Image fusion method and apparatus for improving spatial resolution of shortwave infrared remote sensing images

Abstract

This invention discloses an image fusion method for improving the spatial resolution of shortwave infrared remote sensing images, comprising: step 1, inputting remote sensing images; step 2, adjusting the spatial resolution of remote sensing image B; step 3, principal component analysis of image A; step 4, overlaying and recombining images A, C, and D; step 5, setting a virtual moving window; step 6, determining dominant ground feature pixels; step 7, calculating the spectral composition of dominant ground feature pixels; step 8, spectral fitting; step 9, spectral reconstruction; and step 10, generating a high-resolution shortwave infrared image. This image fusion method for improving the spatial resolution of shortwave infrared remote sensing images has the advantages of high fidelity, simplicity, stability, reliability, and high computational efficiency. It utilizes principal component transformation to extract high-resolution spatial information from spectral images and spectral fitting to fully leverage the spectral information of multispectral and shortwave infrared images, thus improving the spatial resolution of shortwave infrared images while maintaining spectral characteristics.

Description

Technical Field

[0001] This invention relates to the field of remote sensing satellite image processing technology, and in particular to an image fusion method and apparatus for improving the spatial resolution of shortwave infrared remote sensing images. Background Technology

[0002] High spatial and hyperspectral resolution remote sensing imagery plays a crucial role in various aspects of Earth science, including resource surveys, ecosystem dynamic monitoring, and natural disaster assessment and early warning. However, due to technological and budgetary limitations, it is currently difficult for a single satellite sensor to acquire remote sensing images with both high spatial and hyperspectral resolution simultaneously. Therefore, image fusion technology is one of the effective ways to address the trade-off between spatial and spectral resolution.

[0003] WorldView-3 and WorldView-4 satellites are currently the most advanced high-resolution civilian remote sensing satellites. Building upon the eight visible-near-infrared (VNIR) multispectral bands of WorldView-2, they added eight shortwave infrared (SWIR) bands, significantly improving the ability to extract detailed ground feature information. However, the raw spatial resolution of WorldView shortwave infrared imagery is 3.75m, while the spatial resolution of commonly obtained shortwave infrared imagery in China is 7.5m, a significant difference from the 1.24m spatial resolution of multispectral imagery. This limits the application effectiveness of shortwave infrared imagery.

[0004] ASTER (Advanced Spaceborne Thermal Emission and Reflection Radiometer) remote sensing data has been widely used in the field of Earth environmental resource research. The ASTER sensor consists of three independent subsystems operating in the visible-near infrared (VNIR), shortwave infrared (SWIR), and thermal infrared (TIR) ​​bands, with corresponding spatial resolutions of 15m, 30m, and 90m, respectively. Utilizing fusion techniques to enhance the spatial resolution of ASTER's shortwave infrared (SWIR) fluctuations to 15m will significantly improve the application potential and effectiveness of ASTER remote sensing data.

[0005] Image fusion is an effective technique for improving the spatial resolution of images. The key to fusion lies in how to improve the spatial resolution of the fused image while preserving the original spectral characteristics as much as possible. Traditional image fusion mostly involves fusing panchromatic and multispectral images from the same satellite sensor, as their wavelength ranges are roughly the same. Early fusion methods often introduced significant spectral distortion while improving resolution. Currently, multiscale analysis is a commonly used method to reduce spectral distortion. By using wavelet transform, contourlet transform, and other multiscale analysis methods, high-frequency and low-frequency information in images can be separated. By fusing high- and low-frequency information separately, the impact on the overall spectrum can be reduced.

[0006] As described in the literature (Laben et al., Process for Enhancing the Spatial Resolution of Multispectral Imagery Using Pan-Sharpening, US Patent 6,011,875.), the principle and calculation process of the GS method and the principle and calculation process of the PC method described in the literature (Welch, R. and W. Ahlers, 1987. "Merging Multiresolution SPOT HRV and Landsat TM Data." Photogrammetric Engineering & Remote Sensing, 53(3), pp.301-303.) require that the wavelength ranges of the high spatial resolution panchromatic image and the low spatial resolution multispectral image participating in the fusion be basically consistent. The greater the difference in wavelength range, the more severe the color distortion of the fused image will be. Summary of the Invention

[0007] The technical problem to be solved by the present invention is to provide a simple, reliable, efficient, and effective image fusion method that improves the spatial resolution of shortwave infrared remote sensing images by combining principal component transformation and spectral fitting.

[0008] The technical solution adopted by this invention to solve its technical problem is:

[0009] An image fusion method for improving the spatial resolution of shortwave infrared remote sensing images includes the following steps:

[0010] Step 1: Input two different high spatial resolution remote sensing images of the same satellite with different sensor types, after geometric correction;

[0011] Step 2: Adjust the low spatial resolution image to an integer multiple of the high spatial resolution by resampling;

[0012] Step 3: Perform principal component analysis on the high spatial resolution image and take its first component as the spatial information factor of ground features.

[0013] Step 4: Based on the spatial resolution of image A, resample and overlay the three sets of remote sensing images in the order of images D, A, and C.

[0014] Step 5: Set a virtual moving window with a size of T×T pixels, and cover and apply it to all pixels;

[0015] Step 6: Determine and identify dominant objects within the virtual mobile window;

[0016] Step 7: Calculate the pixels of the spectral average value for the dominant features in the virtual moving window to obtain the synthesized reference spectrum;

[0017] Step 8: Calculate the spectral fitting function for each pixel using the least squares method;

[0018] Step 9: Using the fitting function and the reference spectrum in the shortwave infrared band, simulate and reconstruct the new SWIR spectrum;

[0019] Step 10: Construct new high-resolution shortwave infrared band images.

[0020] Preferably, the remote sensing images in step 1 are a high spatial resolution visible-near infrared image A and a low spatial resolution shortwave infrared image B from two different types of sensors on the same satellite, after geometric correction.

[0021] Preferably, step 2 includes the following steps: adjusting the spatial resolution of the shortwave infrared image B with m bands and low spatial resolution to an integer multiple T of the spatial resolution of the visible-near infrared image A by resampling, and generating a new shortwave infrared image C with m bands and low spatial resolution.

[0022] Preferably, the spatial information factor of ground features in step 3 is: after performing principal component analysis on all bands of the high spatial resolution image A with n bands, n components from PC1 to PCn are obtained, and the first component (PC1) is taken as the grayscale image D of the spatial information factor of ground features.

[0023] Preferably, in step 4, the three sets of remote sensing images are resampled and overlaid in the order of images D, A, and C based on the spatial resolution of image A, to generate a new image E. Image E contains 1+n+m bands, and the first component representing the spatial information of ground features, PC1, is the first band of image E.

[0024] Preferably, in step 5, the image size of the low spatial resolution shortwave infrared image C is set to W1×W2 pixels, and the image size of the image E generated after overlay and recombination is set to (T×T)×(W1×W2) pixels. A virtual moving window with a window size of T×T pixels is set, and the image E is divided into W1×W2 virtual windows.

[0025] Preferably, step 6 involves statistically analyzing the mean and standard deviation of the gray values ​​in the red band of the visible-near-infrared image within the T×T virtual moving window image E. Pixels with gray values ​​within the range of (MEAN-0.5×STD, MEAN+0.5×STD) are identified as dominant ground object pixels and marked accordingly.

[0026] Preferably, step 7 involves calculating the average gray values ​​of all pixels identified as dominant features within a T×T virtual moving window in image E, across 1+n+m bands, to obtain the reference spectral vector X(1, 2, ..., 1+n+m) of the T×T virtual moving window. Step 8 involves obtaining the spatial information of the ground features and the gray values ​​of the visible-near-infrared bands for each pixel within the T×T virtual moving window, resulting in the spectral vector Y(1, 2, ..., 1+n) of each pixel. Within the T×T virtual moving window, the spectral vector Y(1, 2, ..., 1+n) and the reference spectrum X1(1, 2, ..., 1+n) of each pixel are obtained respectively, and the least squares method is used for fitting to obtain the values ​​of parameters a and b of the fitting function y = a×x + b for each pixel.

[0027] Preferably, step 9 involves calculating the reconstructed spectral value Y(1, 2, ..., m) of the shortwave infrared band for each pixel within the T×T virtual moving window using the values ​​of parameters a and b in the obtained fitting function y = a×x + b, and the reference spectral vector X2(1, 2, ..., m) of the shortwave infrared band; step 10 involves performing steps 5 to 9 for each of the W1×W2 T×T virtual moving windows, thereby obtaining a high-resolution shortwave infrared band image F with m bands.

[0028] An apparatus for implementing an image fusion method to improve the spatial resolution of shortwave infrared remote sensing images, comprising interconnected components:

[0029] The image acquisition module is used to input geometrically corrected remote sensing images from the same satellite with two different high spatial resolutions and different sensor types.

[0030] The spatial resolution adjustment module adjusts low spatial resolution images to an integer multiple of high spatial resolution through resampling;

[0031] The principal component analysis module performs principal component analysis on high spatial resolution images and takes the first component as the spatial information factor of ground features.

[0032] The image overlay and reconstruction module is used to resample and overlay three sets of remote sensing images in the order of images D, A, and C, based on the spatial resolution of image A.

[0033] The virtual moving window setting module is used to set a virtual moving window with a window size of T×T pixels, and to cover and apply it to all pixels;

[0034] The dominant feature pixel determination module is used to determine and identify dominant feature pixels within a virtual mobile window;

[0035] The dominant feature pixel spectral synthesis calculation module is used to calculate the pixel value of the spectral average of dominant features in the virtual moving window to obtain the synthesized reference spectrum.

[0036] The spectral fitting module is used to calculate the spectral fitting function for each pixel using the least squares method.

[0037] The spectral reconstruction module is used to simulate and reconstruct new shortwave infrared spectra using a fitting function and a reference spectrum in the shortwave infrared band.

[0038] A high-resolution shortwave infrared image construction module for building new high-resolution shortwave infrared images.

[0039] The beneficial effects of this invention are:

[0040] The multispectral images and shortwave infrared images to be fused in this invention are from different sensors and have completely different wavelength ranges. The fusion problem has certain special characteristics. It is necessary to effectively extract the high-resolution spatial information of the multispectral images and minimize the impact on the spectral information of the shortwave infrared images during the fusion process.

[0041] To address the aforementioned problems, this invention proposes a fusion method combining principal component transformation and spectral fitting to improve the spatial resolution of shortwave infrared images. Principal component transformation is used to extract high-resolution spatial information from spectral images, while spectral fitting fully utilizes the spectral information from both multispectral and shortwave infrared images, thereby improving the spatial resolution of shortwave infrared images while preserving their spectral characteristics. Attached Figure Description

[0042] Figure 1 This is a flowchart of an image fusion method for improving the spatial resolution of shortwave infrared remote sensing images in an embodiment of the present invention;

[0043] Figure 2 This refers to the low spatial resolution shortwave infrared image C in this embodiment of the invention;

[0044] Figure 3 The grayscale image D, representing the spatial information factor of ground features in this embodiment of the invention;

[0045] Figure 4 This is a schematic diagram of the determination of dominant land features and the calculation of synthetic reference spectra within a moving window in an embodiment of the present invention;

[0046] Figure 5 This is a schematic diagram of the moving window analysis and spectral fitting reconstruction process in an embodiment of the present invention;

[0047] Figure 6 The fused high spatial resolution shortwave infrared image F in this embodiment of the invention;

[0048] Figure 7 This is a schematic diagram of the device structure for implementing the image fusion method for improving the spatial resolution of shortwave infrared remote sensing images in an embodiment 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] like Figure 1 As shown, the image fusion method for improving the spatial resolution of shortwave infrared remote sensing images according to the present invention is as follows:

[0052] Step 1: Input remote sensing imagery;

[0053] Input geometrically corrected high spatial resolution visible-near infrared image A and low spatial resolution shortwave infrared image B from two different types of sensors on the same satellite.

[0054] In this embodiment, the remote sensing images selected are 1.24-meter high spatial resolution visible-near infrared image A and 3.75-meter low spatial resolution shortwave infrared image B from the WorldView-3 high-resolution remote sensing satellite after systematic geometric correction.

[0055] Step 2: Adjust the spatial resolution of the remote sensing image in B space;

[0056] The spatial resolution of shortwave infrared image B with low spatial resolution in m bands is adjusted to an integer multiple T of the spatial resolution of visible-near infrared image A by resampling, thereby generating a new shortwave infrared image C with low spatial resolution in m bands.

[0057] In this embodiment, the spatial resolution of WorldView-3's 8-band low spatial resolution shortwave infrared image B is 3.75 meters, and the spatial resolution of the visible-near-infrared image A is 1.24 meters. The ratio of the two is 2.98, which rounds to 3. Therefore, the ratio T between the high and low resolution images is equal to 3. The shortwave infrared image B should be resampled to be 3 times that of the visible-near-infrared image A (i.e., 3.72 meters) to obtain a shortwave infrared image C with 8 bands (see...). Figure 2 As shown in the figure, this allows a pixel of image C to cover exactly 9 pixels of image A, creating conditions for subsequent virtual moving window analysis.

[0058] Step 3, principal component analysis of image A;

[0059] Principal component analysis was performed on all bands of the high spatial resolution VNIR image A with 8 bands to obtain 8 components from PC1 to PC8. The first component (PC1) was taken as the grayscale image D with spatial information factor of ground features.

[0060] Specifically, such as Figure 3 The image shown is a grayscale image D of the spatial information factor of ground features in an embodiment of the present invention.

[0061] Step 4: Overlay and reassemble images A, C, and D;

[0062] Based on the 1.24-meter spatial resolution of image A, the three sets of remote sensing images D, A, and C are resampled and overlaid to generate a new image E. Image E contains 17 bands, and the first component representing the spatial information of ground features, PC1, is the first band of image E.

[0063] Step 5, Virtual moving window settings;

[0064] The size of the visible-near-infrared band image A with a spatial resolution of 1.24 meters cut out in this embodiment of the invention is 600.

[0065] If the image size is 600×600 pixels, then the size of the overlaid image E is also 600×600 pixels, and the size of the corresponding shortwave infrared image C is 200×200 pixels. The number of pixels in image E is exactly 9 times the number of pixels in image C (i.e., 3×3).

[0066] Define a virtual moving window with a size of T×T pixels. Specifically, in this implementation, the virtual moving window size is 3×3, so image E can be divided into 200×200 virtual windows.

[0067] Subsequent calculations and remote sensing analysis were all performed within a 3×3 virtual moving window.

[0068] Step 6, determining the dominant object pixels;

[0069] The mean (MEAN) and standard deviation (STD) of the gray values ​​in the red band of the visible-near-infrared image in the 3×3 virtual moving window image E are statistically analyzed. Pixels whose gray values ​​fall within the range of (MEAN-0.5×STD, MEAN+0.5×STD) are identified as dominant ground object pixels and marked accordingly.

[0070] Specifically, in this embodiment, the grayscale values ​​of the red band corresponding to a certain 3×3 virtual moving window are: 709, 701, 709, 702, 712, 715, 709, 708, and 708, with a mean (MEAN) of 708.11 and a standard deviation (STD) of 4.37. Therefore, the grayscale value range of the dominant features is (705.92.710.29). The pixels with grayscale values ​​of 708 and 709 in the red band are the dominant features, totaling 5 dominant feature pixels.

[0071] For details, see Figure 4 , Figure 4 The diagram shown is a schematic diagram of the determination of dominant features and the calculation of synthetic spectra within a moving window in an embodiment of the present invention.

[0072] Step 7, Calculation of spectral synthesis of dominant features;

[0073] In this embodiment, within a 3×3 virtual moving window in image E, for all pixels identified as dominant features, the average gray value of the dominant feature pixels in each of the 17 bands is calculated to obtain the reference spectral vector X(1,2,…,17) of the 3×3 virtual moving window.

[0074] Specifically, the reference spectral vector X of a certain 3×3 virtual moving window =

[0075] (189,351,337,560,552,608,1983,3423,3146,2790,1597,1851,1730,1147,1122,1064,903).

[0076] Step 8, spectral fitting;

[0077] In this embodiment, for each pixel within a 3×3 virtual moving window, spatial information of other objects and grayscale values ​​in the visible-near-infrared band are obtained, resulting in the spectral vector Y(1,2,…,9) for each pixel.

[0078] Within a 3×3 virtual moving window, the spectral vector Y(1,2,…,9) and the reference spectrum X1(1,2,…,9) of each pixel are obtained respectively. The least squares method is used to fit the data, thereby obtaining the values ​​of parameters a and b of the fitting function y=a×x+b for each pixel.

[0079] In this implementation example, the baseline spectrum X1 of a certain 3×3 virtual moving window is (189,351,337,560, ...).

[0080] Given the spectral vector Y = (176,334,314,526,537,573,1856,3223,3078) of a pixel in a 3×3 virtual moving window, and fitting X1 and Y using the least squares method, we obtain the fitting function y = 0.9703×x - 10.142, with a correlation coefficient R0. 2 =0.9983.

[0081] For details, please refer to Figure 5 , Figure 5 This is a schematic diagram of the moving window analysis and spectral fitting reconstruction process in an embodiment of the present invention.

[0082] Step 9, spectral reconstruction;

[0083] For each pixel within the 3×3 virtual moving window, the reconstructed spectral value Y(1,2,…,8) of the shortwave infrared band can be calculated using the values ​​of parameters a and b in the obtained fitting function y=a×x+b and the reference spectral vector X2(1,2,…,8) of the shortwave infrared band.

[0084] In this implementation example, the reference spectral vector X2 of a 3×3 virtual moving window in the shortwave infrared (SWIR) band is (2790,1597,1851,1730,1147,1122,1064,903). Using the fitting function y = 0.9703×x-10.142, the reconstructed spectrum Y = (2697,1539,1786,1669,1103,1079,1022,866) can be calculated.

[0085] Step 10: Generate high-resolution shortwave infrared band images;

[0086] For 200×200 3×3 virtual moving windows, if steps 5 to 9 are performed for each virtual moving window, a high-resolution shortwave infrared band image F with 8 bands reconstructed can be obtained.

[0087] For details, please refer to Figure 6 , Figure 6 The image F is the fused high spatial resolution shortwave infrared image in this embodiment of the invention.

[0088] To evaluate the image fusion method in this invention, quantitative indicators are used in this embodiment to evaluate the spectral fidelity and true spatial resolution of the fused image. The quantitative evaluation indicators used include the deviation index, correlation coefficient, spectral angle, global error, and image quality index. The above quantitative evaluation indicators are derived from the calculation formulas and meanings of the quantitative evaluation indicators recorded in Reference 1 (Xu Ning, Xiao Xinyao, You Hongjian, et al. Remote sensing image fusion combining HCT transform and joint sparse model [J]. Acta Geodaetica et Cartographica Sinica, 2016, 45(4):434-441), and can be calculated by commercial remote sensing software. The meanings of the above quantitative evaluation indicators are as follows: the deviation index, correlation coefficient, global error, and spectral angle can evaluate the similarity between the fused image and the original shortwave infrared image. In particular, the spectral angle can measure the spectral similarity between the two. The smaller the value, the better the spectral fidelity. The image quality index comprehensively considers the spatial and spectral information of the fused image. The larger the image quality index value, the better the quality of the fused image.

[0089] To conduct data comparison, research, and evaluation, this embodiment employs the GS (Gram-Schmidt Spectral Sharpening) and PC (PC Spectral Sharpening) methods built into the commercial remote sensing software ENVI to fuse the low spatial resolution shortwave infrared SWIR1, SWIR2, and SWIR3 bands and the high spatial resolution red band of the WorldView-3 image. The GS method described above originates from the principle and calculation process of the GS method described in reference 2 (Laben et al., Process for Enhancing the Spatial Resolution of Multispectral Imagery Using Pan-Sharpening, US Patent 6,011,875). The PC method described above originates from the principle and calculation process of the PC method described in reference 3 (Welch, R. and W. Ahlers, 1987. "Merging Multiresolution SPOT HRV and Landsat TM Data." Photogrammetric Engineering & Remote Sensing, 53(3), pp.301-303).

[0090] The GS method and PC method require that the wavelength ranges of the high spatial resolution panchromatic image and the low spatial resolution multispectral image participating in the fusion be basically the same. The greater the difference in wavelength range, the more severe the color distortion of the fused image will be.

[0091] The quantitative evaluation indicators of the fusion results obtained by the method of this invention, the GS method, and the PC method are shown in Table 1 below. Among them, the spectral angle and global error are calculated as one index value for the 8-band image, while the deviation index, correlation coefficient, and image quality index are calculated for each band separately. The average value of the 8 bands is given in Table 1.

[0092] Table 1 Quantitative evaluation indicators of the fusion results of three different methods

[0093] Fusion Method Correlation coefficient bias index Spectral angle Global error Quality Index Method of the present invention 0.845 0.113 0.035 8.95 0.679 GS method 0.458 0.331 0.115 10.16 0.437 PC method 0.325 0.376 0.119 0.98 0.316

[0094] The images fused by the three methods all effectively increased the spatial detail information of the shortwave infrared image and improved the spatial resolution, which is consistent with the visual results. Table 1 shows that the correlation coefficients between the GS method and the PC method and the original shortwave infrared image are very low, which is consistent with their visually distorted appearance. Furthermore, compared with quantitative indicators measuring spectral fidelity such as deviation index, spectral angle, and global error, the method of this invention is also superior to the GS method and the PC method. In particular, the deviation index and spectral angle of the fusion method of this invention are significantly superior, further demonstrating its significant advantage in spectral fidelity. For the comprehensive indicator of image quality index, the method of this invention is also significantly superior to the GS method and the PC method. Overall, visual observation and quantitative indicator evaluation demonstrate that the fusion method proposed in this invention can significantly improve the spatial resolution of shortwave infrared images while effectively preserving the original spectral characteristics.

[0095] As can be seen from the above description, the image fusion method for improving the spatial resolution of shortwave infrared remote sensing images of the present invention has the technical advantages of simple and easy-to-implement algorithm, stable and reliable operation, high computational efficiency and excellent results.

[0096] As another embodiment, the method of this embodiment can also be applied to improve the spatial resolution of ASTER's five shortwave infrared remote sensing images from 30 meters to 15 meters. The operation steps and methods are similar to those for WorldView-3 remote sensing images, and will not be repeated here.

[0097] On the other hand, the present invention also provides an apparatus for implementing an image fusion method to improve the spatial resolution of shortwave infrared remote sensing images.

[0098] See Figure 7 , Figure 7 This is a schematic diagram of the device structure for implementing the image fusion method for improving the spatial resolution of shortwave infrared remote sensing images according to an embodiment of the present invention, including:

[0099] The image acquisition module 101 receives geometrically corrected high spatial resolution visible-near infrared image A and low spatial resolution shortwave infrared image B from two different types of sensors on the same satellite.

[0100] The spatial resolution adjustment module 102 adjusts the spatial resolution of the low spatial resolution shortwave infrared image B to T times the spatial resolution of the visible-near infrared band image A through resampling, and generates a new image C.

[0101] Principal component analysis module 103 performs principal component analysis on all bands of high spatial resolution VNIR image A, takes its first component as the spatial information factor of ground features, and generates image D.

[0102] The image overlay and reconstruction module 104 uses the spatial resolution of image A as a reference, and resamples and overlays the three sets of remote sensing images in the order of images D, A, and C to generate a new image E.

[0103] The virtual moving window setting module 105 sets a virtual moving window with a size of T×T pixels. Subsequent calculations and remote sensing analysis are all performed within the virtual moving window of size T×T.

[0104] The dominant feature pixel determination module 106 statistically analyzes the mean and standard deviation of the gray values ​​of the red band of the visible-near-infrared image in the image E within the T×T virtual moving window, and determines and identifies the dominant feature pixels.

[0105] The dominant feature pixel spectral synthesis calculation module 107 calculates the average gray value of all pixels identified as dominant features within the T×T virtual moving window in image E, and obtains the synthesized reference spectral vector.

[0106] The spectral fitting module 108, for each pixel within the T×T virtual moving window, uses the least squares method to obtain the spectral fitting function for each pixel based on the actual spectrum of each pixel and the reference spectrum of the visible-near infrared band.

[0107] The spectral reconstruction module 109 simulates and reconstructs a new shortwave infrared spectrum for each pixel within the T×T virtual moving window using a fitting function and the reference spectral direction of the shortwave infrared band.

[0108] The high-resolution shortwave infrared band image construction module 110 performs spectral reconstruction on each virtual moving window to construct a new high spatial resolution shortwave infrared band image F.

[0109] The image fusion apparatus for improving the spatial resolution of shortwave infrared remote sensing images according to the above embodiments is used to implement the corresponding image fusion method for improving the spatial resolution of shortwave infrared remote sensing images in the foregoing embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0110] 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. An image fusion method for improving the spatial resolution of shortwave infrared remote sensing images, characterized in that, It includes the following steps: Step 1: Input the geometrically corrected high spatial resolution visible-near infrared image and the original low spatial resolution shortwave infrared image from the same satellite; Step 2: The original low spatial resolution shortwave infrared image is resampled and adjusted to an integer multiple of the spatial resolution of the high spatial resolution visible light-near infrared image to generate a resampled low spatial resolution shortwave infrared image. Step 3: Perform principal component analysis on the high spatial resolution visible-near infrared image and take its first component as the grayscale image of the spatial information factor of ground features. Step 4: Based on the spatial resolution of the high spatial resolution visible-near infrared image, the three sets of remote sensing images are resampled and overlaid in the order of ground feature spatial information factor grayscale image, high spatial resolution visible-near infrared image, and resampled low spatial resolution shortwave infrared image to generate the overlaid and reassembled image. Step 5: Set a virtual moving window with a size of T×T pixels, and cover and apply it to all pixels; Step 6: Determine and label dominant ground feature pixels within the virtual moving window; Step 6 specifically involves: statistically analyzing the mean (MEAN) and standard deviation (STD) of the gray values ​​in the red band of the visible-near-infrared image in the overlaid and reconstructed image within a virtual moving window of T×T pixels; pixels whose gray values ​​fall within the range of (MEAN-0.5×STD, MEAN+0.5×STD) are determined to be dominant ground feature pixels and labeled accordingly. Step 7: Calculate the spectral average value of the pixels of the dominant features in the virtual moving window to obtain the synthesized reference spectrum; Step 7 specifically involves: for all pixels identified as dominant features within the T×T virtual moving window in the overlaid and recombined image, calculate the average gray value of the dominant feature pixels in each of the 1+n+m bands to obtain the reference spectral vector X(1,2,...,1+n+m) of the T×T virtual moving window, where n and m are natural numbers, n is the number of bands in the high spatial resolution visible-near infrared image, and m is the number of bands in the low spatial resolution shortwave infrared image; Step 8: Calculate the spectral fitting function for each pixel using the least squares method. Specifically, Step 8 involves: obtaining the spatial information of ground features and the gray values ​​of the visible-near-infrared band for each pixel within the T×T virtual moving window, thus obtaining the spectral vector Y(1,2,...,1+n) for each pixel. Within the T×T virtual moving window, obtain the spectral vector Y(1,2,...,1+n) and the reference spectrum X1(1,2,...,1+n) for each pixel, and fit them using the least squares method to obtain the values ​​of parameters a and b of the fitting function y=a×x+b for each pixel, where x represents the gray value in the reference spectral vector and y represents the gray value in the corresponding pixel spectral vector. Step 9: Using the fitting function and the reference spectrum of the short-wave infrared band, simulate and reconstruct the new short-wave infrared spectrum; Step 10: Construct new high-resolution shortwave infrared band images.

2. The image fusion method for improving the spatial resolution of shortwave infrared remote sensing images according to claim 1, characterized in that, The remote sensing images in step 1 are high spatial resolution visible-near infrared images and raw low spatial resolution shortwave infrared images from two different types of sensors on the same satellite, after geometric correction.

3. The image fusion method for improving the spatial resolution of shortwave infrared remote sensing images according to claim 2, characterized in that, Step 2 includes the following steps: adjusting the spatial resolution of the original low spatial resolution shortwave infrared image with m bands to an integer multiple T of the spatial resolution of the high spatial resolution visible-near infrared image by resampling, and generating a new resampled low spatial resolution shortwave infrared image with m bands.

4. The image fusion method for improving the spatial resolution of shortwave infrared remote sensing images according to claim 3, characterized in that, In step 3, the spatial information factors of ground features are obtained by performing principal component analysis on all bands of a high spatial resolution visible-near-infrared image with n bands, resulting in PC1 to PC2. n There are n components in total. The first component (PC1) is taken as the grayscale image of the spatial information factor of ground features.

5. The image fusion method for improving the spatial resolution of shortwave infrared remote sensing images according to claim 4, characterized in that, Step 4 specifically involves: using the spatial resolution of the high spatial resolution visible-near infrared image as a benchmark, resampling and overlaying the three sets of remote sensing images in the order of grayscale image of ground feature spatial information factor, high spatial resolution visible-near infrared image, and resampled low spatial resolution shortwave infrared image to generate a new overlay image. The overlay image contains 1+n+m bands, and the first component representing the spatial information of ground features, PC1, is the first band of the overlay image.

6. The image fusion method for improving the spatial resolution of shortwave infrared remote sensing images according to claim 5, characterized in that, Step 5 specifically involves setting the image size of the resampled low spatial resolution shortwave infrared image to W1×W2 pixels, then setting the image size of the overlaid and recombined image to (T×T)×(W1×W2) pixels, setting a virtual moving window with a window size of T×T pixels, and dividing the overlaid and recombined image into W1×W2 virtual windows.

7. The image fusion method for improving the spatial resolution of shortwave infrared remote sensing images according to claim 6, characterized in that, Step 9 specifically involves: for each pixel within the T×T virtual moving window, using the obtained fitting function y=a×x+b with the values ​​of parameters a and b, and the reference spectral vector X2(1,2,...,m) of the shortwave infrared band, calculating the reconstructed spectral value Y(1,2,...,m) of each pixel.

8. The image fusion method for improving the spatial resolution of shortwave infrared remote sensing images according to claim 7, characterized in that, Specifically, step 10 involves performing steps 5 to 9 on each of the W1×W2 T×T virtual moving windows to obtain a high-resolution shortwave infrared band image with m bands.

9. An apparatus for implementing the image fusion method for improving the spatial resolution of shortwave infrared remote sensing images as described in claim 1, characterized in that, Including mutual communication connections: The image acquisition module is used to input geometrically corrected high spatial resolution visible-near infrared images and raw low spatial resolution shortwave infrared images from the same satellite. The spatial resolution adjustment module resamples the original low spatial resolution shortwave infrared image to an integer multiple of the spatial resolution of the high spatial resolution visible light-near infrared image, generating a resampled low spatial resolution shortwave infrared image. The principal component analysis module performs principal component analysis on high spatial resolution visible-near infrared images and takes the first component as the grayscale image of the spatial information factor of ground features. The image overlay and reconstruction module is used to resample and overlay three sets of remote sensing images based on the spatial resolution of high spatial resolution visible light-near infrared images, in the order of ground feature spatial information factor grayscale images, high spatial resolution visible light-near infrared images, and resampled low spatial resolution shortwave infrared images, to generate overlay and reconstructed images. The virtual moving window setting module is used to set a virtual moving window with a window size of T×T pixels, and to cover and apply it to all pixels; The dominant feature pixel determination module is used to determine and identify dominant feature pixels within a virtual mobile window; The dominant feature pixel spectral synthesis calculation module is used to calculate the pixel value of the spectral average of dominant features in the virtual moving window to obtain the synthesized reference spectrum. The spectral fitting module is used to calculate the spectral fitting function for each pixel using the least squares method. The spectral reconstruction module is used to simulate and reconstruct new shortwave infrared spectra using a fitting function and a reference spectrum in the shortwave infrared band. A high-resolution shortwave infrared image construction module for building new high-resolution shortwave infrared images.

Images