A multi-band remote sensing image compensation processing method, device and equipment

Abstract

The application relates to the field of remote sensing image processing, and discloses a compensation processing method, device and equipment for multi-band remote sensing images, wherein the method comprises the following steps: acquiring initial remote sensing images of multiple bands; performing fixed difference processing on the initial remote sensing images to obtain first remote sensing images of multiple different bands; determining a first remote sensing image pair formed by any two first remote sensing images; performing sub-pixel level registration on the first remote sensing image pair; determining a relative displacement curve of the first remote sensing image pair based on the registration result; performing frequency spectrum analysis on the relative displacement curve; determining a main frequency component of an optical axis jitter based on the analysis result; determining an optical axis jitter model based on the main frequency component; and performing geometric compensation on the first remote sensing images according to the optical axis jitter model to obtain multiple second remote sensing images after compensation. The technical scheme provided by the application can realize accurate compensation of optical axis jitter of remote sensing images, thereby improving the imaging quality of the remote sensing images.

Description

Technical Field

[0001] This application relates to the field of remote sensing image processing, and in particular to a method, apparatus and equipment for compensating and processing multi-band remote sensing images. Background Technology

[0002] With the rapid development of remote sensing technology, high-resolution Earth observation places increasingly stringent requirements on the attitude stability of remote sensing satellite platforms and the imaging accuracy of payloads. Ideally, the imaging line of sight of remote sensing payloads should remain highly stable during on-orbit operation.

[0003] In practical applications, remote sensing payloads inevitably experience multi-frequency, slight jitter due to a combination of factors such as servo errors, platform micro-vibrations, and complex space thermal environments. However, even slight jitter is enough to introduce significant geometric distortions into ground imaging results, severely reducing the positioning accuracy and quality of the images.

[0004] Therefore, improving the image quality of remote sensing images has become a key research focus in the field of remote sensing image processing. Summary of the Invention

[0005] This application provides a method, apparatus, and device for compensating multi-band remote sensing images, which can achieve precise compensation for line-of-sight jitter in remote sensing images, thereby improving the imaging quality of remote sensing images.

[0006] This application provides a compensation processing method for multi-band remote sensing images. The method includes: acquiring initial remote sensing images of multiple bands; performing fixed difference processing on the initial remote sensing images to obtain multiple first remote sensing images of different bands; determining a first remote sensing image pair composed of any two first remote sensing images; performing sub-pixel level registration on the first remote sensing image pair; determining the relative displacement curve of the first remote sensing image pair based on the registration result, wherein the relative displacement curve characterizes the relative line-of-sight jitter between the first remote sensing image pair; performing spectral analysis on the relative displacement curve; determining the main frequency components of the line-of-sight jitter based on the analysis result; determining the line-of-sight jitter model based on the main frequency components; and performing geometric compensation on the first remote sensing images according to the line-of-sight jitter model to obtain multiple compensated second remote sensing images.

[0007] In one embodiment, the fixed difference processing includes relative radiometric calibration and co-registration processing; the fixed difference processing of the initial remote sensing image includes: performing relative radiometric calibration on the initial remote sensing image to eliminate radiometric response differences between linear array detectors of different bands; and performing co-registration processing on the radiometrically calibrated initial remote sensing image to eliminate fixed position offsets caused by the physical spacing of the arrangement of linear array detectors of different bands.

[0008] In one embodiment, the first remote sensing image includes multiple consecutive image blocks divided according to scanning time; subpixel-level registration is performed on the first remote sensing image pair. The first and second image blocks of the first remote sensing image pair at any scanning time are acquired, and the normalized cross-power spectra of the first and second image blocks are obtained. The normalized cross-power spectra are filtered based on an anisotropic weighted circular mean algorithm. The filtered normalized cross-power spectra are approximated using a rank-one subspace to obtain relevant phase information. The relevant phase information is then subjected to one-dimensional expansion and fitting processing to obtain the subpixel-level offset of the first remote sensing image pair at the scanning time.

[0009] In one embodiment, filtering the normalized cross-power spectrum based on the anisotropic weighted circular mean algorithm includes: within a set filtering window, performing weighted circular mean filtering on the normalized cross-power spectrum according to filtering weight coefficients, wherein the filtering weight coefficients are determined by amplitude weights and spatial weights, the amplitude weights are determined based on the amplitude of the normalized cross-power spectrum, and the spatial weights are determined based on the anisotropic weighted kernel.

[0010] In one embodiment, the anisotropic weighted kernel characterizes the filtering intensity in different directions, and the anisotropic weighted kernel is determined based on a first directional standard deviation and a second directional standard deviation, wherein the first directional standard deviation is greater than the second directional standard deviation.

[0011] In one implementation, determining the relative displacement curve of the first remote sensing image pair based on the registration result includes: obtaining a sub-pixel-level offset sequence of the first remote sensing image pair, the sub-pixel-level offset sequence including the sub-pixel-level offset of the first remote sensing image at each scan time; constructing a relative displacement curve of the first remote sensing image pair based on the sub-pixel-level offset sequence, wherein the horizontal axis of the relative displacement curve represents each scan time of the first remote sensing image pair, and the vertical axis of the relative displacement curve represents the sub-pixel-level offset corresponding to the scan time.

[0012] In one embodiment, the main frequency components include a target fundamental frequency component and harmonic components; performing spectral analysis on the relative displacement curve and determining the main frequency components of the line-of-sight jitter based on the analysis results includes: preprocessing the relative displacement curve, performing a Fourier transform on the preprocessed relative displacement curve to obtain the amplitude spectrum of the relative displacement curve; determining one or more target extreme points in the amplitude spectrum, and traversing within the frequency resolution range of each target extreme point to determine one or more target fundamental frequency components and their harmonic components of the line-of-sight jitter, and using the target fundamental frequency components and their harmonic components as the main frequency components.

[0013] In one embodiment, the line-of-sight jitter model is used to describe the change of line-of-sight jitter over time; compensating the first remote sensing image for line-of-sight jitter based on the line-of-sight jitter model to obtain multiple compensated second remote sensing images includes: determining the amount of line-of-sight jitter at each scanning moment based on the line-of-sight jitter model, wherein the amount of line-of-sight jitter represents the positional offset of the line of sight of the remote sensing payload at the scanning moment; and performing geometric correction on the first remote sensing image through image resampling based on each of the line-of-sight jitter amounts, and using the geometrically corrected first remote sensing image as the second remote sensing image.

[0014] A second aspect of this application provides a compensation processing apparatus for multi-band remote sensing images. The system includes: an image acquisition unit, configured to acquire initial remote sensing images of multiple bands, and perform fixed difference processing on the initial remote sensing images to obtain multiple first remote sensing images, wherein the first remote sensing images represent first remote sensing images of different bands; a registration processing unit, configured to perform sub-pixel-level registration on any pair of first remote sensing images, and determine the relative displacement curve of the first remote sensing image pair based on the registration result, wherein the relative displacement curve represents the relative line-of-sight jitter between the first remote sensing image pairs; and a jitter compensation unit, configured to perform spectral analysis on the relative displacement curve, determine the main frequency components of the line-of-sight jitter based on the analysis result, determine the line-of-sight jitter model based on the main frequency components, and perform line-of-sight jitter compensation on the first remote sensing images according to the line-of-sight jitter model to obtain multiple compensated second remote sensing images.

[0015] A third aspect of this application provides a computer device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the multi-band remote sensing image compensation processing method described in the first aspect.

[0016] The technical solution provided in one or more embodiments of this application extracts line-of-sight jitter features through sub-pixel registration of multi-band images, achieving high-precision geometric compensation of remote sensing images. Specifically, firstly, the initial multi-band images undergo fixed difference processing to eliminate differences in radiometric response and fixed position offsets between detectors. Then, a sub-pixel registration method based on anisotropic weighted circular mean filtering is employed to effectively capture sub-pixel-level offsets, thereby constructing a relative displacement curve that accurately reflects line-of-sight jitter. A combination of spectral analysis and fundamental frequency determination is used, and the line-of-sight jitter model determined based on the relative displacement curve accurately calculates complex jitter containing multi-frequency components, achieving high-precision geometric compensation. It is evident that the technical solution provided in this application, through precise processing at each stage, effectively solves the problem of insufficient accuracy in complex jitter compensation using traditional methods, achieving accurate compensation of line-of-sight jitter in remote sensing images, significantly reducing image blur and distortion, and thus greatly improving the imaging quality of remote sensing images. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the specific embodiments of this application or the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 A schematic diagram illustrating the steps of a multi-band remote sensing image compensation processing method provided in an embodiment of this application; Figure 2 This is a schematic diagram illustrating the steps of subpixel-level registration in one embodiment of this application; Figure 3 A schematic diagram of the structure of a compensation processing device for multi-band remote sensing images provided in one embodiment of this application; Figure 4 This is a schematic diagram of the structure of a computer device provided in one embodiment of this application. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0020] Furthermore, the use of terms such as "first," "second," etc., in this application is for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of embodiments in this application, unless otherwise stated, "multiple" means two or more. Additionally, the use of "based on" or "according to" implies openness and inclusiveness, because processes, steps, calculations, or other actions "based on" or "according to" one or more of the stated conditions or values ​​may in practice be based on additional conditions or beyond the stated values.

[0021] During operation, the line-of-sight (LOS) of a remote sensing satellite payload should remain stable. However, due to factors such as servo errors, platform micro-vibrations, and the complex thermal environment of space, the LOS of remote sensing payloads is prone to jitter. This LOS jitter can be understood as motion caused by the superposition of harmonic components of multiple frequencies. For high-orbit, high-resolution imaging payloads, even small deviations caused by LOS jitter can produce significant geometric errors on the ground, severely affecting image quality and application effectiveness. Furthermore, to achieve more refined Earth observation, remote sensing payloads are developing towards higher spatial resolution, higher temporal resolution, and more flexible observation modes, placing increasingly stringent requirements on the accuracy and stability of the payload's LOS.

[0022] However, jitter detection methods based on parallax observation in related technologies still face challenges. First, line-of-sight jitter is often the result of multiple vibration sources acting together, manifesting in the frequency domain as a complex multi-frequency signal composed of the fundamental frequency and its harmonics. Traditional spectral analysis methods, due to limited frequency resolution and the influence of spectral leakage, struggle to accurately and reliably separate all major frequency components from noise, especially when multiple frequency components are close together, easily leading to estimation bias or omissions. Second, image shift caused by jitter is usually at the sub-pixel level, placing extremely high demands on the accuracy of image registration algorithms. Existing high-precision registration algorithms, when processing remote sensing images, are susceptible to interference from spectral noise and aliasing effects in their core frequency domain calculation process, resulting in registration errors. These errors are then propagated to subsequent jitter resolution stages, ultimately affecting the compensation effect.

[0023] In view of this, how to effectively improve the accuracy and reliability of detecting complex multi-frequency line-of-sight jitter in multi-band remote sensing images and achieve high-quality geometric compensation has become a technical bottleneck that urgently needs to be overcome in the field of high-resolution remote sensing satellite image processing. This application provides one or more embodiments of a compensation processing method, apparatus, and device for multi-band remote sensing images, which can solve the above-mentioned problems, achieve accurate compensation for line-of-sight jitter in remote sensing images, and thus improve the imaging quality of remote sensing images.

[0024] Please see Figure 1One embodiment of this application provides a compensation processing method for multi-band remote sensing images, which may include the following steps: S1: Acquire initial remote sensing images of multiple bands, perform fixed difference processing on the initial remote sensing images to obtain first remote sensing images of multiple different bands; S3: Determine a first remote sensing image pair consisting of any two first remote sensing images, perform subpixel-level registration on the first remote sensing image pair, and determine the relative displacement curve of the first remote sensing image pair based on the registration result, wherein the relative displacement curve characterizes the relative jitter of the line of sight between the first remote sensing image pair. S5: Perform spectral analysis on the relative displacement curve, determine the main frequency components of line-of-sight jitter based on the analysis results, determine the line-of-sight jitter model based on the main frequency components, and perform geometric compensation on the first remote sensing image according to the line-of-sight jitter model to obtain multiple compensated second remote sensing images.

[0025] The aforementioned initial remote sensing images are unprocessed remote sensing images acquired by linear array detectors in different bands. These bands can be understood as different spectral ranges, and the image of each band reflects the reflectance or radiation characteristics of ground objects within that specific spectral range. Specifically, for multi-band scanning remote sensing payloads, the linear array detectors in each band are arranged parallel to each other on the focal plane, scanning the same ground point at fixed time intervals to form parallax observations along the scanning direction, thereby obtaining the initial remote sensing images of the aforementioned multiple bands.

[0026] The errors between the initial remote sensing images and the actual images, besides the time-varying geometric shift caused by line-of-sight jitter, also exhibit significant fixed differences, including differences in radiometric response caused by different linear array detectors and fixed position shifts. The radiometric response differences can be understood as linear grayscale differences in different initial remote sensing images caused by variations in the physical characteristics of each linear array detector (such as filters and sensor sensitivity). The fixed position shifts can be understood as constant shifts caused by the fixed physical spacing between the linear array detectors.

[0027] In this embodiment, to eliminate the aforementioned fixed differences and reduce noise during subsequent line-of-sight jitter compensation, the initial remote sensing image is first subjected to fixed difference processing to eliminate radiometric response differences and fixed position offsets between detectors of different bands. The initial remote sensing image after fixed difference processing is used as the first remote sensing image. This can be understood as representing the processed initial remote sensing images of different bands. After this processing, the first remote sensing image has eliminated fixed differences in radiometric response and geometric position, providing a foundation for further line-of-sight jitter detection and compensation. This allows subsequent line-of-sight jitter detection to more accurately extract the time-varying geometric offset caused by line-of-sight jitter, rather than being interfered with by fixed differences, ensuring accurate compensation for line-of-sight jitter in the remote sensing image.

[0028] The aforementioned first remote sensing image pair is formed by arbitrarily selecting two images from multiple remote sensing images of different bands after fixed difference processing. It should be noted that for images of multiple bands, the line-of-sight jitter is the same at a specific time; selecting any one of the first remote sensing image pairs allows determination of the line-of-sight jitter at each scanning time. This line-of-sight jitter can be understood as the positional offset of the line of sight at different times. This jitter causes the position of ground features in the image to shift, affecting the geometric accuracy of the image. During image offset registration, because the magnitude of line-of-sight jitter is small, the resulting image offset is generally at the sub-pixel level. Sub-pixel level registration is required to improve the image registration accuracy to the sub-pixel level. Accurate image alignment is achieved by calculating the minute offset between the two images. Specifically, the sub-pixel level registration uses an anisotropic weighted circular mean filtering algorithm during the filtering process to improve the accuracy of the phase correlation registration algorithm. Based on the sub-pixel level registration results, the sub-pixel level offset at each scanning time is determined, and the sub-pixel level offsets at all scanning times are collected to form a relative displacement curve.

[0029] The aforementioned relative displacement curves represent the changes in the relative displacement of the first remote sensing image pair over time at different scanning moments. These curves reflect the relative jitter of the line of sight between the first remote sensing image pair, and this relative jitter is a time-varying parameter characterizing the line-of-sight jitter between the first remote sensing image pair. By calculating the relative displacement curves, the changes in line-of-sight jitter over time can be visually displayed and obtained, enabling researchers and users to better understand and analyze the dynamic characteristics of line-of-sight jitter. Through accurate line-of-sight jitter detection, geometric compensation can be performed more effectively, thereby improving the imaging quality of remote sensing images. Traditional methods for constructing relative displacement curves are typically based on simple linear or quadratic curve fitting. These methods have limited accuracy when dealing with complex nonlinear jitter. Constructing relative displacement curves using high-precision offset sequences obtained through sub-pixel level registration can more accurately reflect the dynamic characteristics of line-of-sight jitter.

[0030] In this embodiment, spectral analysis of the relative displacement curve can extract the main frequency components of visual axis jitter. Specifically, the signal in the time domain is converted into a signal in the frequency domain using methods such as Fourier transform, thereby extracting information such as the frequency components, amplitude, and phase of the signal. For example, the relative displacement curve is preprocessed, such as smoothing and denoising, to improve the accuracy of the spectral analysis. A Fourier transform is then performed on the preprocessed relative displacement curve to convert it from the time domain to the frequency domain, obtaining an amplitude spectrum. One or more target extreme points (maximum points) are identified in the amplitude spectrum, and the frequency resolution range of each target extreme point is traversed to determine one or more target fundamental frequency components and their harmonic components of visual axis jitter. These frequency components are used as the main frequency components, providing a reliable data foundation for establishing an accurate visual axis jitter model, making the detection of visual axis jitter more precise.

[0031] In spectral analysis, the dominant frequency components of a signal, including the fundamental frequency component and harmonic components, typically have a significant impact on the signal's characteristics. Further, a line-of-sight jitter model for line-of-sight jitter compensation can be determined based on these dominant frequency components. This model can employ a mathematical model to describe the time-varying nature of line-of-sight jitter, including the amplitude and phase information of each frequency component. Optionally, the model can also be constructed based on deep learning models such as convolutional neural networks (CNNs) or recurrent neural networks (RNNs), automatically identifying and compensating for line-of-sight jitter by learning its characteristics. Based on the amplitude and phase information of the dominant frequency components, this model can predict the positional shift of the line of sight at different times and perform geometric compensation on the first remote sensing image based on this shift. The second remote sensing image, after geometric compensation, shows a significant improvement in geometric accuracy, reducing blur and distortion caused by line-of-sight jitter, thereby achieving accurate compensation for line-of-sight jitter in remote sensing images and improving the imaging quality of remote sensing images.

[0032] For example, the above-mentioned line-of-sight jitter model can be expressed as: ,in, This represents the line-of-sight jitter model, where, , indicating at time The model's predicted value, which depends on the parameters. , The parameters to be solved include the magnitude. and phase etc., used to describe the characteristics of visual axis jitter. To indicate different imaging times, according to It can accurately calculate the amount of line-of-sight jitter along the scanning direction at each imaging moment, and use bilinear interpolation to resample the remote sensing image to achieve accurate compensation for line-of-sight jitter.

[0033] Based on the above ideas, the technical solution provided in this embodiment of the application extracts line-of-sight jitter features through sub-pixel registration of multi-band images to achieve high-precision geometric compensation of remote sensing images. Specifically, firstly, the initial multi-band images undergo fixed difference processing to eliminate differences in radiometric response and fixed position offsets between detectors. Then, a sub-pixel registration method based on an anisotropic weighted circular mean filtering algorithm is adopted, which can effectively capture sub-pixel-level offsets, thereby constructing a relative displacement curve that accurately reflects line-of-sight jitter. By accurately solving the complex jitter containing multi-frequency components through the line-of-sight jitter model determined based on the relative displacement curve, accurate compensation of line-of-sight jitter in remote sensing images can be achieved, thereby improving the imaging quality of remote sensing images.

[0034] In one embodiment, the above-mentioned fixed difference processing includes relative radiometric calibration and co-registration processing; the fixed difference processing of the initial remote sensing image includes: performing relative radiometric calibration on the initial remote sensing image to eliminate the differences in radiometric response between linear array detectors of different bands; and performing co-registration processing on the radiometrically calibrated initial remote sensing image to eliminate the fixed position offset caused by the physical spacing of the arrangement of linear array detectors of different bands.

[0035] In this embodiment, the radiometric characteristics of each initial remote sensing image are adjusted through relative radiometric calibration. Specifically, the initial remote sensing images are processed to adjust the grayscale values ​​of each band image. A linear transformation is used to adjust the grayscale values ​​of each initial remote sensing image so that, under the same illumination conditions, the radiometrically calibrated initial remote sensing images have consistent grayscale response values ​​to the same uniform object, thereby eliminating the differences in radiometric response between linear array detectors of different bands, i.e., linear grayscale differences. For example, the above linear transformation can be determined based on gain, input, and offset.

[0036] In this embodiment, after relative radiometric calibration of the initial remote sensing image, co-registration processing is performed. Specifically, based on the known detector geometry, the radiometrically calibrated initial remote sensing image is translated to geometrically align images of different bands, thereby eliminating fixed positional offsets caused by the physical spacing of the linear array detectors of different bands. Under ideal jitter-free conditions, the same target object appears at exactly the same pixel coordinates in each first remote sensing image.

[0037] The technical solution provided in this embodiment eliminates the fixed differences between each initial remote sensing image through relative radiometric calibration and co-registration processing. This enables subsequent line-of-sight jitter detection to more accurately extract the time-varying geometric offset caused by line-of-sight jitter based on the first remote sensing image, thereby achieving precise compensation for line-of-sight jitter in remote sensing images and improving the imaging quality of remote sensing images.

[0038] In one embodiment, the aforementioned first remote sensing image includes multiple consecutive image blocks divided according to scanning time. Subpixel-level registration of the first remote sensing image pair is performed based on an anisotropic weighted circular mean algorithm. The anisotropic weighted circular mean algorithm is used to perform anisotropic weighted circular mean filtering on the normalized cross-power spectrum of the first remote sensing image pair. Specifically, the first image block and the second image block of the first remote sensing image pair at any scanning time are obtained, and the normalized cross-power spectrum of the first image block and the second image block is obtained. The normalized cross-power spectrum is filtered based on the anisotropic weighted circular mean algorithm. A rank-one subspace approximation is performed on the filtered normalized cross-power spectrum to obtain relevant phase information. The relevant phase information is then subjected to one-dimensional expansion and fitting processing to obtain the subpixel-level offset of the first remote sensing image pair at the scanning time.

[0039] The aforementioned image patch can be understood as image data at each scanning moment (corresponding to a unique scanning direction). For multi-band remote sensing images, each band has its corresponding multiple image patches. For a linear array detector within the same band, during the scanning imaging process, the scanning time of each image patch acquired is different, they are arranged sequentially, and they correspond to different, continuous strip regions on the ground. By stitching together image patches from different scanning moments of the same linear array detector in chronological order, a remote sensing image for the corresponding band can be obtained. The aforementioned first image patch and second image patch can be understood as image patches at that scanning moment in different first remote sensing images, based on the same scanning moment.

[0040] The normalized cross-power spectrum (CPS) described above is used to evaluate the similarity between two image patches and can be calculated using Fourier transform. Since spectral noise and aliasing effects affect the stability and accuracy of the CPS calculation, an anisotropic weighted circular mean algorithm is used to filter the CPS to achieve robust filtering and improve the registration accuracy of the first remote sensing image. The anisotropic weighted circular mean algorithm is used to smooth the signal in the frequency domain, and is particularly suitable for processing directional signals. Through weighted circular mean filtering, noise and aliasing effects can be suppressed while preserving the main characteristics of the signal. The rank-one subspace approximation described above can extract the main phase components, i.e., relevant phase information, from complex spectral data. The one-dimensional expansion described above can map phase information from a high-dimensional space to a one-dimensional space, and the phase change can be estimated through fitting processing (such as polynomial fitting) to obtain sub-pixel-level offset.

[0041] In this embodiment, within a defined filtering window, anisotropic weighted circular mean filtering is performed on the normalized cross-power spectrum based on filtering weight coefficients. Specifically, for each frequency point, its weighted average value within the filtering window is calculated based on the filtering weight coefficients, serving as the filtering weight coefficient for that frequency point, to perform local filtering on the normalized cross-power spectrum. Weighted circular mean filtering effectively suppresses spectral noise and aliasing, improving the accuracy of the registration algorithm. Furthermore, by applying different filtering intensities in different directions using anisotropic weighting kernels, the filtering effect is further enhanced. Moreover, the introduction of anisotropic smoothing mechanisms based on geometric characteristics in the frequency domain allows for more accurate extraction of phase information, thereby improving the accuracy of sub-pixel level registration and achieving precise compensation for line-of-sight jitter, significantly improving the imaging quality of remote sensing images.

[0042] The aforementioned filtering window can be understood as a region defined in the frequency domain, used for local filtering of the normalized cross-power spectrum. The filtering weight coefficients are determined by amplitude weights and spatial weights, used to adjust the weight of each frequency point or spatial point. The amplitude weights are determined based on the amplitude of the normalized cross-power spectrum, reflecting the intensity of different frequency components in the signal. Frequency points with larger amplitudes should be assigned higher weights, while frequency points with smaller amplitudes should be suppressed. Specifically, frequency points with larger amplitudes represent stronger consistency of the first remote sensing image pair at that frequency and should be assigned higher weights, while frequency points with smaller amplitudes should be suppressed.

[0043] The aforementioned spatial weights, determined based on anisotropic weighted kernels, reflect the filtering intensity in different directions and are used to apply different filtering intensities in different directions. In spectral analysis, the peak of the cross-power spectrum generally appears in a specific radial direction, pointing from the origin of the spectrum to the location of the peak, exhibiting strong consistency and energy concentration, while the energy of noise and aliasing is non-directional. Furthermore, the energy of noise and aliasing effects diffuses more significantly in the tangential direction, typically existing in an unstructured form. Therefore, based on the anisotropic weighted circular mean algorithm, anisotropic weighted kernels with different bandwidth characteristics are constructed in the frequency domain, and optimization is performed separately for the radial and tangential directions.

[0044] The anisotropic weighted kernel described above is a kernel function used to define spatial weights, representing the filtering intensity in different directions. This anisotropic weighted kernel is determined based on the standard deviations of a first and second direction. The first standard deviation is greater than the second standard deviation, typically applying stronger filtering in one direction and weaker filtering in others. Specifically, the first standard deviation represents the radial standard deviation, and the second standard deviation represents the tangential standard deviation.

[0045] In one embodiment, the normalized cross-power spectrum can be viewed as a complex field on a unit circle. To maintain phase consistency and avoid phase jumps, an anisotropic weighted circular mean filtering algorithm is employed within the filtering window in the following manner: ,in, This represents the weighted circular mean. Indicates the filter window. Represents the filter weight coefficients. This represents the normalized cross-power spectrum. The calculated circular mean is normalized and expressed as... , This represents a very small number that is not divisible by zero.

[0046] In this embodiment, the filter weight coefficients include amplitude weights and spatial weights, expressed as follows: ,in, Indicates magnitude weight, , and The Fourier transform corresponding to the first remote sensing image pair Indicates spatial weight.

[0047] In this embodiment, weighted kernels with different bandwidths along the radial and tangential directions are constructed in the frequency domain to construct the aforementioned spatial weights, specifically as follows: ,in, This represents the coordinate difference of a pixel within the filter window relative to the center. The first standard deviation represents the radial direction, and a relatively large smoothing scale is used. The second standard deviation represents the standard deviation in the tangential direction, and a narrower bandwidth is used to limit cross-directional diffusion.

[0048] The technical solution provided in this embodiment achieves sub-pixel-level high-precision image registration through a series of advanced signal processing methods, including filtering of the normalized cross-power spectrum based on an anisotropic weighted circular mean filtering algorithm. Specifically, by using anisotropic weighted circular mean filtering and rank-one subspace approximation, spectral noise and aliasing effects can be effectively suppressed, improving the accuracy of the phase correlation registration algorithm, thereby achieving sub-pixel-level high-precision image registration. Through precise sub-pixel-level registration, the alignment between images is more accurate, facilitating precise compensation for line-of-sight jitter in remote sensing images, reducing image blurring and geometric distortion caused by line-of-sight jitter, and thus improving the imaging quality of remote sensing images.

[0049] In one embodiment, determining the relative displacement curve of the first remote sensing image pair based on the registration result includes: obtaining a sub-pixel-level offset sequence of the first remote sensing image pair, the sub-pixel-level offset sequence including the sub-pixel-level offset of the first remote sensing image at each scanning time; constructing a relative displacement curve of the first remote sensing image pair based on the sub-pixel-level offset sequence, wherein the horizontal axis of the relative displacement curve represents each scanning time of the first remote sensing image pair, and the vertical axis of the relative displacement curve represents the sub-pixel-level offset corresponding to the scanning time.

[0050] The technical solution provided in this embodiment, through the acquisition of sub-pixel-level offset sequences and the construction of relative displacement curves, achieves sub-pixel-level accuracy in detecting line-of-sight jitter. Specifically, each scanning moment is used as the abscissa, and the corresponding sub-pixel-level offset at each scanning moment is used as the ordinate to accurately reflect the change of line-of-sight jitter over time. Through accurate line-of-sight jitter detection, geometric compensation can be performed more effectively, thereby improving the imaging quality of remote sensing images.

[0051] In one embodiment, the aforementioned main frequency components include a target fundamental frequency component and harmonic components; performing spectral analysis on the relative displacement curve and determining the main frequency components of the line-of-sight jitter based on the analysis results includes: performing signal preprocessing on the relative displacement curve, performing a Fourier transform on the preprocessed relative displacement curve to obtain the amplitude spectrum of the relative displacement curve; determining one or more target extreme points in the amplitude spectrum, and traversing within the frequency resolution range of each target extreme point to determine one or more target fundamental frequency components and their harmonic components of the line-of-sight jitter, and using the target fundamental frequency components and their harmonic components as the main frequency components.

[0052] The aforementioned target fundamental frequency component is the main frequency component in the line-of-sight jitter signal, corresponding to the most fundamental frequency of line-of-sight jitter and reflecting the basic periodic change of line-of-sight jitter. The aforementioned harmonic components are higher frequency components related to the target fundamental frequency component, typically integer multiples of the fundamental frequency, and together with the target fundamental frequency component, constitute the complete frequency spectrum of line-of-sight jitter. Before performing spectrum analysis, a series of signal preprocessing steps are performed, such as zero-filling and time-domain windowing, to improve the accuracy and reliability of the spectrum analysis. Specifically, a Hamming window is used to smooth and weight the signal, effectively suppressing spectral sidelobes and making the dominant frequency component more prominent, thus improving the accuracy of the spectrum analysis. Through Fourier transform, the signal in the time domain is converted into a signal in the frequency domain, and the amplitude distribution of the Fourier-transformed signal in the frequency domain is obtained as the amplitude spectrum. The aforementioned frequency resolution can be understood as the interval between two adjacent frequency points, determined by the signal's sampling frequency and duration. Preferably, for each target extreme point, i.e. maximum point, in the amplitude spectrum, a recursive traversal is performed at the target extreme point according to the frequency resolution range, and the target fundamental frequency component and harmonic components are determined according to the least common divisor, thereby accurately solving the main frequency components of the signal.

[0053] In one embodiment, due to visual axis jitter Frequency components and The same, through the Spectral analysis can determine The frequency components. Specifically, firstly, a zero-padding operation is performed on the signal, refining the spectral curve by adding zero values, thereby improving the apparent frequency resolution. This frequency resolution is expressed as: ,in, Indicates frequency resolution. and These are the signal sampling frequency and duration, respectively. This represents the number of points in the FFT. The number of zeros to be padded to the signal is calculated as follows: ,in, , and Representing the original signal length and the achieved apparent target frequency resolution, respectively. The required signal length and the amount of zero padding required for the signal.

[0054] In one embodiment, the main frequency components can be preliminarily determined by detecting maxima in the amplitude spectrum. Specifically, maxima detection is performed on the amplitude spectrum after zero-filling and Hamming window processing, and the frequencies of the maxima are used as target candidate frequencies. Furthermore, a detailed search is performed within the candidate frequency range. The optimal frequency component is determined as the target fundamental frequency component by calculating the least common divisor. This component, along with its corresponding harmonic components, is then classified as a major frequency component. The candidate frequency range is determined based on the minimum frequency interval of the amplitude spectrum, and the harmonic components are determined as follows: ,in, This represents the target fundamental frequency component.

[0055] In one embodiment, the frequency of any target extremum point ,exist arrive Within, at 0.001 Iterate through the intervals to find The least common divisor of the fundamental frequency components is taken as a fundamental frequency component. For each possible fundamental frequency component, a value is taken as follows: ,in, According to the threshold Will They are divided into two categories. If... ,but Classified as harmonic frequencies ,like ,but Classified as non-harmonic frequencies .

[0056] Further calculation various frequencies and Mean error of harmonic frequencies ,in The optimal fundamental frequency for the parallax is determined based on the principle of minimizing error. The target fundamental frequency component is used as the primary frequency component. This target fundamental frequency component and its corresponding harmonic frequencies are used as the main frequency components. Preferably, the optimal fundamental frequency is... Corresponding non-harmonic frequency Assign to Repeat the above traversal process until... If the value is empty, the main frequency components can be obtained more accurately.

[0057] The technical solution provided in this embodiment performs spectral analysis on the relative displacement curve through signal preprocessing and Fourier transform to accurately detect the main frequency components of line-of-sight jitter. Signal preprocessing reduces noise and interference, making the spectral analysis more accurate, while Fourier transform converts the time-domain signal into a frequency-domain signal, facilitating frequency component extraction. By traversing each target extremum point in the amplitude spectrum according to the frequency resolution range, the frequency components of line-of-sight jitter can be detected more comprehensively, improving detection reliability and achieving accurate compensation for line-of-sight jitter in remote sensing images, thereby improving the imaging quality of remote sensing images.

[0058] In one embodiment, the aforementioned line-of-sight jitter model is used to describe the change of line-of-sight jitter over time. Line-of-sight jitter compensation is performed on the first remote sensing image based on the model to obtain multiple compensated second remote sensing images. Specifically, the line-of-sight jitter amount is determined at each scanning moment based on the model. The line-of-sight jitter amount characterizes the positional offset of the remote sensing payload's line of sight in the first remote sensing image at the current scanning moment, representing the magnitude and direction of the line-of-sight jitter. Based on each line-of-sight jitter amount, the first remote sensing image is geometrically corrected through image resampling, and the geometrically corrected first remote sensing image is used as the second remote sensing image. For example, geometric correction is performed on each pixel in the image. Its ideal position in the jitter-free state is calculated based on the line-of-sight jitter amount. Each pixel is adjusted to its ideal position using image resampling techniques (such as bilinear interpolation, bicubic interpolation, etc.). The aforementioned image resampling can adjust the position and value of pixels in the image. Through image resampling, each pixel in the image can be adjusted to its proper position, thereby eliminating the impact of line-of-sight jitter on the geometric accuracy of the image.

[0059] The technical solution provided in this embodiment determines the amount of line-of-sight jitter in the first remote sensing image at each scanning moment through a line-of-sight jitter model, and performs image resampling based on these jitter amounts to achieve geometric correction of the first remote sensing image. The second remote sensing image after geometric correction has significantly improved geometric accuracy, reducing blurring and distortion caused by line-of-sight jitter, thereby achieving accurate compensation for line-of-sight jitter in remote sensing images and improving the imaging quality of remote sensing images.

[0060] Please see Figure 2 This application also provides an embodiment for determining subpixel-level translation amount. Figure 2 For a multi-band remote sensing image processing workflow, the sub-pixel offset of the first remote sensing image pair at any scanning time is obtained by following these steps: Step 1: Window function filtering. A window function filter is applied to the first pair of remote sensing images to mitigate the impact of image edge effects on the subsequent amplitude spectrum.

[0061] Step 2: Perform DFT on the first remote sensing image pair after window function filtering. This transforms the first remote sensing image pair from the spatial domain to the frequency domain to reveal the frequency characteristics of the images, providing a basis for subsequent sub-pixel level registration.

[0062] Step 3: Anisotropic weighted circular mean filtering. In the frequency domain, anisotropic weighted circular mean filtering is applied to the normalized cross-power spectrum matrix obtained by DFT to effectively suppress noise and aliasing effects while preserving the main features of the image and improving the accuracy of phase correlation registration.

[0063] Step 4: Rank-one subspace approximation. Performing rank-one subspace approximation on the filtered frequency domain data extracts relevant phase information, which helps to extract the main phase components from complex frequency data, providing a basis for subsequent offset calculations.

[0064] Step 5: One-dimensional expansion. The extracted relevant phase information is expanded in one dimension, converting it from a two-dimensional frequency domain signal to a one-dimensional time domain signal, which facilitates subsequent fitting and analysis.

[0065] Step Six: Fitting Analysis. The one-dimensional unfolded phase information is fitted to accurately estimate the sub-pixel level offset.

[0066] In this embodiment, an anisotropic weighted circular mean filtering algorithm is proposed through steps one to six described above. This algorithm introduces anisotropic smoothing mechanisms based on geometric characteristics in the frequency domain. Traditional sub-pixel-level registration methods use filters to filter the normalized cross-power spectrum, which has certain limitations in suppressing spectral noise and aliasing effects. This embodiment, by employing weighted circular mean filtering in the local neighborhood, can more effectively suppress spectral noise and aliasing effects, improve the accuracy of the phase-correlation registration algorithm, and achieve precise compensation for line-of-sight jitter in remote sensing images, thereby improving the imaging quality of remote sensing images.

[0067] This application also provides an embodiment in which the normalized cross-power spectrum is filtered based on the anisotropic weighted circular mean algorithm according to the following steps: Step 1: Constructing the frequency domain coordinate system. Calculate the radial radius and angle information for each frequency point to facilitate the subsequent construction of a direction-dependent anisotropic weighted kernel, transforming each point in the frequency domain from rectangular coordinates to polar coordinates.

[0068] Step 2: Initial filtering in each direction. The standard deviation is used as... A two-dimensional Gaussian kernel is used to convolve the real and imaginary parts of the amplitude-weighted normalized cross-power spectrum, respectively, to remove those very obvious and severe noise points, thus achieving preliminary noise reduction of the spectrum.

[0069] Step 3: Radial one-dimensional filtering. Perform a convolution on the image using a one-dimensional Gaussian kernel in the horizontal direction to obtain... By performing a convolution on the image using a one-dimensional Gaussian kernel in the vertical direction, we obtain... .

[0070] Step 4: Directional Weighted Fusion. This is based on the angles between the frequency vector and the horizontal and vertical directions. The weights along the radial and tangential directions are determined and weighted fusion is performed to approximately achieve enhanced smoothing along the radial direction and weakened smoothing along the tangential direction, expressed as: ,in, and These represent the horizontal and vertical coordinates of the frequency vector, respectively. , These represent the weights along the radial and tangential directions, respectively. , These represent the filtering results along the radial and tangential directions, respectively.

[0071] Step 5: Unit Circle Normalization. Normalize the results to ensure phase consistency, expressed as: To ensure that the filtered cross-power spectrum remains on the unit circle to maintain its phase characteristics, steps one through five above can be iterated.

[0072] Please see Figure 3 This application also provides a compensation processing apparatus for multi-band remote sensing images, the apparatus comprising: The image acquisition unit 100 is used to acquire initial remote sensing images of multiple bands and perform fixed difference processing on the initial remote sensing images to obtain first remote sensing images of multiple different bands. The registration processing unit 200 is used to perform sub-pixel level registration on the first remote sensing image pair composed of any two first remote sensing images, and determine the relative displacement curve of the first remote sensing image pair based on the registration result, wherein the relative displacement curve characterizes the relative jitter of the line of sight between the first remote sensing image pair. The jitter compensation unit 300 is used to perform spectral analysis on the relative displacement curve, determine the main frequency components of line-of-sight jitter based on the analysis results, determine the line-of-sight jitter model based on the main frequency components, and perform line-of-sight jitter compensation on the first remote sensing image according to the line-of-sight jitter model to obtain multiple compensated second remote sensing images. in, In one embodiment, the image acquisition unit 100 is specifically used to acquire initial remote sensing images of multiple bands, perform relative radiometric calibration on the initial remote sensing images to eliminate radiometric response differences between linear array detectors of different bands, and perform co-registration processing on the radiometrically calibrated initial remote sensing images to eliminate fixed position offsets caused by the physical spacing of the arrangement of linear array detectors of different bands.

[0073] In one embodiment, the registration processing unit 200 is specifically configured to acquire the first image block and the second image block of the first remote sensing image pair at any scanning time, acquire the normalized cross power spectrum of the first image block and the second image block, filter the normalized cross power spectrum based on the anisotropic weighted circular mean algorithm, approximate the filtered normalized cross power spectrum with a rank-one subspace to obtain relevant phase information, and perform one-dimensional expansion and fitting processing on the relevant phase information to obtain the subpixel-level offset of the first remote sensing image pair at the scanning time.

[0074] In one embodiment, the jitter compensation unit 300 is specifically used to perform spectral analysis on the relative displacement curve, determine the main frequency components of the line-of-sight jitter based on the analysis results, determine the line-of-sight jitter model based on the main frequency components, determine the line-of-sight jitter amount at each scanning moment according to the line-of-sight jitter model, wherein the line-of-sight jitter amount characterizes the positional offset of the line of sight of the remote sensing payload in the first remote sensing image at the scanning moment, and geometrically correct the first remote sensing image by image resampling based on each of the line-of-sight jitter amounts, and use the geometrically corrected first remote sensing image as the second remote sensing image.

[0075] Further functional descriptions of the above modules and units are the same as those in the corresponding embodiments described above, and will not be repeated here.

[0076] The compensation processing device for multi-band remote sensing images in this application embodiment is presented in the form of a functional unit. Here, a unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that execute one or more software or fixed programs, or other devices that can provide the above functions.

[0077] Please see Figure 4 , Figure 4 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application, such as... Figure 4As shown, the computer device includes one or more processors 10, memory 20, and interfaces for connecting the components, including high-speed interfaces and low-speed interfaces. The components communicate with each other via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the computer device, including instructions stored in or on memory to display graphical information of a GUI on external input / output devices (such as display devices coupled to the interfaces). In some alternative implementations, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple computer devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system). Figure 4 Take a processor 10 as an example.

[0078] Processor 10 may be a central processing unit, a network processor, or a combination thereof. Processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The programmable logic device may be a complex programmable logic device (CAMP), a field-programmable gate array (FPGA), a general-purpose array logic (GDA), or any combination thereof.

[0079] The memory 20 stores instructions executable by at least one processor 10 to cause the at least one processor 10 to perform the method shown in the above embodiments.

[0080] The memory 20 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the computer device. Furthermore, the memory 20 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some alternative embodiments, the memory 20 may optionally include memory remotely located relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0081] The memory 20 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, hard disk or solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0082] The computer device also includes a communication interface 30 for communicating with other devices or communication networks.

[0083] This application also provides a computer-readable storage medium. The methods described in this application can be implemented in hardware or firmware, or implemented as recordable on a storage medium, or implemented as computer code downloaded over a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and subsequently stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the methods shown in the above embodiments are implemented.

[0084] The apparatus, module, or unit described in the above embodiments can be implemented by a computer chip or entity, or by a product having a certain function. A typical implementation device is a computer. Specifically, the computer can be, for example, a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email device, game console, tablet computer, wearable device, or any combination of these devices.

[0085] For ease of description, the above devices are described separately by function as various units. Of course, in implementing this application, the functions of each unit can be implemented in one or more software and / or hardware.

[0086] Those skilled in the art will understand that embodiments of this application can be provided as methods, apparatus, or computer devices. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0087] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus, or computer devices according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0088] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0089] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0090] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0091] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the apparatus embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0092] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

[0093] Although embodiments of this application have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of this application, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A compensation processing method for multi-band remote sensing images, characterized in that, The method includes: Acquire initial remote sensing images of multiple bands, and perform fixed difference processing on the initial remote sensing images to obtain first remote sensing images of multiple different bands. The fixed difference processing includes relative radiometric calibration and co-registration processing. A first remote sensing image pair is determined by any two first remote sensing images. Based on the anisotropic weighted circular mean algorithm, the first remote sensing image pair is registered at the subpixel level. Based on the registration result, the relative displacement curve of the first remote sensing image pair is determined, wherein the relative displacement curve represents the relative jitter of the line of sight between the first remote sensing image pair. Spectral analysis is performed on the relative displacement curve. Based on the analysis results, the main frequency components of line-of-sight jitter are determined. Based on the main frequency components, a line-of-sight jitter model is determined. Geometric compensation is then performed on the first remote sensing image according to the line-of-sight jitter model to obtain multiple compensated second remote sensing images. The anisotropic weighted circular mean algorithm is used to perform anisotropic weighted circular mean filtering on the normalized cross power spectrum of the first remote sensing image pair according to the filtering weight coefficients. The filter weight coefficients are determined by amplitude weights and spatial weights. The amplitude weights are determined based on the amplitude of the normalized cross-power spectrum, and the spatial weights are determined based on an anisotropic weighted kernel. The spatial weights are used to apply different filter intensities in the radial and tangential directions.

2. The method according to claim 1, characterized in that, The fixed difference processing of the initial remote sensing image includes: The initial remote sensing image is subjected to relative radiometric calibration to eliminate differences in radiometric response between linear array detectors of different bands; The initial remote sensing images after radiometric calibration are co-registered to eliminate fixed position offsets caused by the physical spacing of linear array detectors in different bands.

3. The method according to claim 1, characterized in that, The first remote sensing image comprises multiple consecutive image blocks divided according to scanning time; subpixel-level registration of the first remote sensing image pair includes: The first image block and the second image block of the first remote sensing image pair at any scanning time are obtained, and the normalized cross power spectrum of the first image block and the second image block is obtained. The normalized cross power spectrum is filtered based on the anisotropic weighted circular mean algorithm. The normalized cross-power spectrum after filtering is approximated by a rank-one subspace to obtain relevant phase information. The relevant phase information is then subjected to one-dimensional expansion and fitting to obtain the subpixel-level offset of the first remote sensing image pair at the scanning time.

4. The method according to claim 3, characterized in that, The filtering process for the normalized cross-power spectrum based on the anisotropic weighted circular mean algorithm includes: Within the set filtering window, the normalized cross-power spectrum is subjected to weighted circular mean filtering based on the filtering weight coefficients.

5. The method according to claim 4, characterized in that, The anisotropic weighted kernel represents the filtering intensity in different directions. The anisotropic weighted kernel is determined based on the standard deviation of the first direction and the standard deviation of the second direction, wherein the standard deviation of the first direction is greater than the standard deviation of the second direction.

6. The method according to claim 1 or 3, characterized in that, The relative displacement curves of the first remote sensing image pair determined based on the registration results include: Obtain the subpixel-level offset sequence of the first remote sensing image pair, the subpixel-level offset sequence including the subpixel-level offset of the first remote sensing image at each scanning time; The relative displacement curve of the first remote sensing image pair is constructed based on the subpixel-level offset sequence, wherein the horizontal axis of the relative displacement curve represents each scanning time of the first remote sensing image pair, and the vertical axis of the relative displacement curve represents the subpixel-level offset corresponding to the scanning time.

7. The method according to claim 1, characterized in that, The main frequency components include the target fundamental frequency component and harmonic components; spectral analysis is performed on the relative displacement curve, and based on the analysis results, the main frequency components of the line-of-sight jitter are determined to include: The relative displacement curve is preprocessed, and the preprocessed relative displacement curve is subjected to Fourier transform to obtain the amplitude spectrum of the relative displacement curve. One or more target extreme points are determined in the amplitude spectrum, and the frequency resolution range of each target extreme point is traversed to determine one or more target fundamental frequency components and their harmonic components of the line-of-sight jitter, and the target fundamental frequency components and their harmonic components are used as the main frequency components.

8. The method according to claim 1, characterized in that, The line-of-sight jitter model is used to describe the change of line-of-sight jitter over time; line-of-sight jitter compensation is performed on the first remote sensing image according to the line-of-sight jitter model to obtain multiple compensated second remote sensing images, including: The line-of-sight jitter amount is determined at each scanning moment according to the line-of-sight jitter model, wherein the line-of-sight jitter amount represents the positional offset of the line of sight of the remote sensing payload at the scanning moment; Based on the jitter of each line of sight, the first remote sensing image is geometrically corrected by image resampling, and the geometrically corrected first remote sensing image is used as the second remote sensing image.

9. A compensation processing device for multi-band remote sensing images, characterized in that, The device includes: The image acquisition unit is used to acquire initial remote sensing images of multiple bands and perform fixed difference processing on the initial remote sensing images to obtain first remote sensing images of multiple different bands. The fixed difference processing includes relative radiometric calibration and co-registration processing. The registration processing unit is configured to perform sub-pixel-level registration on any pair of first remote sensing images composed of two first remote sensing images, based on an anisotropic weighted circular mean algorithm, and determine the relative displacement curve of the first remote sensing image pair based on the registration result, wherein the relative displacement curve characterizes the relative jitter of the line of sight between the first remote sensing image pairs; wherein the anisotropic weighted circular mean algorithm is configured to perform anisotropic weighted circular mean filtering on the normalized cross power spectrum of the first remote sensing image pair according to the filtering weight coefficients; wherein the filtering weight coefficients are determined by amplitude weight and spatial weight, the amplitude weight is determined based on the amplitude of the normalized cross power spectrum, the spatial weight is determined based on the anisotropic weighting kernel, and the spatial weight is used to apply different filtering intensities in the radial and tangential directions; The jitter compensation unit is used to perform spectral analysis on the relative displacement curve, determine the main frequency components of line-of-sight jitter based on the analysis results, determine the line-of-sight jitter model based on the main frequency components, and perform line-of-sight jitter compensation on the first remote sensing image according to the line-of-sight jitter model to obtain multiple compensated second remote sensing images.

10. A computer device, characterized in that, include: The system includes a memory and a processor, which are interconnected. The memory stores computer instructions, and the processor executes the computer instructions to perform the compensation processing method for multi-band remote sensing images according to any one of claims 1 to 8.

Images