Deep space target data fusion enhancement method based on multi-scale low-rank structure characterization

Abstract

The invention discloses a deep space target data fusion enhancement method based on multi-scale low-rank structure representation. The method comprises the following steps: collecting a deep space target multi-source image and partial non-image data; preprocessing the collected data and representing the data as a multi-scale data form; performing significant feature extraction on the multi-scale data by using low-rank sparse matrix decomposition; fusing the low-rank structure features on each scale; post-processing operations of denoising, sharpening and enhancing are carried out on the fused image; and inputting the fused feature representation into a classification or detection network, and executing a specific task. According to the method, the problems of redundancy, noise and scale difference of multispectral data are solved, the space target detection precision and robustness are improved, redundancy and noise interference are effectively reduced, and the accuracy of data fusion is improved. Through a convolution fusion multi-source data fusion strategy, the data dimension is reduced, the complexity between different data sources is eliminated, and efficient fusion of multi-source data represented by a low-rank structure is realized.

Description

Technical Field

[0001] The present invention relates to a deep space target data fusion enhancement method based on multi-scale low-rank structure representation, and belongs to the technical field of deep space detection. Background Art

[0002] Traditional deep space target data fusion technology is a method of fusing visible light images, infrared images and infrared spectrometer data obtained in deep space target detection. Visible light imaging technology generates images by capturing radiation in the visible light band (usually 400nm to 700nm) through optical systems such as astronomical telescopes. This technology provides high-resolution image information for deep space detection and is widely used in astronomy, space exploration, satellite navigation and other fields. Infrared imaging technology usually consists of infrared sensors and optical systems, and acquires images by detecting infrared radiation emitted by objects (wavelengths are usually between 0.75 microns and 1000 microns). These images can not only reveal the visible characteristics of objects, but also provide key information about the temperature and material composition of objects. Infrared spectrometer technology is based on the absorption and emission characteristics of infrared radiation. By detecting the reflection, absorption or emission spectrum of objects in the infrared band, information such as the molecular structure, chemical composition and temperature of the substance is obtained.

[0003] Traditional deep space target data fusion technology has certain defects in multi-scale and multi-modal data processing. The traditional processing method is often to directly fuse the multi-modal data after processing them separately. This processing method may lead to the loss of key information, thus affecting the effect of data fusion. Since infrared, visible light image data and non-image infrared spectrum data lack synchronization in time, space and scale, traditional fusion technology may not be able to handle this data inconsistency, resulting in information loss in the fused data. Moreover, due to the dim characteristics of space targets, the fused data obtained by traditional methods often cannot fully utilize the cross-modal correlation between multi-spectral data for feature enhancement, resulting in information loss. In addition, traditional data fusion methods often involve a large number of operations such as alignment, interpolation and registration when processing multi-modal data, which increases the burden of calculation and storage.

[0004] Low-rank feature representation methods can effectively learn the correlation between different modes when fusing multimodal data, thereby achieving more comprehensive information integration, and can effectively reduce the redundant information of the data, making the calculation more efficient. Deep space target data fusion technology based on low-rank feature representation has stronger noise removal ability, higher fusion accuracy, robustness and flexibility, and can achieve multi-scale fusion and feature enhancement of multimodal deep space target data, while meeting the real-time and accuracy requirements of deep space target detection tasks. Summary of the invention

[0005] The purpose of the present invention is to provide a deep space target data fusion enhancement method based on multi-scale low-rank structure characterization.

[0006] In order to achieve the above object, the present invention provides a deep space target data fusion enhancement method based on multi-scale low-rank structure representation, comprising the following steps: Step 1: Collect multispectral data from the same space target under different time sequence conditions, including visible light images, infrared images and infrared spectrum data; Step 2, preprocessing the visible light and infrared image data collected in Step 1 by using mean filter denoising and adaptive histogram equalization (AHE) methods, and preprocessing the infrared spectrum data by using Gaussian smoothing denoising and polynomial baseline correction methods; Step 3: Use the Gaussian pyramid method to decompose the multi-source data obtained in Step 2 into a multi-scale matrix, use the alternating least squares optimization algorithm to solve the low-rank sparse matrix decomposition (LRaSMD), and extract shared features from multi-time series and multi-spectral data; Step 4: Use the cross-modal attention mechanism to fuse multi-scale low-rank structural features to obtain complete fused feature data; Step 5: Perform denoising and image sharpening on the fused feature data obtained in Step 4 to obtain the final multi-source fused feature data.

[0007] Furthermore, in the Step 2, the collected visible light and infrared image data are preprocessed by the method of mean filtering denoising and adaptive histogram equalization, and the infrared spectrum data are preprocessed by the method of Gaussian smoothing denoising and polynomial baseline correction. Step 2.1-1, read in visible light and infrared image data, and use the mean filtering based denoising method to denoise the image. Mean filtering is a common image denoising method that replaces the value of each pixel with the average value of its neighboring pixels to smooth the image and remove noise. Indicates the position in the image The pixel value of Indicates that the image The filter neighborhood is centered on . A 5×5 filter size is selected and the image boundary is processed by replicating the edges. The output of the mean filter It is expressed by the following formula: ; Assign the calculated mean to the current pixel to smooth the image.

[0008] Step 2.1-2, use the adaptive histogram equalization method to enhance the visible light and infrared images. Adaptive Histogram Equalization (AHE) is an image enhancement method that aims to improve local details by adjusting the contrast of the image. The entire image is divided into 8×8 windows. For each local window , calculate its histogram normalized cumulative distribution function , the calculation formula is as follows: ; in, It is a local window In pixel value The normalized histogram frequency on , is the number of pixels in the window, and L is the number of gray levels of the image data; For each original pixel in the image , according to the histogram normalized cumulative distribution function value of the local window where it is located, update the value of the pixel to , the calculation formula is: ; Step 2.2-1, use Gaussian smoothing method to denoise the infrared spectrum data. Gaussian smoothing is a common image denoising method. It smoothes the image and removes noise by applying Gaussian filter. Gaussian filter is based on Gaussian function (normal distribution function). Its weight gradually decreases as the distance from the center point increases. Therefore, it can effectively smooth the details and noise in the image while retaining the main structure and edge information of the image. For each pixel point , use Gaussian filter to perform weighted averaging on the surrounding neighborhood to obtain the smoothed pixel value , the calculation formula is as follows: ; Step 2.2-2, use the polynomial baseline correction method to preprocess the infrared spectrum data. Polynomial baseline correction is a technology used to remove baseline drift or background noise in the signal. It is widely used in spectral data processing. This method represents the baseline part of the signal by fitting a polynomial curve (usually a low-order polynomial, such as a first-order or second-order polynomial) of the signal, and then removes the baseline part from the original signal to correct the real information of the signal. Assuming that the original data is a Wavelength values ​​and corresponding absorption intensities An array of: ; in, is the wavelength, is the corresponding absorption intensity; Then, the baseline drift region in the spectral data is identified and a polynomial function of a certain order is selected. , the baseline drift in the spectrum is fitted by the least squares method, and the obtained baseline drift polynomial form is: ; The baseline drift polynomial is fitted , from the original spectrum The baseline drift is removed to obtain the spectrum data after the baseline drift is removed. , the calculation formula is: ; Furthermore, in Step 3, the Gaussian pyramid method is used to decompose the multi-source data obtained in Step 2 into a multi-scale matrix form, and the alternating least squares optimization algorithm is used to solve the low-rank sparse matrix decomposition (LRSMD). The steps of extracting shared features from multi-time series and multi-spectral data are as follows: Step 3.1. In order to extract the main components of multi-source data, the Gaussian pyramid analysis method is used to decompose the data into multiple scales. The Gaussian pyramid analysis method is a multi-scale image processing technology that generates a series of images with different resolutions by continuously applying Gaussian filtering to the image and gradually downsampling it. Represent the preprocessed deep space target multi-source data as a matrix , as the first layer of the pyramid, that is, the bottom layer of the Gaussian pyramid; First, apply Gaussian filtering to the first layer of data. The filter calculation formula is: ; in, is the standard deviation, which determines the width of the filter and controls the degree of smoothing; Downsample the Gaussian filtering result to get the second layer of data ; Then, Gaussian filtering and downsampling are continued on the second layer data to obtain the third layer data. , thus obtaining multi-source data information at three different scales. Each layer of the Gaussian pyramid represents the characteristics of the data at different scales; Step 3.2-1, in order to extract the shared features of multi-source and multi-scale data, the alternating least squares optimization algorithm is used to perform low-rank sparse matrix decomposition (LRaSMD) on the data. This method aims to simultaneously recover the low-rank part and the sparse part from a given matrix to remove noise and extract the main structural information; set up For data of various scales, the purpose of low-rank sparse matrix decomposition is to Decompose into and The sum of the matrices of the two parts, where is a low-rank matrix that represents the main structure or trend of the data, It is a sparse matrix, representing outliers, noise or rare features in the data. This method uses the alternating least squares optimization method to optimize the low-rank sparse matrix decomposition, that is, in each step, one matrix is ​​fixed and the other matrix is ​​minimized, so as to gradually approach the optimal solution; Initially, the matrix and Initialize, Initialized as the singular value decomposition low-rank approximation matrix of matrix M, The matrix is ​​initialized to a zero matrix. The optimization process of low-rank sparse matrix decomposition is to minimize the decomposition error between the low-rank matrix and the sparse matrix. The error formula is: ; in, represents the norm, rank represents the matrix rank, card represents the number of matrix components, and r represents the low-rank background matrix The maximum rank, k represents a sparse matrix The sparsity reflects the sparse components in the image and is usually defined as of norm; Step 3.2-2, for low-rank matrix and sparse matrices Perform alternating optimization to gradually reduce the decomposition error of low-rank matrices and sparse matrices through the alternating optimization process; First fix the sparse matrix , optimize the low-rank matrix , the calculation formula of the optimization process is: ; Where t is the dimension of the data matrix; Then, fix the low-rank matrix , optimize the sparse matrix , the calculation formula of the optimization process is: ; As the number of iterations increases, the decomposition error decreases monotonically; ( is the selected fault tolerance coefficient), the optimization process is judged to be converged, and a low-rank matrix corresponding to the background information and a sparse matrix corresponding to the abnormal target information are obtained; Furthermore, the steps of using the cross-modal attention mechanism in Step 4 to fuse multi-scale low-rank structural features to obtain complete fused feature data are as follows: By adaptively learning the relationship between different modalities and different scales through the cross-modal attention mechanism, the model can automatically focus on important features and suppress redundant or irrelevant information, thereby effectively fusing multi-source information together. Specifically, for image features and non-image features Define the query, key, and value respectively, and the calculation formula is: ; in, , , is the learned weight matrix, which is used to map queries, keys, and values ​​respectively; Query by calculation and key The dot product of , we get the attention weight, and the calculation formula is: ; in, is the dimension of the key, used to scale the dot product to prevent gradient explosion; Image features are concatenated through feature concatenation and non-image features Fusion, get fusion features , the calculation formula is: ; Furthermore, in Step 5, the fused feature data obtained in Step 4 is subjected to data post-processing of denoising and image sharpening to obtain the final multi-source fused feature data: The low-rank recovery algorithm is used to denoise the fused feature information. The data matrix is ​​decomposed into the sum of a low-rank matrix and a sparse matrix. The low-rank matrix represents the main pattern and structure of the data, while the sparse matrix contains noise or outliers, thereby achieving the denoising effect. Specifically, the fused feature matrix F obtained in Step 4 is first decomposed by SVD to obtain: ; in, and are singular vectors, is the singular value matrix; Then, the singular value matrix is ​​thresholded, that is, singular values ​​less than a certain threshold are set to zero, and we get: ; in, is the threshold parameter, which controls the degree of compression of singular values; Finally, the low-rank matrix is ​​obtained by matrix reconstruction : ; Through the above denoising operation, a low-rank matrix is ​​obtained that removes noise and retains the main structure of the original matrix ; The Laplace filtering method is used to sharpen and enhance the feature data. Specifically, in the application process, the original feature data is first subjected to Laplace filtering to obtain the second-order derivative of the data (i.e., the rate of change). Then, the edge part of the image or signal is enhanced by weighted superposition of the result after Laplace filtering and the original data (usually the weighting coefficient is set to a positive value), thereby improving the detail contrast and structural information, and achieving the effect of feature enhancement; set up Represents the matrix data before processing. The calculation formula of the Laplace filter method is: ; in, is a parameter that controls the strength of sharpening.

[0009] After the above post-processing operations, the final multi-scale fusion enhanced image data of deep space targets is obtained, which can meet the needs of subsequent deep space target detection and other tasks.

[0010] Beneficial effects: It solves the problems of redundancy, noise and scale difference of multi-spectral data, can make full use of multi-source deep space target images and spectral data, improves data fusion effect and fused image quality, has strong robustness and robustness, effectively reduces redundancy and noise interference, improves the accuracy of data fusion, and can reduce the complexity of calculation and storage, and realize the efficient fusion of multi-source data characterized by low-rank structure, which is suitable for the detection and recognition tasks of deep space dim target objects. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0012] Figure 1 It is a schematic diagram of the process of the present invention; Figure 2 It is a flowchart of multi-scale decomposition of multi-source data using Gaussian pyramid method; Figure 3 It is a flowchart of low-rank sparse matrix decomposition using alternating least squares optimization algorithm; Figure 4 This is a schematic diagram of the principle of using a cross-modal attention mechanism to fuse multi-source low-rank structural features. DETAILED DESCRIPTION

[0013] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0014] like Figure 1 As shown, the deep space target data fusion enhancement method based on multi-scale low-rank structure representation includes the following steps: Step 1: Collect multispectral data from the same space target under different time sequence conditions, including visible light images, infrared images and infrared spectrum data; Step 2, preprocessing the visible light and infrared image data collected in Step 1 by using mean filter denoising and adaptive histogram equalization (AHE) methods, and preprocessing the infrared spectrum data by using Gaussian smoothing denoising and polynomial baseline correction methods; Step 3: Use the Gaussian pyramid method to decompose the multi-source data obtained in Step 2 into a multi-scale matrix, use the alternating least squares optimization algorithm to solve the low-rank sparse matrix decomposition (LRaSMD), and extract shared features from multi-time series and multi-spectral data; Step 4: Use the cross-modal attention mechanism to fuse multi-scale low-rank structural features to obtain complete fused feature data; Step 5: Perform denoising and image sharpening on the fused feature data obtained in Step 4 to obtain the final multi-source fused feature data.

[0015] As a preferred implementation, the specific steps of Step 2 are: Step 2.1-1, read in visible light and infrared image data, and use the mean filtering based denoising method to denoise the image. Mean filtering is a common image denoising method that replaces the value of each pixel with the average value of its neighboring pixels to smooth the image and remove noise. Indicates the position in the image The pixel value of Indicates that the image The filter neighborhood is centered on . A 5×5 filter is selected and the image boundary is processed by copying the edge. The output of the mean filter It is expressed by the following formula: ; Assign the calculated mean to the current pixel to smooth the image; Step 2.1-2, use the adaptive histogram equalization method to enhance the visible light and infrared images. Adaptive Histogram Equalization (AHE) is an image enhancement method that aims to improve local details by adjusting the contrast of the image. The entire image is divided into 8×8 windows. For each local window , calculate its histogram normalized cumulative distribution function , the calculation formula is as follows: ; in, It is a local window In pixel value The normalized histogram frequency on , is the number of pixels in the window, and L is the number of gray levels of the image data; For each original pixel in the image , according to the histogram normalized cumulative distribution function value of the local window where it is located, update the value of the pixel to , the calculation formula is: ; Step 2.2-1, use Gaussian smoothing method to denoise the infrared spectrum data. Gaussian smoothing is a common image denoising method. It smoothes the image and removes noise by applying Gaussian filter. Gaussian filter is based on Gaussian function (normal distribution function). Its weight gradually decreases as the distance from the center point increases. Therefore, it can effectively smooth the details and noise in the image while retaining the main structure and edge information of the image. For each pixel point , use Gaussian filter to perform weighted averaging on the surrounding neighborhood to obtain the smoothed pixel value , the calculation formula is as follows: ; Step 2.2-2, use the polynomial baseline correction method to preprocess the infrared spectrum data. Polynomial baseline correction is a technology used to remove baseline drift or background noise in the signal. It is widely used in spectral data processing. This method represents the baseline part of the signal by fitting a polynomial curve (usually a low-order polynomial, such as a first-order or second-order polynomial) of the signal, and then removes the baseline part from the original signal to correct the real information of the signal. Assuming that the original data is a Wavelength values ​​and corresponding absorption intensities An array of: ; in, is the wavelength, is the corresponding absorption intensity; Then, the baseline drift region in the spectral data is identified and a polynomial function of a certain order is selected. , the baseline drift in the spectrum is fitted by the least squares method, and the obtained baseline drift polynomial form is: ; The baseline drift polynomial is fitted , from the original spectrum The baseline drift is removed to obtain the spectrum data after the baseline drift is removed. , the calculation formula is: ; As a preferred implementation, the specific steps of Step 3 are: Step 3.1, in order to extract the main components of multi-source data, the Gaussian pyramid analysis method is used to decompose the data into multiple scales. The Gaussian pyramid analysis method is a multi-scale image processing technology that generates a series of images with different resolutions by continuously applying Gaussian filtering to the image and gradually downsampling it, such as Figure 2 As shown; Represent the preprocessed deep space target multi-source data as a matrix , as the first layer of the pyramid, that is, the bottom layer of the Gaussian pyramid; First, apply Gaussian filtering to the first layer of data, and the calculation formula is: ; in, is the standard deviation, which determines the width of the filter and controls the degree of smoothing; Downsample the Gaussian filtering result to get the second layer of data ; Then, Gaussian filtering and downsampling are continued on the second layer data to obtain the third layer data. , thus obtaining multi-source data information at three different scales. Each layer of the Gaussian pyramid represents the characteristics of the data at different scales; Step 3.2-1, in order to extract the shared features of multi-source and multi-scale data, the alternating least squares optimization algorithm is used to perform low-rank sparse matrix decomposition (LRaSMD) on the data. This method aims to simultaneously recover the low-rank part and the sparse part from a given matrix to remove noise and extract the main structural information; set up For data of various scales, the purpose of low-rank sparse matrix decomposition is to Decompose into and The sum of the matrices of the two parts, where is a low-rank matrix that represents the main structure or trend of the data, is a sparse matrix, representing outliers, noise or rare features in the data. This method uses the alternating least squares optimization method to optimize the low-rank sparse matrix decomposition, that is, in each step, one matrix is ​​fixed and the other matrix is ​​minimized, so as to gradually approach the optimal solution, such as Figure 3 As shown; Initially, the matrix and Initialize, Initialized as the singular value decomposition low-rank approximation matrix of matrix M, The matrix is ​​initialized to a zero matrix. The optimization process of low-rank sparse matrix decomposition is to minimize the decomposition error between the low-rank matrix and the sparse matrix. The error formula is: ; in, represents the norm, rank represents the matrix rank, card represents the number of matrix components, and r represents the low-rank background matrix The maximum rank, k represents a sparse matrix The sparsity reflects the sparse components in the image and is usually defined as of norm; Step 3.2-2, for low-rank matrix and sparse matrices Perform alternating optimization to gradually reduce the decomposition error of low-rank matrices and sparse matrices through the alternating optimization process; First fix the sparse matrix , optimize the low-rank matrix , the calculation formula of the optimization process is: ; Where t is the dimension of the data matrix; Then, fix the low-rank matrix , optimize the sparse matrix , the calculation formula of the optimization process is: ; As the number of iterations increases, the decomposition error decreases monotonically. ( is the selected fault tolerance coefficient), the optimization process is judged to converge, and a low-rank matrix corresponding to the background information and a sparse matrix corresponding to the abnormal target information are obtained.

[0016] As a preferred implementation, the specific steps of Step 4 are: By adaptively learning the relationship between different modalities and different scales through the cross-modal attention mechanism, the model can automatically focus on important features and suppress redundant or irrelevant information, thereby effectively fusing multi-source information together, such as Figure 4 As shown, specifically, the image features and non-image features Define the query, key, and value respectively, and the calculation formula is: ; in, , , is the learned weight matrix, which is used to map queries, keys, and values ​​respectively; Query by calculation and key The dot product of , we get the attention weight, and the calculation formula is: ; in, is the dimension of the key, used to scale the dot product to prevent gradient explosion; Image features are concatenated through feature concatenation and non-image features Fusion, get fusion features , the calculation formula is: ; As a preferred implementation, the specific steps of Step 5 are: The low-rank recovery algorithm is used to denoise the fused feature information. The data matrix is ​​decomposed into the sum of a low-rank matrix and a sparse matrix. The low-rank matrix represents the main pattern and structure of the data, while the sparse matrix contains noise or outliers, thereby achieving the denoising effect. Specifically, the fused feature matrix F obtained in Step 4 is first decomposed by SVD to obtain: ; in, and are singular vectors, is the singular value matrix; Then, the singular value matrix is ​​thresholded, that is, singular values ​​less than a certain threshold are set to zero, and we get: ; in, is the threshold parameter, which controls the degree of compression of singular values; Finally, the low-rank matrix is ​​obtained by matrix reconstruction : ; Through the above denoising operation, a low-rank matrix is ​​obtained that removes noise and retains the main structure of the original matrix .

[0017] The Laplace filtering method is used to sharpen and enhance the feature data. Specifically, in the application process, the original feature data is first subjected to Laplace filtering to obtain the second-order derivative of the data (i.e., the rate of change). Then, the edge part of the image or signal is enhanced by weighted superposition of the result after Laplace filtering and the original data (usually the weighting coefficient is set to a positive value), thereby improving the detail contrast and structural information, and achieving the effect of feature enhancement; set up Represents the matrix data before processing. The calculation formula of the Laplace filter method is: ; in, is a parameter that controls the strength of sharpening.

[0018] After the above post-processing operations, the final deep space target multi-scale fusion enhanced image data is obtained, which can meet the needs of subsequent deep space target detection and other tasks.

[0019] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A deep space target data fusion enhancement method based on multi-scale low-rank structure representation, characterized in that: The steps include: Step 1: Collect multispectral data from the same space target under different time sequence conditions, including visible light images, infrared images and infrared spectrum data; Step 2: Preprocess the visible light and infrared image data collected in Step 1 by using mean filtering denoising and adaptive histogram equalization, and preprocess the infrared spectrum data by using Gaussian smoothing denoising and polynomial baseline correction; Step 3: Use the Gaussian pyramid method to decompose the multi-source data obtained in Step 2 into a multi-scale matrix form, use the alternating least squares optimization algorithm to solve the low-rank sparse matrix decomposition, and extract shared features from multi-time series and multi-spectral data; Step 4: Use the cross-modal attention mechanism to fuse multi-scale low-rank structural features to obtain complete fused feature data; Step 5: Perform denoising and image sharpening on the fused feature data obtained in Step 4 to obtain the final multi-source fused feature data.

2. The deep space target data fusion enhancement method based on multi-scale low-rank structure representation according to claim 1 is characterized in that: The specific steps of Step 2 are: Step 2.1-1, read the visible light and infrared image data, and use the median filter-based denoising method to denoise the image. Indicates the position in the image The pixel value of Indicates that the image The filter neighborhood is centered, a 5×5 filter is selected, and the image boundary is processed by copying the edge. The output of the median filter It is expressed by the following formula: ； Assign the calculated mean to the current pixel to smooth the image; Step 2.1-2, use the adaptive histogram equalization method to enhance the visible light and infrared images; divide the entire image into 8×8 windows, and for each local window , calculate its histogram normalized cumulative distribution function , the calculation formula is as follows: ； in, It is a local window In pixel value The normalized histogram frequency on , is the number of pixels in the window, and L is the number of gray levels of the image data; For each original pixel in the image , according to the histogram normalized cumulative distribution function value of the local window where it is located, update the value of the pixel to , the calculation formula is: ； Step 2.2-1, use Gaussian smoothing method to denoise the infrared spectrum data. For each pixel , use Gaussian filter to perform weighted averaging on the surrounding neighborhood to obtain the smoothed pixel value , the calculation formula is as follows: ； Step 2.2-2, use the polynomial baseline correction method to preprocess the infrared spectrum data, assuming that the original data is a Wavelength values ​​and corresponding absorption intensities An array of: ； in, is the wavelength, is the corresponding absorption intensity; Subsequently, the baseline drift regions in the spectral data were identified and the polynomial function was selected. , the baseline drift in the spectrum is fitted by the least squares method, and the obtained baseline drift polynomial form is: ； The baseline drift polynomial is fitted , from the original spectrum The baseline drift is removed to obtain the spectrum data after baseline drift is removed. , the calculation formula is: 。 3. The deep space target data fusion enhancement method based on multi-scale low-rank structure representation according to claim 1 is characterized in that: The specific steps of Step 3 are: Step 3.1-1, in order to extract the main components of multi-source data, the Gaussian pyramid analysis method is used to decompose the data into multiple scales; Represent the preprocessed deep space target multi-source data as a matrix , as the first layer of the pyramid, that is, the bottom layer of the Gaussian pyramid; First, apply Gaussian filtering to the first layer of data. The filter calculation formula is: ； in, is the standard deviation, which determines the width of the filter and controls the degree of smoothing; Downsample the Gaussian filtering result to get the second layer of data ; Then, Gaussian filtering and downsampling are continued on the second layer data to obtain the third layer data. , thus obtaining multi-source data information at three different scales. Each layer of the Gaussian pyramid represents the characteristics of the data at different scales; Step 3.2-1, in order to extract the shared features of multi-source and multi-scale data, the alternating least squares optimization algorithm is used to perform low-rank sparse matrix decomposition on the data. For data of various scales, the purpose of low-rank sparse matrix decomposition is to Decompose into and The sum of the matrices of the two parts, where is a low-rank matrix that represents the main structure or trend of the data, It is a sparse matrix, representing outliers, noise or rare features in the data. The alternating least squares optimization method is used to optimize the low-rank sparse matrix decomposition, that is, in each step, one matrix is ​​fixed and the other matrix is ​​minimized, so as to gradually approach the optimal solution; Initially, the matrix and Initialize, Initialized as the singular value decomposition low-rank approximation matrix of matrix M, The matrix is ​​initialized to a zero matrix. The optimization process of low-rank sparse matrix decomposition is to minimize the decomposition error between the low-rank matrix and the sparse matrix. The error formula is: ； in, represents the norm, rank represents the matrix rank, card represents the number of matrix components, and r represents the low-rank background matrix The maximum rank, k represents a sparse matrix The sparsity reflects the sparse components in the image and is defined as of norm; Step 3.2-2, for low-rank matrix and sparse matrices Perform alternating optimization: First fix the sparse matrix , optimize the low-rank matrix , the calculation formula of the optimization process is: ； Where t is the dimension of the data matrix; Then, fix the low-rank matrix , optimize the sparse matrix , the calculation formula of the optimization process is: ； As the number of iterations increases, the decomposition error decreases monotonically. When , the optimization process converges, and the low-rank matrix corresponding to the background information and the sparse matrix corresponding to the abnormal target information are obtained. is the selected fault tolerance factor.

4. The deep space target data fusion enhancement method based on multi-scale low-rank structure representation according to claim 1 is characterized in that: The specific steps of Step 4 are: Step 4.1-1, image features and non-image features Define the query, key, and value respectively, and the calculation formula is: ； in, , , is the learned weight matrix, which is used to map queries, keys, and values ​​respectively; Step 4.1-2, query by calculation and key The dot product of , we get the attention weight, and the calculation formula is: ； in, is the dimension of the key; Step 4.1-3, image features are stitched together and non-image features Fusion, get fusion features , the calculation formula is: 。 5. The deep space target data fusion enhancement method based on multi-scale low-rank structure representation according to claim 1 is characterized in that: The specific steps of Step 5 are: Step 5.1, use low-rank recovery algorithm to denoise the fused feature information; First, perform SVD decomposition on the fusion feature matrix F obtained in Step 4 to obtain: ； in, and are singular vectors, is the singular value matrix; Thresholding the singular value matrix, that is, setting the singular values ​​less than a certain threshold to zero, we get: ； in, is the threshold parameter, which controls the degree of compression of singular values; Finally, the low-rank matrix is ​​obtained by matrix reconstruction : ； Through the above denoising operation, a low-rank matrix is ​​obtained that removes noise and retains the main structure of the original matrix ; Step 5.2-1, use Laplace filtering method to sharpen and enhance the feature data. Represents the matrix data before processing, and the calculation formula of the Laplace filter method is: ； in, is the parameter that controls the sharpening strength; After the above post-processing operations, the final deep space target multi-scale fusion enhanced image data is obtained.

Images