a schematic diagram of the target range unit automatic recommendation interface in this embodiment. In this embodiment, rows 182, 202, and 222, located in the center of the scene, are selected as representative rows for pre-scanning. For each representative row, the pitch axis is moved to the corresponding position and then fixed, while the azimuth axis performs a full-range unidirectional linear scan at a scanning speed of 40 mm / s. The radar front end is in continuous acquisition mode, acquiring multiple frames of pre-scan echo data for that row.

[0049] After three rows of scanning are completed, the data for each row is converted into a 512-row × frame-column data matrix. Since the actual number of frames acquired in each row may vary due to serial port transmission jitter, rows with fewer than 400 frames are padded symmetrically to 400 columns with the first frame on the left and the last frame on the right; rows with more than 400 frames are truncated to the first 400 columns. Then, the three 512×400 matrices are stacked along the dimension representing the rows and the arithmetic mean is taken to obtain a 512×400 average pre-scan data matrix.

[0050] By pre-scanning at multiple pitch positions in the middle of the scene and averaging the results, random noise and occasional clutter interference in a single scan are effectively suppressed, making the subsequent spectral characteristics more stable and reliable.

[0051] Automatic target range cell selection is performed on the averaged pre-scan data: First, the mean vector is subtracted in the slow time dimension to suppress static clutter and DC components; then, a Hanning window is applied in the fast time dimension and a one-dimensional FFT is performed, and the amplitude of the first half of the spectrum (256 range cells) is taken to obtain the range-time amplitude spectrum (RTI); for each range cell, the energy characteristics (frame amplitude summation) and variance characteristics (frame amplitude variance) are calculated, and after max-min normalization, a weighted linear combination is obtained to obtain the comprehensive score. ; in, This represents the overall score of the k-th distance unit; This represents the normalized energy of the k-th distance cell; This represents the normalized variance of the k-th distance cell; Indicates the distance unit number; The comprehensive score in this embodiment is calculated using the following formula: ; The scores of the first 5 near-range units are set to zero to eliminate DC leakage interference; the range unit with the highest score is selected as the target range unit.

[0052] like Figure 3 As shown, the system provides a human-computer interaction interface to display the RTI heatmap, comprehensive score curve and energy curve, and marks the automatically recommended distance units with an asterisk. Users can also manually select distance units and confirm the application by clicking with the mouse.

[0053] Step 3, Adaptive center-expanding square spiral scan trajectory generation process: like Figure 4 As shown, Figure 4 This is a schematic diagram of the adaptive center-expanding square spiral scanning trajectory in this embodiment.

[0054] Users input the 1-based coordinates of the starting grid point through the interactive interface (e.g., ...). The system converts to a 0-based index. Using this point as the center of the spiral, calculate the maximum spiral radius: ; in, This represents the image after logarithmic compression. This represents the image after amplitude normalization; in, The spiral trajectory covers a region centered at the starting point with a side length of [missing information]. The largest inscribed square area of 1 grid point; This indicates the maximum number of rotations in an adaptive, outward-expanding square spiral trajectory. In this embodiment =400、 =401.

[0055] Before performing the formal scan, the system allows users to move the platform to the designated location for verification using the "Reach Starting Point" button. Once verification is complete, the formal spiral scan will begin.

[0056] The spiral scan begins at the initial grid point, which serves as the first sampling point. Subsequently, it expands outwards circle by circle, starting from the first circle. Each lap follows a fixed sequence of five movements: Move 1 step in the positive direction; Pitch forward movement step; Negative azimuth movement step; negative pitch movement step; Directional movement step.

[0057] One lap total Step, after the r-th lap, the current position reaches This refers to the bottom right corner of this circle. Throughout the entire helical scan, each grid point is visited exactly once, ensuring complete and non-overlapping data acquisition. The system maintains a visited marker matrix, checking and marking it when writing to each grid point. If duplicate visits are detected, an abnormal termination is triggered to ensure path correctness.

[0058] in, Indicates the current number of spiral rotations; The beneficial effects of this step are: Compared to traditional serpentine grid trajectories, spiral trajectories expand outward from the user's area of ​​interest, ensuring that the area of ​​interest receives data coverage and imaging is completed first; the non-overlapping path design avoids the time wasted due to repeated acquisition.

[0059] Step 4, segment-by-segment motion acquisition and effective frame uniform resampling process: as follows Figure 5 As shown, Figure 5 This diagram illustrates the effective frame selection and uniform resampling during a uniform motion segment in this embodiment. Each motion segment corresponds to one single-axis linear motion, with the motion distance being the number of steps multiplied by the grid step size. This step addresses the mismatch between the number of acquired frames and the number of grid steps caused by the varying motion distances of each segment in helical scanning.

[0060] The specific process of selecting effective frames of the uniform speed segment for each motion segment and uniformly resampling is as follows: Based on the acceleration of the axis where this segment is located deceleration and scanning speed Calculate the acceleration stroke separately and deceleration stroke Determine the physical distance of this segment. Does it meet the requirements? Therefore, a uniform velocity interval exists. If a uniform velocity interval exists and the number of frames acquired is not less than the required number of grid steps... Then it is based on the first frame timestamp plus the acceleration time. The starting time of the uniform speed segment is obtained, and frames within the uniform speed interval are selected as candidate sets using the timestamps of each frame. If the condition is not met, all acquired frames are used as the candidate set.

[0061] If the number of candidate frames Extract by equally spaced index: ; like The first frame is padded with ⌊ missing number / 2⌋ on the left, and the last frame is padded with ⌈ missing number / 2⌉ on the right.

[0062] Resampled The 512-point fast-time dimensional signal of each frame is sequentially written into the corresponding grid positions of the 3D data cube according to the motion direction and step sequence. At the same time, the raw frame data of each motion segment is asynchronously saved to disk in MAT format, and the acquisition and saving are performed in parallel through a dual-thread mechanism.

[0063] in, This indicates the total number of valid candidate frames for the uniform velocity segment obtained from the current motion segment screening. Indicates the required number of grid steps; This indicates the frame sequence number index during equal-interval index extraction; Step 5, range compression and wavenumber domain focusing imaging process: as follows Figure 6 As shown, Figure 6 This is a schematic diagram of the distance compression and wavenumber domain focusing processing link in this embodiment. Range processing and two-dimensional focusing are performed on the three-dimensional data cube.

[0064] First, a Hilbert transform is performed on the current 3D data cube replica along the pitch dimension to obtain the analytical signal, and then a 512-point IFFT is performed along the fast time dimension to complete the range compression. This is based on the target range cell index determined in step 2. Extract the corresponding two-dimensional slices from the distance image data and set them as azimuth × pitch dimensions.

[0065] Subsequently, a wavenumber domain focusing operator was constructed.

[0066] A wavenumber axis of 512 points is constructed based on the azimuth and elevation sampling intervals. , ; Calculate the longitudinal wavenumber , in, is the free space wavenumber.

[0067] Distance from target to cell center , in, .

[0068] Focusing Operator ; right The evanescent wave region sets H to zero and performs FFT shifting along both axes.

[0069] in, Indicates the radar carrier frequency; Represents the speed of light; Indicates the resolution of the distance cell; Indicates the frequency modulation slope; Indicates the sampling period; Represents the number of points in the fast time-dimension FFT; The 2D slices are aligned with the focusing operator in size (symmetric filling for insufficient dimensions). A 2D FFT is performed on the slices, and then a 2D IFFT is performed after element-wise multiplication with H to obtain a complex-valued focusing image. The modulus value is extracted and spatially cropped according to a preset imaging size range (±250mm).

[0070] The focusing operator is calculated and cached during the first touch image and can be reused directly during subsequent refreshes, avoiding the time overhead of repeated calculations.

[0071] Step 6: Amplitude display and real-time refresh process for each lap. like Figure 7 As shown, Figure 7 This is a schematic diagram of the host computer software interface and real-time imaging refresh per circle in this embodiment. The cropped amplitude image is normalized to a maximum value of 1, and values ​​below 1 are normalized to a maximum value of 1. The value is truncated and logarithmically compressed: The output is rendered to the real-time imaging panel with a dynamic range of [-25,0]dB and jet pseudo-color mapping.

[0072] During the helical scan, steps 5 and 6 are automatically triggered after each complete rotation (i.e., reaching the lower right corner of the rotation) and data writing. To avoid interface lag caused by frequent reconstruction, a mutex lock mechanism and a minimum refresh interval (0.6 seconds) are used for throttling control. As the helical scan continues, the imaging area gradually expands from the center outwards, and the image gradually becomes clearer and more complete, allowing operators to observe the imaging results of the area of ​​interest even before the scan is finished.

[0073] After all scans are completed, the system performs a final image refresh and saves the complete 3D data cube, along with metadata such as starting point coordinates, spiral radius, and grid spacing, to a MAT format file for easy offline processing and reproduction.

[0074] like Figure 8 As shown, Figure 8 This is the real-time imaging result after acquisition. The target area has concentrated energy and continuous texture, while the background area has lower energy and uniform distribution, indicating that noise and stray echoes have been effectively suppressed. The imaging image has the usability and stability to be used for subsequent feature extraction and target recognition. Example 3

[0075] This embodiment also provides a data acquisition system consisting of a millimeter-wave frequency-modulated continuous wave radar front-end and a two-dimensional motion platform.

[0076] The real-time synthetic aperture radar imaging system based on pre-scan automatic range cell selection and adaptive center-expanding square spiral scanning provided in this embodiment includes a pre-scan control and acquisition module, a target range cell automatic selection module, a spiral scanning and data organization module, a real-time imaging processing module, and a display and refresh module. The pre-scan control and acquisition module is used to control the two-dimensional motion platform to perform azimuth-dimensional linear pre-scans at multiple representative pitch positions, acquire pre-scan echo frame data for each row, and perform multi-row element-wise averaging. This module corresponds to... Figure 1 In the S1 phase, multi-line pre-scanning and element-by-element averaging operations are performed, and the averaged data is output to the target range unit automatic selection module. Specifically, this module controls the pitch axis to move sequentially to a preset position via the motion controller interface, controls the azimuth axis to perform a full-range linear scan at a preset speed, and simultaneously receives radar echo frame data via serial port, parses the frame header and tail protocol to extract frame payload, timestamp, and hardware frame count information, and performs continuity detection on the frame sequence number to count frame drops. The multi-line scan data is then output to the target range unit automatic selection module after frame number alignment and element-by-element averaging.

[0077] The target range cell automatic selection module is used to perform static clutter suppression and spectral analysis on the averaged pre-scan data, calculate the energy and variance characteristics of each range cell, obtain a comprehensive score through normalization and weighted combination, and automatically determine the range cell with the highest score as the target range cell. This module corresponds to... Figure 1 In the S2 stage, static clutter suppression and windowed FFT are performed to obtain the RTI amplitude spectrum. A comprehensive score is calculated based on energy and variance, and the cell with the maximum scoring distance is selected after near-field shielding. This module is also linked to a human-computer interaction interface, which presents the RTI heatmap and scoring curve in a graphical manner, allowing users to manually confirm or modify them by clicking with a mouse.

[0078] The spiral scanning and data organization module generates an adaptive, outward-expanding square spiral trajectory based on the target distance unit and the user-specified starting grid point. The control platform moves segment by segment along the trajectory and acquires echo frame data for each segment in real time. Valid frames of the uniform speed segment are selected from each data segment and uniformly resampled. The 3D data cube is then filled and updated point by point. This module determines the spiral radius based on the minimum distance from the starting point to the four boundaries, generates a path according to a fixed sequence of five segments per revolution, filters uniform speed segments using frame timestamps and motion parameters, resamples them uniformly, writes them point by point into the 3D data cube, and updates the visited marker matrix to ensure no path overlap. Simultaneously, a dual-thread mechanism asynchronously saves each segment of original frame data to disk, avoiding write operations that could block the acquisition process.

[0079] The real-time imaging processing module performs Hilbert transform, distance compression, and wavenumber domain two-dimensional focusing processing based on target distance cells on the three-dimensional data cube to obtain a complex-valued focused image, and then performs amplitude normalization and logarithmic compression to generate the imaging result. This module corresponds to... Figure 1In the S4 stage, Hilbert transform, IFFT distance compression, extraction of two-dimensional slices of target range cells, and multiplication of 2D-FFT with the focusing operator and 2D-IFFT using the wavenumber domain focusing operator (calculated for the first time and reused later) are performed to obtain a complex value focusing map, which is triggered by the completion of one scan. The wavenumber domain focusing operator is calculated and cached during the first imaging and reused directly during subsequent refreshes.

[0080] The display and refresh module is used to display the imaging results and triggers the real-time imaging processing module to process and refresh the display after each complete spiral scan. This module corresponds to... Figure 1 In the S5 stage, modulus extraction, maximum value normalization, truncation, logarithmic compression, pseudo-color mapping, and display are performed. After all revolutions are completed, the complete 3D data cube and metadata are saved. A mutex lock mechanism and minimum refresh interval are used for throttling control. The scanning progress, axis position coordinates, frame statistics, and frame drop rate are displayed synchronously through a graphical interface.

[0081] This system also includes a human-machine interface module, whose functions include: displaying pre-scan RTI heatmaps, scoring curves, energy curves, and automatically recommended target distance unit identifiers; users can manually select and confirm these by clicking on the heatmap or curves; providing input functions for the coordinates of the starting grid point for helical scanning (azimuth dimension 1-400, elevation dimension 1-401) and scanning mode switching functions (serpentine grid or helical scan), allowing users to drive the platform to the starting point to verify the physical location before performing the formal scan; and providing basic operation control functions such as serial port connection management, motion controller connection management, homing operation, acquisition start and stop, and log display. This module is integrated with... Figure 1 The interaction between the S2 and S3 stages provides users with a visual interactive interface for inputting starting grid points and selecting distance cells.

[0082] The above-described embodiments are merely preferred embodiments provided to fully illustrate the present invention, and the scope of protection of the present invention is not limited thereto. Equivalent substitutions or modifications made by those skilled in the art based on the present invention are all within the scope of protection of the present invention. The scope of protection of the present invention is defined by the claims.

Claims

1. A pre-scanning range selection and adaptive helical scanning real-time SAR imaging method, comprising the steps of controlling a two-dimensional motion platform to scan the scene under test and acquiring echo frame data, characterized in that, It also includes the following steps: S1: Control the two-dimensional motion platform to perform azimuth dimension linear pre-scan at multiple preset pitch positions in the scene to be tested, collect the pre-scan echo frame data of each row, and perform element-wise averaging on the multiple rows of pre-scan echo frame data to obtain the average pre-scan echo frame data. S2: Perform static clutter suppression and fast time-dimensional spectrum analysis on the average pre-scan echo frame data, calculate a comprehensive score based on the energy characteristics and variance characteristics of each range cell, and automatically determine the target range cell based on the comprehensive score; S3: Based on the target distance unit and the user-specified starting grid point, control the two-dimensional motion platform to perform formal scanning according to the adaptive center-outward square spiral trajectory. In each motion segment, real-time echo frame data is collected. Based on the motion parameters of each segment and the timestamp of the echo frame, select the effective frame of the uniform speed segment and uniformly resample it to the same number of frames as the grid step number of that segment. Fill the three-dimensional data cube point by point with the resampled echo frame data. S4: Perform Hilbert transform and distance compression on the three-dimensional data cube, extract two-dimensional data slices based on the target distance unit, and use wavenumber domain focusing operators to perform two-dimensional focusing to obtain a complex value focusing map; S5: Perform amplitude normalization and logarithmic compression on the complex-valued focusing image, and output the imaging result; During the execution of steps S3 to S5, after each complete spiral scan and update of the three-dimensional data cube, steps S4 and S5 are triggered to refresh the displayed imaging results.

2. The pre-scan range selection and adaptive spiral scanning real-time SAR imaging method as described in claim 1, characterized in that, In step S1, the plurality of preset pitch positions are at least two different pitch positions selected within the pitch range of the scene to be tested. In step S1, multiple preset pitch positions are selected as representative rows; for each representative row, the azimuth axis is controlled to perform a full-range linear scan at a preset scanning speed and acquire echo frame data; after aligning the data matrix of each representative row by frame number, the arithmetic mean is taken element by element along the row dimension.

3. The pre-scan range selection and adaptive spiral scanning real-time SAR imaging method as described in claim 1, characterized in that, In step S2, calculating the comprehensive score further includes: calculating the energy characteristics and variance characteristics of each distance cell, normalizing them, and then linearly combining them according to preset weights to obtain the comprehensive score. ； in, This represents the overall score of the k-th distance unit; This represents the normalized energy of the k-th distance cell; This represents the normalized variance of the k-th distance cell; Indicates the distance unit number; and The weighting coefficients and ; After setting the comprehensive score of a preset number of near-range units corresponding to the radar blind zone to zero, the range unit with the highest score is selected as the target range unit.

4. The pre-scan range selection and adaptive spiral scanning real-time SAR imaging method as described in claim 1, characterized in that, In step S3, the method for generating the adaptive center-expanding square spiral trajectory is as follows: Starting from the user-specified grid point Taking the center of the spiral as the minimum distance from the starting point to the four boundaries as the maximum spiral radius. ; Starting from the first rotation, expand outwards with each rotation until the maximum spiral radius is reached. The circle is executed in five segments: "positive azimuth, positive pitch, negative azimuth, negative pitch, and positive azimuth," with the number of steps for each segment being as follows. , , , , This ensures that each grid point is visited exactly once.

5. The pre-scan range selection and adaptive spiral scanning real-time SAR imaging method as described in claim 1, characterized in that, In step S3, selecting effective frames of the uniform speed segment for each motion segment and uniformly resampling further includes: The existence of a uniform speed interval is determined based on the acceleration, deceleration, and scanning speed parameters of the motion segment. If it exists, the frames within the uniform speed segment are selected as candidate sets using the timestamps of each frame. The number of frames with the same grid step number is extracted according to the equally spaced index. If the number of candidate frames is insufficient, the first and last frames are symmetrically padded to make up the difference.

6. The pre-scan range selection and adaptive spiral scanning real-time SAR imaging method as described in claim 1, characterized in that, In step S4, the wavenumber domain focusing operator is pre-calculated and cached, and reused each time imaging is triggered; the focusing operator is calculated according to the following formula: ； in, Indicates the focusing operator; This indicates the distance from the target to the center of the cell; Indicates the longitudinal wavenumber; The According to The following formula is used for calculation: ； in, For free space wavenumber, These are the space wavenumbers in the azimuth and elevation directions.

7. The pre-scan range selection and adaptive spiral scanning real-time SAR imaging method as described in claim 1, characterized in that, In step S5, the process of refreshing and displaying the imaging results is controlled by a mutex lock mechanism and a minimum refresh interval.

8. The pre-scan range selection and adaptive spiral scanning real-time SAR imaging method as described in claim 1, characterized in that, It also includes the steps of controlling the two-dimensional motion platform to move to the user-specified starting grid point position for physical position verification before the formal scan; and the steps of asynchronously saving the original echo frame data of each motion segment to the disk during the scan.

9. A pre-scanning range selection and adaptive helical scanning real-time SAR imaging system, comprising a two-dimensional motion platform, a radar front-end, a motion controller, and a host computer, characterized in that, The host computer includes: The pre-scan control and acquisition module is used to control the two-dimensional motion platform to perform azimuth linear pre-scan at multiple preset pitch positions in the scene to be tested, acquire the pre-scan echo frame data of each row, and perform element-wise averaging of the multiple rows of pre-scan echo frame data to obtain the average pre-scan echo frame data. The target distance cell automatic selection module is used to perform static clutter suppression and fast time-dimensional spectrum analysis on the average pre-scan echo frame data, calculate a comprehensive score based on the energy characteristics and variance characteristics of each distance cell, and automatically determine the target distance cell according to the comprehensive score. The spiral scanning and data organization module is used to generate an adaptive center-expanding square spiral trajectory based on the target distance unit and the user-specified starting grid point, control the two-dimensional motion platform to perform formal scanning according to the trajectory, collect echo frame data in real time in each motion segment, select effective frames of the uniform speed segment based on the motion parameters of each segment and the timestamp of the echo frame, and uniformly resample to the number of frames consistent with the number of grid steps of that segment, and fill the three-dimensional data cube point by point with the resampled echo frame data. The real-time imaging processing module is used to perform Hilbert transform and distance compression on the three-dimensional data cube, extract two-dimensional data slices based on the target distance unit, and perform two-dimensional focusing using wavenumber domain focusing operators to obtain complex value focusing images. The display and refresh module is used to normalize the amplitude and logarithmically compress the complex-valued focusing image, output the imaging result, and trigger the real-time imaging processing module to process and refresh the display after each complete spiral scan and update of the three-dimensional data cube.

10. The pre-scanning range selection and adaptive spiral scanning real-time SAR imaging system as described in claim 9, characterized in that, The target distance unit automatic selection module is also used to display the distance-time amplitude spectrum heatmap, comprehensive scoring curve and energy curve of the pre-scan echo data on the human-computer interaction interface, and mark the automatically recommended target distance units. At the same time, it supports users to manually select and confirm the target distance units by clicking. The spiral scanning and data organization module generates an adaptive center-expanding square spiral trajectory by using a user-specified starting grid point. Taking the center of the spiral as the minimum distance from the starting point to the four boundaries as the maximum spiral radius. Starting from the first turn, expand outwards with each turn until the maximum spiral radius is reached. The circle is executed in five segments: "positive azimuth, positive pitch, negative azimuth, negative pitch, and positive azimuth," with the number of steps for each segment being as follows. , , , , This ensures that each grid point is visited exactly once; The spiral scanning and data organization module selects valid frames of the uniform speed segment and resamples them uniformly for each motion segment as follows: it determines whether there is a uniform speed interval based on the acceleration, deceleration and scanning speed parameters of the motion segment. If there is, it uses the timestamp of each frame to filter the frames in the uniform speed segment as a candidate set; it extracts the number of frames consistent with the grid step number according to the equally spaced index. If the number of candidate frames is insufficient, it uses the method of symmetrical padding of the first and last frames to make up the number.

Images