A method and apparatus for dual-window simulation reconstruction of shortwave infrared remote sensing images
By using a dual-window simulation reconstruction method for shortwave infrared remote sensing images, the problem of high spectral resolution and high spatial resolution images being difficult to coexist is solved, achieving simple and efficient image reconstruction while maintaining the spatial resolution and spectral characteristics of the images.
Patent Information
- Application Number
- CN202211520657.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-11-29
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2042-11-29
AI Technical Summary
In existing technologies, it is difficult for high spectral resolution and high spatial resolution remote sensing images to coexist. Existing image simulation methods are complex and computationally inefficient, and they have high requirements for the similarity of ground features in images with different spatial resolutions, resulting in spectral distortion.
A dual-window simulation reconstruction method for shortwave infrared remote sensing images is adopted. The size of the moving window is determined by calculating the image resolution ratio. The images are resampled and overlaid to reconstruct the images. The high spatial resolution SWIR band images are reconstructed using the optimal similar pixels and fitting functions.
It significantly improves the spatial resolution of shortwave infrared images, maintains the original spectral characteristics, simplifies the calculation process, and improves computational efficiency and accuracy.
Smart Images

Figure CN115830185B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of digital image processing technology in photogrammetry and remote sensing, and particularly to a dual-window simulation reconstruction method and apparatus for shortwave infrared remote sensing images. Background Technology
[0002] With the development of satellite remote sensing technology, a three-dimensional spatial observation system with multiple spatial scales, hyperspectral and multi-temporal resolution has been gradually improved and has been widely used in various fields closely related to people's lives, such as environmental monitoring, land surveys, and travel navigation.
[0003] However, due to limitations of remote sensors themselves, high spectral resolution and high spatial resolution are usually difficult to achieve simultaneously. Most high spatial resolution satellites, such as IKONOS, QuickBird, WorldView-1, and WorldView-2, are limited to the Visible Near Infrared (VNIR) band and lack the Shortwave Infrared (SWIR) band. Shortwave infrared imaging has advantages such as less atmospheric scattering, stronger penetration through smoke, fog, or haze, and longer effective detection distances, making it significantly more adaptable to climatic conditions and environments than visible light imaging. Addressing the lack of high spatial resolution shortwave infrared data has become a pressing research hotspot in the field of remote sensing applications. Image simulation is a technique that, based on existing remote sensing images, prior remote sensing knowledge, and remote sensing theoretical models, uses mathematical calculations or physical modeling under certain constraints to obtain simulated images. Image simulation technology can compensate for the lack of shortwave infrared imagery in high spatial resolution satellites, and has significant theoretical and practical implications.
[0004] Spatial-spectral image simulation is a widely used and relatively mature image simulation technique. Over many years of application, various data simulation methods have been developed. These include different images acquired by the same sensor using different imaging methods, image fusion from different sensors, and combining images of the same ground target acquired at different times. The goal is to achieve a simulated image that possesses both highly detailed spatial features and spectral characteristics of a multispectral image. Mainstream methods include high-pass filtering, Brovey transform, PCA substitution, IHS transform, and wavelet transform, with numerous improved methods based on these principles.
[0005] In the prior art, invention patent CN106384340A discloses a method for image fusion of remote sensing images with different spatial resolutions from different periods. This method analyzes early and later low-spatial-resolution images to obtain change data, thereby using early high-spatial-resolution images to obtain later high-spatial-resolution images. The construction mode of the solution model is evolved from a planar window mode to a linear spiral mode, allowing for the targeted construction of solution pixel sets and increasing the upper limit of the number of mixed pixel components that the downscaling fusion method can solve.
[0006] Patent application CN112102218A discloses a fusion method for generating high spatial resolution multispectral images. This method fuses high spatial resolution images with fewer spectral bands and low spatial resolution images with more spectral bands to generate a high spatial resolution image with multiple spectral bands. The method classifies high spatial resolution pixels, calculates and identifies a c-th class of clean low spatial resolution pixels, and uses matrix singular value decomposition to obtain the spectral basis vectors corresponding to the c-th class of clean low spatial resolution pixels. By moving information from neighboring similar pixels within a window and utilizing residual compensation, the method eliminates patchiness in the fusion result and improves fusion accuracy.
[0007] However, the above method is relatively complex, the computational efficiency is still not very high, and it requires a high degree of similarity between the two remote sensing images used for calculation. When the temporal difference between the two images with different spatial resolutions is large or the difference in land surface type is large, it is easy to produce large spectral distortion. Summary of the Invention
[0008] The technical problem to be solved by the present invention is to provide a simpler, less time-consuming, and more accurate method and apparatus for simulating and reconstructing shortwave infrared remote sensing images.
[0009] The technical solution adopted by this invention to solve its technical problem is:
[0010] A dual-window simulation reconstruction method for shortwave infrared remote sensing images includes the following steps:
[0011] Step 1: Input geometrically corrected remote sensing images A and B at different spatial resolutions;
[0012] Step 2: Calculate the ratio R of the spatial resolution of remote sensing images B and A, and determine the size N×N of the moving window;
[0013] Step 3: Resample the low spatial resolution image B to N times the resolution of the high resolution remote sensing image A to generate a new image C1;
[0014] Step 4: Using the spatial resolution of image A as a reference, overlay and reassemble images A and C1 to generate a new image D;
[0015] Step 5: Perform window neighborhood operation on the low spatial resolution remote sensing image C1 in image D using an N×N moving window to obtain image C2 in order to achieve spatial scale restoration.
[0016] Step 6: Determine the dominant ground feature pixels in the N×N window of image D, calculate the synthetic spectral vector Y, and scale it up to match that of image C2.
[0017] Step 7: Determine the optimal similar pixel in the N×N window of the low-resolution image to obtain the optimal spectral vector Xmax;
[0018] Step 8: Reconstruct the SWIR band spectrum with high spatial resolution using the optimal spectral vector Xmax and the fitting function;
[0019] Step 9: Simulate and reconstruct to generate a high spatial resolution SWIR band image E.
[0020] Preferably, the remote sensing images A and B in step 1 are geometrically corrected high-resolution remote sensing images A and B, which contain visible-near-infrared (VNIR) bands and shortwave infrared bands, respectively.
[0021] Low spatial resolution remote sensing image B with (VNIR+SWIR)m2 bands.
[0022] Preferably, step 2 involves calculating the ratio R between the spatial resolution of the visible-near-infrared and short-wave infrared (VNIR+SWIR) band remote sensing image B and the spatial resolution of the visible-near-infrared band remote sensing image A, and determining the size of the moving window as N×N based on the range of values for R.
[0023] Preferably, the method for generating the new image C1 in step 3 is as follows: all m2 visible-near infrared and shortwave infrared (VNIR+SWIR) bands of the low spatial resolution image B are resampled to N times the size of the high-resolution remote sensing image A to generate a new m2 image C1 containing the visible-near infrared and shortwave infrared (VNIR+SWIR) bands.
[0024] Preferably, the method for generating image D in step 4 is as follows: based on the spatial resolution of image A, the two sets of remote sensing images are resampled and overlaid in the order of images A and C1 to generate a new image D with m1+m2 bands; the image size of the low spatial resolution visible-near infrared and shortwave infrared (VNIR+SWIR) band image C1 with m2 bands is set to W1×W2 pixels, and the image size of the overlaid and reassembled image D with m1+m2 bands is set to (N×N)×(W1×W2) pixels.
[0025] As a preferred embodiment, the specific method for achieving spatial scale restoration in step 5 is as follows: using an N×N moving window, perform window neighborhood operations on the m2 visible-near-infrared and shortwave infrared (VNIR+SWIR) band images corresponding to the low spatial resolution remote sensing image C1 in image D to achieve spatial scale restoration.
[0026] Preferably, step 6 specifically involves taking an N×N window centered on each pixel of image D, calculating the mean (MEAN) and standard deviation (STD) of the grayscale value (DN) of the high spatial resolution red (RED) band image, and statistically analyzing these values. Pixels whose grayscale value (DN) falls within the interval (MEAN-0.5×STD, MEAN+0.5×STD) are identified as dominant ground feature pixels and marked accordingly. The average value of the VNIR spectrum of each dominant ground feature pixel within the window is calculated and assigned to the center pixel of the window, resulting in the composite spectral vector Y(1,2,…,m1) of the N×N window bands 1 to m1, thereby increasing the spatial scale of the composite spectrum Y to the same spatial scale as image C2.
[0027] Preferably, step 7 specifically involves: synthesizing a spectral vector Y(1,2,…,m1) from an N×N window of the high spatial resolution image D corresponding to a band 1 to m1 of the high spatial resolution image A, and corresponding to a pixel of the low spatial resolution image C2; taking another N×N window in the low spatial resolution image C2 to obtain N×N spectral vectors X(1,2,…,m2) of the visible-near infrared and short-wave infrared (VNIR+SWIR) bands; fitting the values X(1,2,…,m1) of the first m1 bands of the spectral vector X(1,2,…,m2) with Y(1,2,…,m1) using the least squares method, and identifying the pixel of image C2 corresponding to the maximum correlation coefficient as the optimal similar pixel; the visible-near infrared and short-wave infrared (VNIR+SWIR) band spectrum of the image C2 corresponding to the optimal similar pixel is the optimal spectral vector Xmax(1,2,…,m2).
[0028] Preferably, step 8 specifically involves: processing the optimal spectral vector obtained in the above steps.
[0029] Xmax(1,2,…,m2) takes the values X(1,2,…,m1) of its first m1 bands and combines them with the synthesized spectral vector Y.
[0030] The least squares method is used to fit the spectrum of (1,2,…,m1) to obtain the values of parameters a and b in the fitting function y=a×x+b. Then, the spectral values of the shortwave infrared (SWIR) band are obtained using the optimal spectral vector Xmax(1,2,…,m2).
[0031] Substituting Xmax(m1+1,m2) into the fitting function yields Y(m1+1,m2), which determines the reconstructed spectra of the m2-m1 SWIR bands with high spatial resolution. Step 9 specifically involves performing steps 6 to 8 for each pixel of the high spatial resolution image D, thereby simulating and reconstructing the m2-m1 shortwave infrared (SWIR) band images E with high spatial resolution.
[0032] An apparatus for implementing the dual-window simulation reconstruction method for shortwave infrared remote sensing images as described in claim 1, comprising a remote sensing image input module that is communicatively connected to each other, for inputting geometrically corrected remote sensing images A and B of different spatial resolutions;
[0033] The module for determining the size of the moving window calculates the ratio R of the spatial resolution of remote sensing images B and A, and determines the size N×N of the moving window.
[0034] The spatial resolution adjustment module is used to resample the low spatial resolution image B to N times the resolution of the high-resolution remote sensing image A, generating a new image C1.
[0035] The image overlay and recombination module is used to overlay and recombine images A and C1 based on the spatial resolution of image A to generate a new image D.
[0036] The image spatial scale restoration module is used to perform window neighborhood operations on the low spatial resolution remote sensing image C1 in image D using an N×N moving window to obtain image C2 in order to achieve spatial scale restoration.
[0037] The image spatial scale upgrading module is used to determine the dominant ground feature pixels in the N×N window of image D, calculate the synthetic spectral vector Y, and upgrade its spatial scale to be consistent with that of image C2.
[0038] The optimal similar pixel calculation module is used to determine the optimal similar pixel in an N×N window of low-resolution image and obtain the optimal spectral vector Xmax.
[0039] The spectral simulation and reconstruction module is used to reconstruct high spatial resolution shortwave infrared (SWIR) band spectra using the optimal spectral vector Xmax and the fitting function.
[0040] A high spatial resolution SWIR band image construction module is used to simulate and reconstruct high spatial resolution shortwave infrared (SWIR) band image E.
[0041] The beneficial effects of the present invention are as follows: The shortwave infrared remote sensing image simulation reconstruction method based on dual window analysis and spatial scale transformation proposed in this invention can significantly improve the spatial resolution of shortwave infrared images while maintaining the original spectral characteristics well. It has high computational efficiency, is simpler, consumes less computation time, and provides more accurate calculation results. Attached Figure Description
[0042] Figure 1 This is a flowchart of the shortwave infrared remote sensing image simulation and reconstruction method based on dual window analysis and spatial scale transformation in Embodiment 11 of the present invention.
[0043] Figure 2 This is the high spatial resolution visible light image A in this embodiment of the invention;
[0044] Figure 3 This refers to the low spatial resolution shortwave infrared image B in Embodiment 1 of the present invention;
[0045] Figure 4 This is a schematic diagram of the dual-window analysis and spatial scale conversion in Embodiment 11 of the present invention;
[0046] Figure 5 The high spatial resolution SWIR band image E generated by simulation reconstruction in this embodiment of the invention;
[0047] Figure 6 This is a scatter plot of the predicted and actual spectral values of the SWIR band in Embodiment 1 of the present invention.
[0048] Figure 7 This is a schematic diagram of the device structure for implementing the shortwave infrared remote sensing image simulation reconstruction method based on dual-window analysis and spatial scale transformation in Embodiment 1 of the present invention. Detailed Implementation
[0049] The present invention will be further described below with reference to the accompanying drawings and embodiments, but these specific embodiments do not limit the scope of protection of the present invention in any way.
[0050] Example 1
[0051] The present invention discloses a dual-window simulation reconstruction method for shortwave infrared remote sensing images, comprising the following steps:
[0052] refer to Figure 1 Step 1: Input remote sensing image;
[0053] Input images A, consisting of four VNIR bands from the QuickBird satellite with a geometrically corrected spatial resolution of 2.4 meters, and B, consisting of six VNIR+SWIR bands from the Landsat-5 TM satellite with a spatial resolution of 30 meters.
[0054] The QuickBird satellite is a commercial satellite capable of providing sub-meter resolution imagery, offering panchromatic (0.61-meter resolution) and multispectral (2.4-meter resolution) imagery in four multispectral bands. The Landsat-5 satellite carries the ThematicMapper™, which has four VNIR bands and two SWIR bands with a spatial resolution of 30 meters, as well as a thermal infrared band with a spatial resolution of 120 meters.
[0055] In this embodiment, the remote sensing data used are QuickBird imagery from September 14, 2011, and Landsat 5-TM imagery from October 16, 2010, for a region in Sudan, Africa (center coordinates E 36°3′, N21°9′). The QuickBird imagery was cut into 1800×1800 pixels, and the corresponding spatial range of the Landsat 5-TM imagery was cut into 144×144 pixels to serve as experimental data for this embodiment of the invention.
[0056] Specifically, such as Figure 2 The image shown is a high spatial resolution QuickBird image A in this embodiment.
[0057] Step 2: Determine the size of the movable window, N×N;
[0058] The spatial resolution of Landsat 5–TM imagery in the visible-near-infrared and short-wave infrared bands is 30 meters, while the spatial resolution of QuickBird imagery in the visible-near-infrared bands is 2.4 meters. The ratio R between the two is 12.5, which is within the range where the ratio R is greater than 10. Therefore, the size of the moving window is determined to be 9×9.
[0059] Step 3: Adjust the spatial resolution of the remote sensing image in B space;
[0060] The spatial resolution of QuickBird's VNIR band remote sensing image A is 2.4 meters, which multiplied by 9 is 21.6 meters. By resampling, the spatial resolution of six TM VNIR+SWIR band remote sensing images B (with a spatial resolution of 30 meters) is adjusted to 21.6 meters. Therefore, the newly generated image C1, corresponding to the area of the 1800×1800 pixel QuickBird image, has a size of 200×200 pixels. In other words, the spatial range of one pixel in TM image C1 corresponds exactly to 81 pixels in QuickBird image A.
[0061] Specifically, such as Figure 3 The image shown is a low spatial resolution Landsat 5-TM image C1 in an embodiment of the present invention.
[0062] Step 4: Overlay and reassemble images A and C1;
[0063] Based on the spatial resolution of 2.4 meters of QuickBird image A, the two sets of remote sensing images were resampled and overlaid in the order of images A and C1 to generate a new image D with 10 (i.e. 4+6) bands.
[0064] Therefore, in step 3, the image size of the six VNIR+SWIR band images C1 of the low spatial resolution TM is 200×200 pixels, and the image size of the superimposed and recombined image D with 10 bands is (9×9)×(200×200) pixels.
[0065] Step 5: Spatial scale restoration of VNIR+SWIR band images;
[0066] Spatial scale restoration is achieved by performing window neighborhood operations on the six VNIR+SWIR band images corresponding to the low spatial resolution TM image C1 in image D using a 9×9 moving window.
[0067] The specific operational rules are as follows: Image D is divided into 200×200 windows of size 9×9. For each band in the 6-band TM image, the spectral average value of the pixels in each 9×9 window is calculated sequentially and assigned to the center pixel of the window, generating a new image C2 with a size of 200×200 and 6 VNIR+SWIR bands. This restores the spatial scale of the 6 VNIR+SWIR bands in image D to the scale level of image C1. This ensures a strict spatial correspondence between image C2 and image D, meaning that a pixel in image C2 corresponds to 9×9 pixels in image D, creating conditions for subsequent dual-window analysis.
[0068] Step 6: Spatial scale enhancement of high-resolution VNIR band images;
[0069] For each pixel in image D, a 9×9 window is taken centered on the pixel. The mean (MEAN) and standard deviation (STD) of the gray value (DN) of the high spatial resolution QuickBird red (RED) band image are calculated. Pixels with DN values in the range of (MEAN-0.5×STD, MEAN+0.5×STD) are identified as dominant ground feature pixels and marked accordingly. The average value of the VNIR spectrum of each dominant ground feature pixel within the window is calculated and assigned to the center pixel of the window, resulting in the composite spectral vector Y(1,2,…,4) of bands 1 to 4 of the 9×9 window. This increases the spatial scale of the composite spectrum Y to the same spatial scale as image C2.
[0070] Specifically, in this embodiment, the composite spectral vector of bands 1 to 4 of a certain 9×9 window QuickBird image is Y(350,584,470,498).
[0071] Specifically, such as Figure 4 The diagram shown is a schematic representation of dual-window analysis and spatial scale conversion in an embodiment of the present invention.
[0072] Step 7: Calculate the optimal similarity pixels for low-resolution VNIR+SWIR band images;
[0073] A 9×9 window of the high spatial resolution image D corresponds to a composite spectral vector Y(1,2,…,4) of bands 1–4 of the high spatial resolution QuickBird image, and also corresponds to a pixel of the low spatial resolution TM image C2. Another 9×9 window is taken in the low spatial resolution TM image C2 to obtain 9×9 spectral vectors X(1,2,…,6) of VNIR+SWIR bands. The values of the first 4 bands X(1,2,…,6) of the spectral vector X(1,2,…,6) are fitted with Y(1,2,…,4) using the least squares method. The pixel of image C2 corresponding to the maximum correlation coefficient is identified as the optimal similar pixel and marked. The VNIR+SWIR band spectrum of image C2 corresponding to the optimal similar pixel is the optimal spectral vector Xmax(1,2,…,6).
[0074] Specifically, in this implementation case, the composite spectral vector of QuickBird bands 1-4 in a 9×9 window is Y(584,450,498,470); a certain spectral vector of TM image bands 1, 2, 3, 4, 5, 7 (VNIR+SWIR bands) in the 9×9 window is X(85,42,64,52,120,72); among the pixels of 81 TM images in the 9×9 window, the largest correlation coefficient is 0.9821. The pixel corresponding to the largest correlation coefficient is identified and the optimal similar pixel is marked, thus obtaining the corresponding optimal spectral vector Xmax(87,47,62,58,123,69) for TM image bands 1, 2, 3, 4, 5, 7.
[0075] Step 8: Spectral reconstruction of high-resolution SWIR band image;
[0076] The optimal spectral vector Xmax(1,2,…,6) obtained in the above steps is used to take the values of its first four bands, X(1,2,…,4), and is then fitted with the synthetic spectral vector Y(1,2,…,4) using the least squares method to obtain the values of parameters a and b in the fitting function y=a×x+b. Then, the spectral values Xmax(5,6) of the SWIR band of the optimal spectral vector Xmax(1,2,…,6) are substituted into the fitting function to obtain Y(5,6), which is the reconstructed spectrum of the two SWIR bands with high spatial resolution. That is, the spectra of the two corresponding SWIR bands of the QuickBird image are reconstructed by simulating the spectra of bands 5 and 7 of the TM image.
[0077] Specifically, in this embodiment, the optimal spectral vector of a certain 9×9 window TM
[0078] The spectral vector Xmax(87,47,62,58), obtained by taking the values of its four bands, is Xmax(87,47,62,58). This vector is then combined with the composite spectral vector Y(584,450,498,470) of a 9×9 window QuickBird bands 1-4. A least-squares method is used to fit the spectrum, yielding the fitting function y = 3.4621×x + 280.66. Substituting the optimal spectral vector Xmax(123,69) of the two SWIR bands from the TM image into the fitting function, the reconstructed spectral values of the two SWIR bands of the QuickBird image can be obtained.
[0079] Step 9: Reconstruct and generate high-resolution SWIR band images;
[0080] For each pixel of the high spatial resolution image D, and by performing steps 6 to 8, high spatial resolution m2-m1 SWIR band images E can be simulated and reconstructed.
[0081] Specifically, in this implementation case, two new high spatial resolution SWIR band images E corresponding to TM image bands 5 and 7 are obtained through reconstruction, such as... Figure 5 As shown, Figure 5 The image E is a high spatial resolution SWIR band image generated by simulation reconstruction in an embodiment of the present invention.
[0082] To conduct a comparative study, the results of three methods were compared and analyzed: the method of this invention, the method based on hyperspectral super-resolution (Hy Sure), and the method for improving the accuracy of multispectral image fusion by residual compensation disclosed in invention patent application document CN112102218A (hereinafter referred to as the residual compensation method).
[0083] Reference 1 (Simoes M, Bioucas-Dias J, Almeida L B & Chanussot Jocelyn. A convexformation for hyperspectral image superresolution via subspace-based regularization[J]. IEEE Transactions on Geoscience and Remote Sensing, 2015, 53(6): 3373-3388) gives the principle and calculation process of the hyperspectral superresolution method (Hy Sure).
[0084] Specifically, such as Figure 6 As shown, Figure 6 This is a scatter plot of the predicted and actual spectral values for the SWIR band in an embodiment of the present invention.
[0085] In this embodiment, the computation time for the proposed method, the HySure method, and the residual compensation method are 3.5 seconds, 7.8 seconds, and 4.6 seconds, respectively, demonstrating the significant advantage of the proposed method in terms of computational efficiency. The predicted images from the three methods show a high degree of consistency between the shape and spectral values of ground features and the original M5 and TM7 band images, while also preserving spatial details well and significantly improving spatial resolution. The grayscale images from the calculation results of the three methods are all quantized using 8 bits (0-255), and the correlation coefficients between the reconstructed SWIR band spectra and the TM5 and TM7 band spectral values are statistically analyzed.
[0086] Figure 6 Scatter plots are provided showing the actual values of the TM5 and TM7 band spectra and the predicted values of the corresponding SWIR band spectra for the three methods. Figure 6In the diagram, a, b, and c represent scatter plots corresponding to the prediction results of the method of this invention, the Hy Sure method, and the residual compensation method, respectively. The correlation coefficients between the predicted and actual values of the three methods for SWIR1 and SWIR2 bands are 0.9456 and 0.9502, 0.9263 and 0.9189, and 0.9237 and 0.9205, respectively. It can be seen that the correlation coefficients obtained by the method of this invention are all above 0.94, which is better than the values of around 0.92 for the other two methods. The closer the scatter points in the scatter plot are to the 1:1 baseline, the better the consistency between the fusion result and the actual result. Therefore, the scatter points in the scatter plot of the predicted results of the method described in this embodiment are closer to the 1:1 baseline and are very concentrated. In contrast, the prediction results of the Hy Sure method and the residual compensation method are more dispersed and deviate further from the 1:1 baseline. Visually, the Hy Sure method and the residual compensation method exhibit some color distortion in the predicted images. The Hy Sure method involves complex parameter optimization, posing a significant challenge for users. Furthermore, Hy Sure is computationally inefficient, with extremely high computation time consumption when dealing with large-scale images. Residual compensation methods, when faced with significant temporal differences or substantial differences in ground cover types between two image types, may also fail to construct a spectral fitting function, leading to local pixel spectral reconstruction failures.
[0087] In summary, visual observation and correlation coefficients and scatter plots between the predicted and actual spectra demonstrate that the shortwave infrared remote sensing image simulation and reconstruction method based on dual-window analysis and spatial scale transformation proposed in this invention can significantly improve the spatial resolution of shortwave infrared images while maintaining the original spectral characteristics well, and also has high computational efficiency.
[0088] On the other hand, embodiments of the present invention also provide an apparatus for implementing a shortwave infrared remote sensing image simulation reconstruction method based on dual-window analysis and spatial scale transformation.
[0089] refer to Figure 7 This is a schematic diagram of the implementation device for the shortwave infrared remote sensing image simulation reconstruction method based on dual-window analysis and spatial scale transformation, according to an embodiment of the present invention. The implementation device is interconnected and includes:
[0090] The remote sensing image input module 101 inputs a geometrically corrected high spatial resolution visible-near infrared image A and a low spatial resolution shortwave infrared (SWIR) image B.
[0091] The moving window size determination module 102 calculates the ratio R of the spatial resolution of remote sensing images B and A, and determines the size N×N of the moving window.
[0092] The spatial resolution adjustment module 103 resamples the low spatial resolution image B to N times the resolution of the high-resolution remote sensing image A, generating a new image C1.
[0093] The image overlay and recombination module 104 uses the spatial resolution of image A as a reference to overlay and recombine images A and C1 to generate a new image D.
[0094] The image spatial scale restoration module 105 performs window neighborhood calculations on the low spatial resolution remote sensing image C1 in image D using an N×N moving window to obtain image C2, thereby achieving spatial scale restoration.
[0095] The image spatial scale upscaling module 106 determines the dominant ground feature pixels in the N×N window of image D, calculates the synthetic spectral vector Y, and upscales its spatial scale to match that of image C2.
[0096] The optimal similar pixel calculation module 107 determines the optimal similar pixels in an N×N window of the low-resolution image and obtains the optimal spectral vector X. max .
[0097] Spectral simulation and reconstruction module 108 utilizes the optimal spectral vector X max By fitting the function y = a × x + b, a high spatial resolution SWIR band spectrum is reconstructed.
[0098] The high spatial resolution SWIR band image construction module 109 repeatedly executes steps 6 to 8 to simulate and reconstruct the high spatial resolution SWIR band image E.
[0099] The above embodiment provides an implementation apparatus for a shortwave infrared remote sensing image simulation and reconstruction method based on dual-window analysis and spatial scale transformation. This apparatus is used to implement the corresponding shortwave infrared remote sensing image simulation and reconstruction method based on dual-window analysis and spatial scale transformation in the foregoing embodiment, and has the beneficial effects of the corresponding method embodiment, which will not be elaborated here.
[0100] The foregoing has shown and described the basic principles, main features, and advantages of this invention. Those skilled in the art should understand that this invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of this invention. Various changes and modifications can be made to this invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention.
Claims
1. A dual-window simulation reconstruction method for shortwave infrared remote sensing images, characterized in that, Includes the following steps: Step 1: Input geometrically corrected remote sensing images A and B with different spatial resolutions. The remote sensing image A is a high spatial resolution remote sensing image containing m1 bands of visible light to near infrared, and the remote sensing image B is a low spatial resolution remote sensing image containing m2 bands of visible light to near infrared and shortwave infrared. Step 2: Calculate the ratio R of the spatial resolution of remote sensing image B and remote sensing image A, and determine the size N×N of the moving window; Step 3: Resample the low spatial resolution image B to N times the resolution of the high spatial resolution remote sensing image A to generate a new image C1; Step 4: Using the spatial resolution of image A as a reference, overlay and reassemble images A and C1 to generate a new image D; Step 5: Perform window neighborhood operation on the low spatial resolution remote sensing image C1 in image D using an N×N moving window to obtain image C2, so as to achieve spatial scale restoration. Step 6: For each pixel in image D, take an N×N window centered on that pixel, calculate the mean (MEAN) and standard deviation (STD) of the gray values in the high spatial resolution red band image, identify the pixels with gray values in the range of (MEAN-0.5×STD, MEAN+0.5×STD) as dominant ground features and mark them, calculate the average value of the visible-near infrared spectrum of each dominant ground feature pixel in the window and assign it to the center pixel of the window, and obtain the composite spectral vector Y (1, 2, ..., m1) of the 1 to m1 bands, thereby increasing the spatial scale of the composite spectral vector Y to the same spatial scale as image C2; Step 7: Assign the composite spectral vector Y (1, 2, ..., m1) of bands 1 to m1 of high spatial resolution image A corresponding to an N×N window of high spatial resolution image D to a pixel of low spatial resolution image C2. In low spatial resolution image C2, take another N×N window to obtain N×N spectral vectors X (1, 2, ..., m2) of visible-near infrared and short-wave infrared bands. Perform least squares fitting on the first m1 bands X (1, 2, ..., m1) of each spectral vector X and the composite spectral vector Y (1, 2, ..., m1). Identify the pixel of image C2 corresponding to the maximum correlation coefficient as the optimal similar pixel and mark it. Then, determine the visible-near infrared + short-wave infrared spectrum of image C2 corresponding to the optimal similar pixel as the optimal spectral vector Xmax (1, 2, ..., m2). Step 8: Take the first m1 band values X(1,2,...,m1) of the optimal spectral vector Xmax(1,2,...,m2) obtained above, and perform least squares spectral fitting with the synthesized spectral vector Y(1,2,...,m1) to obtain the parameters a and b of the fitting function y=a×x+b. Then, substitute the spectral values Xmax(m1+1,m2) of the short-wave infrared band of the optimal spectral vector Xmax(1,2,...,m2) into the fitting function to obtain Y(m1+1,m2), and determine the reconstructed spectra of the m2-m1 short-wave infrared bands with high spatial resolution. Step 9: Perform steps 6 to 8 on each pixel of the high spatial resolution image D to simulate and reconstruct the generation of m2-m1 shortwave infrared band images E with high spatial resolution.
2. The dual-window simulation reconstruction method for shortwave infrared remote sensing images according to claim 1, characterized in that: Step 2 specifically involves: calculating the ratio R between the spatial resolution of the visible-near-infrared and short-wave infrared remote sensing image B and the spatial resolution of the visible-near-infrared remote sensing image A, and determining the size of the moving window as N×N based on the range of values for R.
3. The dual-window simulation reconstruction method for shortwave infrared remote sensing images according to claim 2, characterized in that, The method for generating the new image C1 in step 3 is as follows: all m2 visible-near infrared and shortwave infrared bands of the low spatial resolution image B are resampled to N times the size of the high-resolution remote sensing image A to generate a new m2 image C1 containing visible-near infrared and shortwave infrared bands.
4. The dual-window simulation reconstruction method for shortwave infrared remote sensing images according to claim 3, characterized in that, The image D generation method in step 4 is as follows: based on the spatial resolution of image A, the two sets of remote sensing images are resampled and overlaid in the order of images A and C1 to generate a new image D with m1+m2 bands; the image size of the low spatial resolution visible light-near infrared + shortwave infrared band image C1 with m2 bands is set to W1×W2 pixels, and the image size of the overlaid and reassembled image D with m1+m2 bands is set to (N×N)×(W1×W2) pixels.
5. The dual-window simulation reconstruction method for shortwave infrared remote sensing images according to claim 4, characterized in that: The specific method for achieving spatial scale restoration in step 5 is as follows: using an N×N moving window, perform window neighborhood operations on the m2 visible-near-infrared + shortwave infrared band images corresponding to the low spatial resolution remote sensing image C1 in image D to achieve spatial scale restoration.
6. An apparatus for implementing the dual-window simulation reconstruction method for shortwave infrared remote sensing images as described in claim 1, characterized in that, It includes a remote sensing image input module that is interconnected and used to input geometrically corrected remote sensing images A and B at different spatial resolutions; The module for determining the size of the moving window calculates the ratio R of the spatial resolution of remote sensing images B and A, and determines the size N×N of the moving window. The spatial resolution adjustment module is used to resample the low spatial resolution image B to N times the resolution of the high-resolution remote sensing image A to generate a new image C1. The image overlay and recombination module is used to overlay and recombine images A and C1 based on the spatial resolution of image A to generate a new image D. The image spatial scale restoration module is used to perform window neighborhood operations on the low spatial resolution remote sensing image C1 in image D using an N×N moving window to obtain image C2 in order to achieve spatial scale restoration. The image spatial scale upgrading module is used to determine the dominant ground feature pixels in the N×N window of image D, calculate the synthetic spectral vector Y, and upgrade its spatial scale to be consistent with that of image C2. The optimal similar pixel calculation module is used to determine the optimal similar pixel in an N×N window of low-resolution image and obtain the optimal spectral vector Xmax. The spectral simulation and reconstruction module is used to reconstruct high spatial resolution short-wave infrared spectra using the optimal spectral vector Xmax and the fitting function. A high spatial resolution shortwave infrared band image construction module is used to simulate and reconstruct high spatial resolution shortwave infrared band image E.
Citation Information
Patent Citations
Remote sensing space-time image downscaling fusion method
CN106384340A
Fusion method for generating high-spatial-resolution multispectral image
CN112102218A
