Remote sensing image space-time fusion method combining optical flow and spectrum correction factor

Abstract

The invention relates to a remote sensing image space-time fusion method combining an optical flow and a spectrum correction factor. The method comprises the following steps: dividing a surface feature vegetation index difference image into a change region and a non-change region; calculating a time weight matrix and a global space weight matrix for the change area and the non-change area according to the motion sum of all the pixels on the optical flow motion vector, and fusing the time weight matrix and the global space weight matrix to obtain a global space-time weight matrix; performing spatial interpolation on the low-space high-temporal-resolution image at the T1 moment by using cubic spline interpolation to obtain a spatial interpolation image; performing Gaussian low-pass filtering on the high-space low-time-resolution image at the T0 moment, and performing correlation calculation on the filtered image and the low-space high-time-resolution image at the T0 moment to obtain a spectral difference correction factor matrix; and fusing the spatial interpolation image and the high-space low-temporal-resolution image at the T0 moment by using the spectral difference correction factor matrix and the global space-time weight matrix to obtain a high-time high-spatial-resolution image at the T1 moment. According to the invention, the fusion precision and efficiency are improved.

Description

Technical Field

[0001] The invention belongs to the field of remote sensing image processing, and in particular to a remote sensing image spatiotemporal fusion method combining optical flow and spectral correction factors. Background Art

[0002] In the field of remote sensing technology, achieving high precision, high spatial resolution and continuity in surface observation has always been a technical bottleneck. Limited by factors such as sensor technology, funding and weather, traditional remote sensing data acquisition methods often have a trade-off between temporal resolution and spatial resolution. Remote sensing images with high spatial resolution can provide rich surface detail information, but their temporal resolution is often low and cannot capture surface changes in real time; while remote sensing images with high temporal resolution can reflect surface changes in real time, their spatial resolution is low and cannot meet the precise demand for surface detail information.

[0003] In order to solve this problem, spatiotemporal fusion technology came into being. This technology aims to obtain surface observation data with both high temporal resolution and high spatial resolution by fusing remote sensing images with different resolution characteristics. However, there are still many problems with the existing weight-based spatiotemporal fusion methods. For example, most methods use a fixed-size sliding window for local time or space weighting, which makes it difficult to define the optimal sliding window size, and the local information obtained ignores the continuity of the ground objects. In addition, although the object-based method takes into account the integrity of the shape of the ground object, the reliability of its classifier directly affects the fusion result. Although the deep learning-based method can obtain multi-level local and continuous spatial information at multiple scales and multiple perspectives, it ignores the extraction of dynamic time information.

[0004] In terms of residual calculation, existing methods often use TPS (Thin Plate Spline) or learning-based spatial interpolation methods, which are inefficient when processing large-scale input data and seriously affect prediction accuracy. More importantly, existing methods only consider spatial or temporal information in residual calculation, while ignoring the spectral differences caused by different sensors. The existence of this spectral difference will affect the fusion accuracy to a certain extent, resulting in the fused remote sensing image being unable to accurately reflect the true situation of the surface. Summary of the invention

[0005] In order to solve the problems existing in the background technology and predict remote sensing images with high temporal and spatial resolution, the present invention provides a remote sensing image spatiotemporal fusion method combining optical flow and spectral correction factor, comprising:

[0006] S1: Based on the vegetation index difference image calculated from the low spatial and high temporal resolution images at time T1 and time T0, an adaptive threshold factor is constructed to divide the vegetation index difference image into a change area and a non-change area;

[0007] S2: Based on the dense optical flow between the low spatial and high temporal resolution images from time T0 to time T1, the sum of the motions of all pixels on the optical flow motion vector is calculated;

[0008] S3: Calculate the temporal weights for the changed area and the unchanged area respectively according to the sum of the motions of all pixels on the optical flow motion vector, and obtain the global temporal weight matrix of all pixels;

[0009] S4: Divide each pixel value in the high spatial and low temporal resolution image at time T0 by the average value of all pixels to calculate the global spatial weight matrix of all pixels;

[0010] S5: Fusing the global time weight matrix and the global space weight matrix to obtain the global space-time weight matrix;

[0011] S6: Using cubic spline interpolation to perform spatial interpolation operation on the low spatial and high temporal resolution image at time T1 to obtain a spatial interpolation image;

[0012] S7: Perform Gaussian low-pass filtering on the high spatial and low temporal resolution image at time T0, and perform correlation calculation on the image after Gaussian low-pass filtering and the low spatial and high temporal resolution image at time T0 to obtain a spectral difference correction factor matrix;

[0013] S8: Multiply the spectral difference correction factor matrix and the spatial interpolation image to obtain the residual image, use the global spatiotemporal weight matrix to weight the high spatial and low temporal resolution image at time T0, and add the weighted image and the residual image to obtain the high temporal and high spatial resolution image at time T1.

[0014] Preferably, the step S1 specifically includes:

[0015] S11: Calculate the vegetation index images of low spatial and high temporal resolution images at time T1 and time T0 respectively;

[0016] S12: subtracting the vegetation index image at time T0 from the vegetation index image at time T1 to obtain a vegetation index difference image;

[0017] S13: Calculate the absolute value of each pixel in the vegetation index difference image and the absolute value of the average value of all pixels, and use the standard deviation calculated by subtracting the absolute value of the average value of all pixels from the absolute value of each pixel as the adaptive threshold factor;

[0018] S14: using a decision tree to determine the size of the pixel value of each pixel in the vegetation index difference image and the adaptive threshold factor; when the pixel value of the pixel is less than the adaptive threshold factor, the pixel is divided into a non-changing area; otherwise, the pixel is divided into a changing area.

[0019] Preferably, the adaptive threshold factor comprises:

[0020]

[0021] Among them, |.| represents the absolute value, D represents the difference image of vegetation index, represents the average value of all pixels, std(.) represents the standard deviation calculation function, and V represents the adaptive threshold factor.

[0022] Preferably, calculating the time weights for the changed area and the unchanged area respectively comprises:

[0023] i belongs to the changing area

[0024] i belongs to the no-change region

[0025] in, and The temporal weights of the pixels belonging to the changed area and the unchanged area, represents the average pixel value of the high spatial and low temporal resolution image in the changing area at time T0, i represents the i-th pixel, and D i represents the pixel value of pixel i in the vegetation index difference image, F represents the sum of the motion of all pixels on the optical flow motion vector, N represents the number of pixels, It represents the average pixel value of the high spatial and low temporal resolution image in the changing area at time T0.

[0026] Preferably, the dense optical flow between the low spatial and high temporal resolution images from time T0 to time T1 is calculated by using the Python calcOpticalFlowFarneback method.

[0027] Preferably, the global spatial weight matrix comprises:

[0028]

[0029] Among them, F0(i) represents the pixel value of pixel i in the high spatial and low temporal resolution image at time T0, represents the pixel average of the high spatial and low temporal resolution image at time T0, SW i Represents the global spatial weight corresponding to pixel i.

[0030] Preferably, the global spatiotemporal weight matrix comprises:

[0031] STW i =SW i +TW i

[0032] Among them, STW i represents the global spatiotemporal weight corresponding to pixel i, TW i represents the global temporal weight corresponding to pixel i; SW i Represents the global spatial weight corresponding to pixel i.

[0033] Preferably, the spectrum difference correction factor matrix includes:

[0034]

[0035] Among them, SD i represents the spectral difference correction factor corresponding to pixel i, represents the pixel value corresponding to pixel i in the high spatial and low temporal resolution image at time T0 after Gaussian low-pass filtering, C0(i) represents the pixel value corresponding to pixel i in the high spatial and low temporal resolution image at time T0, and |.| represents the absolute value.

[0036] Preferably, the residual image comprises:

[0037]

[0038] Among them, R(i) is the pixel value corresponding to pixel i in the residual image, is the pixel value corresponding to pixel i in the spatial interpolation image, SD i Represents the spectral difference correction factor corresponding to pixel i.

[0039] Preferably, the high temporal and high spatial resolution image at time T1 includes:

[0040] F1(i)=F0(i)×STW i +R(i)

[0041] Among them, F1(i) represents the pixel value corresponding to pixel i in the high temporal and high spatial resolution image at time T1, STW i represents the global spatiotemporal weight corresponding to pixel i, R(i) is the pixel value corresponding to pixel i in the residual image, and F0(i) represents the pixel value of pixel i in the high spatial and low temporal resolution image at time T0.

[0042] The present invention has at least the following beneficial effects

[0043] The present invention divides the change area and the unchanged area by constructing an adaptive threshold factor, and uses dense optical flow to calculate the sum of pixel motion, thereby obtaining a global time weight matrix. At the same time, combined with the global spatial weight matrix, a global spatiotemporal weight matrix is ​​obtained. By performing spatial interpolation operations on low spatial and high temporal resolution images, and performing Gaussian low-pass filtering and correlation calculations on high spatial and low temporal resolution images, a spectral difference correction factor matrix is ​​obtained. Finally, the spectral difference correction factor matrix is ​​multiplied by the spatial interpolation image to obtain a residual image, and the high spatial and low temporal resolution image at time T0 is weighted using the global spatiotemporal weight matrix, and the weighted image and the residual image are added to obtain a high temporal and high spatial resolution image at time T1. This method can not only solve the problem of balancing time resolution and spatial resolution in traditional remote sensing data acquisition methods, but also comprehensively consider dynamic time information and spectral differences to improve fusion accuracy and efficiency. Therefore, this method has broad application prospects and important practical value in the fields of urban planning, agricultural production, environmental monitoring, etc. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a schematic diagram of the technical process of the present invention;

[0045] Figure 2 It is a schematic diagram for comparing the effects of the present invention and the existing model. DETAILED DESCRIPTION

[0046] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner, and the following embodiments and features in the embodiments can be combined with each other without conflict.

[0047] The present invention is mainly divided into two parts: introducing optical flow estimation in the global spatiotemporal weight to capture dynamic time change information; introducing spectral correction factors in residual calculation to reduce the impact of heterogeneous remote sensing spectral differences on fusion results. The detailed information is as follows:

[0048] 1. Calculate the time and space weights step by step and combine them with simple linear equations. When calculating the time weight, combine the optical flow information prediction method and adaptive threshold factor commonly used in video target tracking to make up for the defect that the differences between different image pixels cannot be captured dynamically and enhance the robustness of the model so that it can adapt to various scenarios; when calculating the space weight, use simple global parameters instead of sliding windows and multi-scale feature extraction to reduce the image pixel differences caused by large fluctuations in local information.

[0049] 2. The residual calculation is divided into two parts: spatial and spectral residuals. In the spatial residual calculation, cubic spline interpolation is used instead of thin plate spline interpolation to reduce the complexity of interpolation calculation and improve fusion efficiency; in the spectral residual calculation, a spectral difference correction factor is added to balance the spectral non-similarity of heterogeneous images caused by different sensors, and improve the color continuity of the image while smoothly fusing the spectral details of the image.

[0050] See also Figure 1 and Figure 2 , Embodiment, the present invention provides a remote sensing image spatiotemporal fusion method combining optical flow and spectral correction factor, comprising:

[0051] S1: Based on the vegetation index difference image calculated from the low spatial and high temporal resolution images at time T1 and time T0, an adaptive threshold factor is constructed to divide the vegetation index difference image into a change area and a non-change area;

[0052] S2: Based on the dense optical flow between the low spatial and high temporal resolution images from time T0 to time T1, the sum of the motions of all pixels on the optical flow motion vector is calculated;

[0053] S3: Calculate the temporal weights for the changed area and the unchanged area respectively according to the sum of the motions of all pixels on the optical flow motion vector, and obtain the global temporal weight matrix of all pixels;

[0054] S4: Divide each pixel value in the high spatial and low temporal resolution image at time T0 by the average value of all pixels to calculate the global spatial weight matrix of all pixels;

[0055] S5: Fusing the global time weight matrix and the global space weight matrix to obtain the global space-time weight matrix;

[0056] S6: Using cubic spline interpolation to perform spatial interpolation operation on the low spatial and high temporal resolution image at time T1 to obtain a spatial interpolation image;

[0057] S7: Perform Gaussian low-pass filtering on the high spatial and low temporal resolution image at time T0, and perform correlation calculation on the image after Gaussian low-pass filtering and the low spatial and high temporal resolution image at time T0 to obtain a spectral difference correction factor matrix;

[0058] S8: Multiply the spectral difference correction factor matrix and the spatial interpolation image to obtain the residual image, use the global spatiotemporal weight matrix to weight the high spatial and low temporal resolution image at time T0, and add the weighted image and the residual image to obtain the high temporal and high spatial resolution image at time T1.

[0059] Before executing the scheme of the present invention, the following steps are included:

[0060] Acquire multi-phase and multi-location Landsat (high spatial and low temporal) and MODIS (high temporal and low spatial) resolution images, and perform radiation correction, atmospheric correction, geometric correction and vegetation index band calculation preprocessing on them to obtain high temporal and low spatial resolution images at T0 and T1. The low temporal and high spatial resolution images at T0 are used as the input images and the low temporal and high spatial resolution reference images at T1.

[0061] Preferably, the step S1 specifically includes:

[0062] S11: Calculate the vegetation index images of low spatial and high temporal resolution images at time T1 and time T0, such as NDVI, Normalized Difference Vegetation Index;

[0063] For each moment of image, the Normalized Difference Vegetation Index (NDVI) formula is used to calculate the vegetation index image of the ground object. NDVI is usually calculated using the following formula:

[0064]

[0065] Among them, NIR represents the reflectivity of the near-infrared band, and RED represents the reflectivity of the red light band.

[0066] S12: Subtract the vegetation index image at time T0 from the vegetation index image at time T1 to obtain a vegetation index difference image; subtract the vegetation index image at time T1 from the vegetation index image at time T0 pixel by pixel to obtain a vegetation index difference image. This image reflects the change in vegetation coverage between the two times.

[0067] S13: Calculate the absolute value of each pixel in the vegetation index difference image and the absolute value of the average value of all pixels, and use the standard deviation calculated by subtracting the absolute value of the average value of all pixels from the absolute value of each pixel as the adaptive threshold factor;

[0068] Absolute value calculation: Calculate the absolute value of each pixel in the vegetation index difference image.

[0069] Mean and standard deviation calculation: Calculate the mean of the absolute values ​​of all pixels and calculate the standard deviation based on the mean. The standard deviation is used as an indicator to measure the fluctuation of pixel values ​​and is used to construct an adaptive threshold factor.

[0070] Adaptive threshold factor: The difference between the absolute value of each pixel and the absolute value of the average value of all pixels is divided by the standard deviation, and the result is used as the adaptive threshold factor. This factor can dynamically adjust the threshold according to the overall change of the image to adapt to different surface changes.

[0071] S14: using a decision tree to determine the size of the pixel value of each pixel in the vegetation index difference image and the adaptive threshold factor; when the pixel value of the pixel is less than the adaptive threshold factor, the pixel is divided into a non-changing area; otherwise, the pixel is divided into a changing area.

[0072] The decision tree algorithm is used to divide the image according to the relationship between the pixel value of each pixel in the vegetation index difference image and the adaptive threshold factor.

[0073] Region division: If the pixel value of a pixel is less than the adaptive threshold factor, the pixel is divided into a unchanged region; conversely, if the pixel value is greater than or equal to the adaptive threshold factor, the pixel is divided into a changed region.

[0074] Through the above steps, the present invention can obtain the division result of the changed area and the unchanged area based on the ground object vegetation index difference image. This result provides important input information for the subsequent spatiotemporal fusion process, helps to more accurately fuse remote sensing images of different resolutions, and improves the accuracy and reliability of the fusion result.

[0075] Preferably, the adaptive threshold factor comprises:

[0076]

[0077] Among them, |.| represents the absolute value, D represents the difference image of vegetation index, represents the average value of all pixels, std(.) represents the standard deviation calculation function, and V represents the adaptive threshold factor.

[0078] Preferably, calculating the time weights for the changed area and the unchanged area respectively comprises:

[0079] i belongs to the changing area

[0080] i belongs to the no-change region

[0081] in, and The temporal weights of the pixels belonging to the changed area and the unchanged area, represents the average pixel value of the high spatial and low temporal resolution image in the changing area at time T0, i represents the i-th pixel, and D i represents the pixel value of pixel i in the vegetation index difference image, F represents the sum of the motion of all pixels on the optical flow motion vector, N represents the number of pixels, It represents the average pixel value of the high spatial and low temporal resolution image in the changing area at time T0.

[0082] Preferably, the dense optical flow between the low spatial and high temporal resolution images from time T0 to time T1 is calculated by using the Python calcOpticalFlowFarneback method.

[0083] Preferably, the global spatial weight matrix comprises:

[0084]

[0085] Among them, F0(i) represents the pixel value of pixel i in the high spatial and low temporal resolution image at time T0, represents the pixel average of the high spatial and low temporal resolution image at time T0, SW i Represents the global spatial weight corresponding to pixel i.

[0086] Preferably, the global spatiotemporal weight matrix comprises:

[0087] STW i =SW i +TW i

[0088] Among them, STW i represents the global spatiotemporal weight corresponding to pixel i, TW i represents the global temporal weight corresponding to pixel i; SW i Represents the global spatial weight corresponding to pixel i.

[0089] Preferably, the spectrum difference correction factor matrix includes:

[0090]

[0091] Among them, SD i represents the spectral difference correction factor corresponding to pixel i, represents the pixel value corresponding to pixel i in the high spatial and low temporal resolution image at time T0 after Gaussian low-pass filtering, C0(i) represents the pixel value corresponding to pixel i in the high spatial and low temporal resolution image at time T0, and |.| represents the absolute value.

[0092] Preferably, the residual image comprises:

[0093]

[0094] Among them, R(i) is the pixel value corresponding to pixel i in the residual image, is the pixel value corresponding to pixel i in the spatial interpolation image, SD i Represents the spectral difference correction factor corresponding to pixel i.

[0095] Preferably, the high temporal and high spatial resolution image at time T1 includes:

[0096] F1(i)=F0(i)×STW i +R(i)

[0097] Among them, F1(i) represents the pixel value corresponding to pixel i in the high temporal and high spatial resolution image at time T1, STW i represents the global spatiotemporal weight corresponding to pixel i, R(i) is the pixel value corresponding to pixel i in the residual image, and F0(i) represents the pixel value of pixel i in the high spatial and low temporal resolution image at time T0.

[0098] Experimental simulation:

[0099] See also Figure 2 , Figure 2 This is a model comparison effect diagram of the present invention. Figure 2 (a) to (d) are the predicted images of STARFM, FSDAF, Fit-FC and the present spatiotemporal fusion model (i.e., the predicted high temporal and high spatial resolution images at time T1), and (e) is the reference image. Figure 2 It can be seen from the figure that the predicted image of the present invention retains more continuous spatial detail information and fewer blurred pixels.

[0100] See Table 1:

[0101] Table 1 Comparison indicators of STARFM, FSDAF, Fit-FC and this model, where RMSE is the root mean square error, SSIM is the structural similarity index and PSNR is the peak signal-to-noise ratio

[0102]

[0103] It can be concluded from Table 1 that, compared with STARFM, FSDAF and Fit-FC, the present invention achieves higher correlation coefficient, structural similarity index and peak signal-to-noise ratio, and lower root mean square error.

[0104] In summary, the present invention divides the change area and the unchanged area by constructing an adaptive threshold factor, and uses dense optical flow to calculate the sum of pixel motion, thereby obtaining a global time weight matrix. At the same time, combined with the global spatial weight matrix, a global spatiotemporal weight matrix is ​​obtained. By performing spatial interpolation operations on low spatial and high temporal resolution images, and performing Gaussian low-pass filtering and correlation calculations on high spatial and low temporal resolution images, a spectral difference correction factor matrix is ​​obtained. Finally, the spectral difference correction factor matrix is ​​multiplied by the spatial interpolation image to obtain a residual image, and is added to the global spatiotemporal weight matrix to obtain a remote sensing image with high temporal and high spatial resolution. This method can not only solve the problem of balancing time resolution and spatial resolution in traditional remote sensing data acquisition methods, but also comprehensively consider dynamic time information and spectral differences to improve fusion accuracy and efficiency. Therefore, this method has broad application prospects and important practical value in the fields of urban planning, agricultural production, environmental monitoring, etc.

[0105] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution, which should be included in the scope of the claims of the present invention.

Claims

1. A remote sensing image spatiotemporal fusion method combining optical flow and spectral correction factor, characterized in that: include: S1: Based on the vegetation index difference image calculated from the low spatial and high temporal resolution images at time T1 and time T0, an adaptive threshold factor is constructed to divide the vegetation index difference image into a change area and a non-change area; S2: Based on the dense optical flow between the low spatial and high temporal resolution images from time T0 to time T1, the sum of the motions of all pixels on the optical flow motion vector is calculated; S3: Calculate the temporal weights for the changed area and the unchanged area respectively according to the sum of the motions of all pixels on the optical flow motion vector, and obtain the global temporal weight matrix of all pixels; S4: Divide each pixel value in the high spatial and low temporal resolution image at time T0 by the average value of all pixels to calculate the global spatial weight matrix of all pixels; S5: Fusing the global time weight matrix and the global space weight matrix to obtain the global space-time weight matrix; S6: Using cubic spline interpolation to perform spatial interpolation operation on the low spatial and high temporal resolution image at time T1 to obtain a spatial interpolation image; S7: Perform Gaussian low-pass filtering on the high spatial and low temporal resolution image at time T0, and perform correlation calculation on the image after Gaussian low-pass filtering and the low spatial and high temporal resolution image at time T0 to obtain a spectral difference correction factor matrix; S8: Multiply the spectral difference correction factor matrix and the spatial interpolation image to obtain the residual image, use the global spatiotemporal weight matrix to weight the high spatial and low temporal resolution image at time T0, and add the weighted image and the residual image to obtain the high temporal and high spatial resolution image at time T1.

2. The remote sensing image spatiotemporal fusion method combining optical flow and spectral correction factor according to claim 1, characterized in that: The step S1 specifically includes: S11: Calculate the vegetation index images of low spatial and high temporal resolution images at time T1 and time T0 respectively; S12: subtracting the vegetation index image at time T0 from the vegetation index image at time T1 to obtain a vegetation index difference image; S13: Calculate the absolute value of each pixel in the vegetation index difference image and the absolute value of the average value of all pixels, and use the standard deviation calculated by subtracting the absolute value of the average value of all pixels from the absolute value of each pixel as the adaptive threshold factor; S14: using a decision tree to determine the size of the pixel value of each pixel in the vegetation index difference image and the adaptive threshold factor; when the pixel value of the pixel is less than the adaptive threshold factor, the pixel is divided into a non-changing area; otherwise, the pixel is divided into a changing area.

3. The remote sensing image spatiotemporal fusion method combining optical flow and spectral correction factor according to claim 2 is characterized in that: The adaptive threshold factors include: Among them, |.| represents the absolute value, D represents the difference image of vegetation index, represents the average value of all pixels, std(.) represents the standard deviation calculation function, and V represents the adaptive threshold factor.

4. The remote sensing image spatiotemporal fusion method combining optical flow and spectral correction factor according to claim 2, characterized in that: The calculating of the time weights for the changed area and the unchanged area respectively comprises: i belongs to the changing area i belongs to the no-change region in, and The temporal weights of the pixels belonging to the changed area and the unchanged area, represents the average pixel value of the high spatial and low temporal resolution image in the changing area at time T0, i represents the i-th pixel, and D i represents the pixel value of pixel i in the vegetation index difference image, F represents the sum of the motion of all pixels on the optical flow motion vector, N represents the number of pixels, It represents the average pixel value of the high spatial and low temporal resolution image in the changing area at time T0.

5. The remote sensing image spatiotemporal fusion method combining optical flow and spectral correction factor according to claim 1, characterized in that: The dense optical flow between the low spatial and high temporal resolution images from time T0 to time T1 is calculated by the Python calcOpticalFlowFarneback method.

6. The remote sensing image spatiotemporal fusion method combining optical flow and spectral correction factor according to claim 1, characterized in that: The global spatial weight matrix includes: Among them, F0(i) represents the pixel value of pixel i in the high spatial and low temporal resolution image at time T0, represents the pixel average of the high spatial and low temporal resolution image at time T0, SW i Represents the global spatial weight corresponding to pixel i.

7. The remote sensing image spatiotemporal fusion method combining optical flow and spectral correction factor according to claim 1, characterized in that: The global spatiotemporal weight matrix includes: STW i =SW i +TW i Among them, STW i represents the global spatiotemporal weight corresponding to pixel i, TW i represents the global temporal weight corresponding to pixel i; SW i Represents the global spatial weight corresponding to pixel i.

8. The remote sensing image spatiotemporal fusion method combining optical flow and spectral correction factor according to claim 1, characterized in that: The spectrum difference correction factor matrix includes: Among them, SD i represents the spectral difference correction factor corresponding to pixel i, represents the pixel value corresponding to pixel i in the high spatial and low temporal resolution image at time T0 after Gaussian low-pass filtering, C0(i) represents the pixel value corresponding to pixel i in the high spatial and low temporal resolution image at time T0, and |.| represents the absolute value.

9. The remote sensing image spatiotemporal fusion method combining optical flow and spectral correction factor according to claim 1, characterized in that: The residual image includes: Among them, R(i) is the pixel value corresponding to pixel i in the residual image, is the pixel value corresponding to pixel i in the spatial interpolation image, SD i Represents the spectral difference correction factor corresponding to pixel i.

10. The remote sensing image spatiotemporal fusion method combining optical flow and spectral correction factor according to claim 1, characterized in that: The high temporal and high spatial resolution images at time T1 include: F1(i)=F0(i)×STW i +R(i) Among them, F1(i) represents the pixel value corresponding to pixel i in the high temporal and high spatial resolution image at time T1, STW i represents the global spatiotemporal weight corresponding to pixel i, R(i) is the pixel value corresponding to pixel i in the residual image, and F0(i) represents the pixel value of pixel i in the high spatial and low temporal resolution image at time T0.

Images