A band interpolation method for improving spectral resolution of hyperspectral remote sensing image

Abstract

The present application relates to the field of hyperspectral remote sensing, and particularly relates to a band interpolation method for improving spectral resolution of a hyperspectral remote sensing image, comprising the following steps: obtaining a hyperspectral remote sensing image and reading metadata; performing radiation correction on the hyperspectral remote sensing image; performing atmospheric correction on the radiation-corrected hyperspectral remote sensing image; removing redundant bands of the hyperspectral remote sensing image; removing periodic stripes of the hyperspectral remote sensing image; performing filtering processing on the hyperspectral remote sensing image; performing band interpolation processing on the hyperspectral remote sensing image to improve the resolution of the hyperspectral remote sensing image; eliminating background spectral trends of the hyperspectral remote sensing image; determining the deepest absorption band between the required interpolation bands; and extracting absorption peak characteristics. The present application proposes a spectral band interpolation method to improve the spectral resolution of spaceborne hyperspectral remote sensing data, so as to realize accurate identification of different minerals or different sub-species of the same mineral.

Description

Technical Field

[0001] This invention relates to the field of hyperspectral remote sensing, and more specifically to a band interpolation method for improving the spectral resolution of hyperspectral remote sensing images. Background Technology

[0002] As a rapidly developing detection method, remote sensing technology is widely used in the geological field due to its characteristics of speed, non-invasiveness, wide coverage, and low cost, and is also being promoted and applied in various fields such as vegetation and soil. Spectroscopy is relatively sensitive to the mineralogical and chemical properties of minerals and rocks. Compared with multispectral remote sensing data, hyperspectral data can contain hundreds of continuous, narrow-band spectral bands, and is better able to capture the subtle spectral features of ground objects in specific bands. Multispectral data, due to its broad bands, may obscure these key spectral details. Therefore, hyperspectral remote sensing has demonstrated its irreplaceable role in fine classification and quantitative inversion.

[0003] For example, muscovite minerals exhibit distinct spectral characteristics between 2180-2228 nm due to the influence of Al-OH, and the wavelength position of this absorption characteristic varies with the n(Al) content in the mica. Ⅵ The content of mica decreases, leading to a shift towards longer wavelengths. In other words, based on elemental content, mica can be further divided into sodium mica (<2198nm, shifting towards shorter wavelengths), common muscovite (around 2200nm), and polysilica (>2208nm, shifting towards longer wavelengths). However, due to limitations in sensor hardware such as the physical constraints of spectroscopic devices and detector sensitivity, the bandwidth (5-20nm) of currently acquired hyperspectral data still has limitations in distinguishing minerals with similar spectral characteristics. Wavelength plotting methods have proven to be effective, reliable, repeatable, and sufficiently accurate. Several wavelength plotting methods have been proposed, including quadratic polynomial and cubic spline interpolation. Interpolation algorithms can generate virtual bands between original bands, constructing more continuous spectral curves. After performing short-wave infrared (SWIR) interpolation on hyperspectral data, the weak absorption peak characteristics of clay minerals can be enhanced, thereby improving spectral matching accuracy, for example, it can be used to distinguish different varieties of mica. This method not only compensates for the shortcomings of sensor performance, but also allows for comparison with standard spectral libraries (such as USGS, JPL spectral libraries, etc.) and indoor ASD measurement spectra, providing more accurate data support for mineral identification.

[0004] However, existing quadratic and cubic polynomial interpolation methods still have certain shortcomings and defects. Their mathematical continuity and curve fitting are poor, and their agreement with real-world conditions in field verification is not as good as that of quartic polynomials. Using quintic or higher-order polynomials carries the risk of overfitting, which can lead to non-physical, spurious "waves" and "peaks" in the interpolated spectral curves, severely distorting the true spectral characteristics and reducing the accuracy of mineral identification. Existing quartic polynomial interpolation methods, due to their good balance between flexibility and stability, are important tools in many scientific and engineering fields. For example, in computer graphics and animation, they are used to generate smooth animation curves, design font outlines, and create free curves in computer-aided design; in robotics and trajectory planning, they are used to plan the motion trajectory of robotic arm end effectors or mobile robots from start to finish; and in engineering mechanics and structural analysis, they are used to analyze the bending deformation of structural components such as beams and plates under complex loads. In the field of remote sensing and image processing, fourth-order polynomial interpolation is also used for image geometric correction and image scaling. However, due to the complex shape of hyperspectral curves and the large amount of data, the calculation process is quite cumbersome. Furthermore, research on the fine identification of minerals using spaceborne hyperspectral remote sensing images is not widespread. Therefore, no one has applied it to the improvement of the spectral resolution of hyperspectral remote sensing images involved in this invention patent.

[0005] Chinese invention patent application No. 202411443910.7 discloses a method for estimating nitrogen content in tea canopy leaves based on hyperspectral data. It employs hyperspectral data interpolation to ensure the continuity and convenience of the calculation, but does not disclose the specific processing method or procedure. Furthermore, it targets data collected by a portable hyperspectral ground object spectrometer and does not process spaceborne hyperspectral data.

[0006] Therefore, there is an urgent need for a spectral band interpolation method to improve the spectral resolution of hyperspectral remote sensing data, so as to achieve accurate identification of different minerals or different subspecies of the same mineral. Summary of the Invention

[0007] To address the problem of inaccurate identification of minerals with similar spectral characteristics in existing technologies, this invention provides a band interpolation method for improving the spectral resolution of hyperspectral remote sensing images, comprising the following steps:

[0008] Acquire hyperspectral remote sensing images and read metadata;

[0009] Radiometric correction is performed on hyperspectral remote sensing images, converting the digital quantization values ​​of the hyperspectral remote sensing images into radiance.

[0010] Atmospheric correction is performed on the radiometrically corrected hyperspectral remote sensing image. The radiance is input into the atmospheric correction model to eliminate the effects of atmospheric absorption and scattering, and then the surface reflectance is output.

[0011] Remove redundant bands from hyperspectral remote sensing images;

[0012] Fourier transform is used to remove periodic stripes from hyperspectral remote sensing images, making the grayscale of uniform ground features in hyperspectral remote sensing images consistent.

[0013] The hyperspectral remote sensing image after removing periodic stripes is filtered.

[0014] A fourth-order polynomial is used to perform band interpolation on the filtered hyperspectral remote sensing image to improve the spectral resolution of the hyperspectral remote sensing image.

[0015] The background spectral trend of the hyperspectral remote sensing image after interpolation is eliminated by using continuum removal. The deepest absorption band between the required interpolation bands is determined by the derivative method, and the absorption peak features are extracted to identify different minerals or different subspecies of the same mineral.

[0016] Furthermore, the process of acquiring hyperspectral remote sensing images and reading metadata includes:

[0017] Log in to the Natural Resources Satellite Remote Sensing Cloud Service Platform to obtain hyperspectral remote sensing images of the study area and adjust the spectral range to cover the characteristic absorption bands of the target minerals;

[0018] Obtain the raw data file and read the metadata of the hyperspectral remote sensing image.

[0019] Furthermore, gain parameters and offset parameters are extracted from the hyperspectral remote sensing image metadata to convert the digital quantization values ​​of the hyperspectral remote sensing image into radiance.

[0020] Furthermore, the process of removing redundant bands from the spectral remote sensing image includes:

[0021] Extract the center wavelength and bandwidth of each band from the metadata of the hyperspectral remote sensing image;

[0022] Remove the overlapping wavelength ranges caused by channel design or instrument errors and check the spectral curves.

[0023] Furthermore, the hyperspectral image redundant bands include: the overlapping band between the visible and near-infrared wavelength range of 395~1040nm and the short-wave infrared wavelength range of 1005~2501nm.

[0024] Furthermore, a sliding window least squares filtering method is used to filter the hyperspectral remote sensing image. Within the sliding window, the local spectral curve is fitted by a polynomial, and the fitted value is used to replace the original noise value, thus preserving the spectral features while suppressing noise.

[0025] Furthermore, the fourth-degree polynomial is expressed as:

[0026]

[0027] in, Indicates the pixel at wavelength The reflectivity value at that location, , , , , For the coefficients of this function, model each pixel independently and solve for the coefficients using data from neighboring bands. , , , , .

[0028] Furthermore, local maximum points are selected as nodes on the spectral curve of the hyperspectral remote sensing image, and adjacent nodes are connected by straight lines to form a continuum curve, thereby performing continuum removal.

[0029] Furthermore, the process of determining the deepest absorption band between the required interpolation bands of a hyperspectral remote sensing image using the derivative method includes:

[0030] Calculate the first derivative of the spectral reflectance function of a hyperspectral remote sensing image;

[0031] Find the minimum point of the first derivative to determine the center position of the deepest absorption band between the required interpolation bands of the hyperspectral remote sensing image;

[0032] Set a depth threshold to retain only absorption bands with depths exceeding the threshold in order to eliminate noise interference.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] This invention provides a method for interpolation of the influence band of airborne hyperspectral imaging. After preprocessing the remote sensing image, the method uses a fourth-order polynomial interpolation method to improve the spectral resolution of the image, thereby extracting the spectral features of ground objects more accurately, thus improving the accuracy of mineral identification and further distinguishing different subspecies of the same mineral. This provides reliable technical support for the in-depth application of hyperspectral remote sensing in the fields of geological exploration and mineral classification. Attached Figure Description

[0035] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0036] Figure 1 This is a flowchart of a band interpolation method for improving the spectral resolution of hyperspectral remote sensing images according to the present invention.

[0037] Figure 2 This is a spatial distribution map of the interpolation wavelength positions in an open-pit iron ore area according to an embodiment of the present invention.

[0038] Figure 3 This is a spatial distribution map of iron-rich chlorite and magnesium-rich chlorite in an open-pit iron ore area after interpolation, as shown in an embodiment of the present invention. Detailed Implementation

[0039] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0040] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0041] Taking a certain iron mine as an example, ZY1-02D ​​hyperspectral remote sensing image was used for band interpolation to improve spectral resolution and identify magnesium-rich chlorite and iron-rich chlorite, thus providing an effective means for finding magnetite-rich deposits.

[0042] This invention provides the following technical solutions:

[0043] like Figure 1 The method for improving the spectral resolution of hyperspectral remote sensing images, as shown, specifically includes the following steps:

[0044] S1: Acquire hyperspectral remote sensing images and read metadata.

[0045] In one embodiment, the user logs into the Natural Resources Satellite Remote Sensing Cloud Service Platform and downloads the ZY1-02D ​​image covering an open-pit iron mine in a specific area. The entire image is unaffected by cloud cover and exhibits good ground features.

[0046] In one embodiment, the raw data file is acquired, and the metadata of the hyperspectral remote sensing image is read.

[0047] S2: Perform radiometric correction on hyperspectral remote sensing images, converting the digital quantization values ​​of the hyperspectral remote sensing images into radiance.

[0048] In one embodiment, open the ENVI 5.3 Toolbox, select Radiometric Correction / Radiometric Calibration, and then select VNSW data for radiometric calibration.

[0049] In one embodiment, the radiometric calibration converts the raw brightness value of each pixel in the remote sensing image into radiance: radiance = raw brightness value x gain + offset, where the gain and offset are obtained from metadata.

[0050] S3: Perform atmospheric correction on the radiometrically corrected hyperspectral remote sensing image, input the radiance into the atmospheric correction model, eliminate the effects of atmospheric absorption and scattering, and then output the surface reflectance.

[0051] In one embodiment, in ENVI 5.3, the FLAASH Atmospheric Correction module is accessed via Toolbox / Radiometric Correction / Atmospheric Correction Module. Input parameters such as radiance, sensor height, imaging time, and atmospheric model; output is a cube representing the surface reflectance.

[0052] S4: Remove redundant bands from hyperspectral remote sensing images.

[0053] In one embodiment, the process of removing redundant bands from the spectral remote sensing image includes:

[0054] Extract the center wavelength and bandwidth of each band from the metadata of the hyperspectral remote sensing image;

[0055] Remove the overlapping wavelength ranges caused by channel design or instrument errors and check the spectral curves.

[0056] In one embodiment, the hyperspectral image redundancy band includes an overlapping band between the visible near-infrared wavelength range of 395-1040 nm and the short-wave infrared wavelength range of 1005-2501 nm.

[0057] In one embodiment, the center wavelength and bandwidth (FWHM) of each band are extracted from the metadata to calculate the actual coverage area of ​​the band:

[0058]

[0059]

[0060] in The initial band, To end the band, The center wavelength, For bandwidth. Traverse all bands; if a band exists... of band of ( ), then mark and These are overlapping band pairs. Compare the signal-to-noise ratio (SNR) of the overlapping band pairs and retain the band with the higher SNR.

[0061] In one embodiment, a typical land cover (such as bare soil or vegetation) is selected, and the spectral curves before and after overlap are compared. If the removal is correct, the spectral shape should remain smooth without abrupt changes due to band loss, and key absorption peaks should be preserved intact.

[0062] S5: Fourier transform is used to remove periodic stripes from hyperspectral remote sensing images, making the grayscale of uniform ground features in hyperspectral remote sensing images consistent.

[0063] In one embodiment, ZY1-02D ​​uses pushbroom imaging. Due to inconsistent responses from different detectors, vertical stripes appear in the image. This noise typically manifests as periodic or non-periodic brightness differences. A two-dimensional Fast Fourier Transform (FFT) is used to transform the image from the spatial domain to the frequency domain. Periodic noise in the image will appear as specific high-frequency components in the frequency domain. A logarithmic amplitude spectrum is generated, the position of the vertical bright lines is observed, and their coordinates are recorded. A Gaussian notch filter is used to smooth the transition. The mask is multiplied by the Fourier transform for frequency domain filtering, and an inverse Fourier transform is performed to shift back to zero frequency and transform back to the spatial domain. The original image and the destriped image are compared to verify the destriping effect.

[0064] S6: Filter the hyperspectral remote sensing image after removing periodic stripes.

[0065] In one embodiment, a sliding window least squares filtering method is used to filter the hyperspectral remote sensing image. Within the sliding window, a local spectral curve is fitted by a polynomial, and the fitted value is used to replace the original noise value, thus preserving the spectral features while suppressing noise.

[0066] In one embodiment, Savitzky-Golay filtering is applied to the spectral curve of each pixel, which is essentially a sliding window least squares filter. Window size:

[0067] (Odd number) =

[0068] in, It refers to the wavelength range covered by the window, which typically includes a wavelength range of 5 consecutive bands (e.g., ), For spectral resolution.

[0069] S7: A fourth-order polynomial is used to perform band interpolation on the filtered hyperspectral remote sensing image to improve the spectral resolution of the hyperspectral remote sensing image.

[0070] In one embodiment, the fourth-degree polynomial is expressed as:

[0071]

[0072] in, Indicates the pixel at wavelength The reflectivity value at that location, , , , , For the coefficients of this function, model each pixel independently and solve for the coefficients using data from neighboring bands. , , , , .

[0073] In one embodiment, the least squares method is used to solve the problem: constructing the matrix equation. ,in, For matrix transpose,

[0074]

[0075]

[0076]

[0077] Solution .

[0078] Input the original SWIR band and perform interpolation calculations for each pixel to improve the spectral resolution to 1 nm.

[0079] S8: The background spectral trend of the hyperspectral remote sensing image after interpolation is eliminated by using the continuum removal method. The deepest absorption band between the required interpolation bands is determined by the derivative method, the absorption peak characteristics are extracted, and different minerals or different subspecies of the same mineral are identified.

[0080] In one embodiment, local maximum points are selected as nodes on the spectral curve of a hyperspectral remote sensing image, and adjacent nodes are connected by straight lines to form a continuum curve, thereby performing continuum removal.

[0081] In one embodiment, the process of determining the deepest absorption band between the required interpolation bands of a hyperspectral remote sensing image using the derivative method includes:

[0082] Calculate the first derivative of the spectral reflectance function of a hyperspectral remote sensing image;

[0083] Find the minimum point of the first derivative to determine the center position of the deepest absorption band between the required interpolation bands of the hyperspectral remote sensing image;

[0084] Set a depth threshold to retain only absorption bands with depths exceeding the threshold in order to eliminate noise interference.

[0085] In one embodiment, the first derivative of the interpolated reflectance spectrum is calculated to find the minimum point of the derivative in the characteristic spectral range of Fe-OH in chlorite. = 0 and ),in Reflectivity function The first derivative, Reflectivity function The second derivative. The Fe-OH absorption characteristics of a certain subway mining area were extracted, and the results are shown in Table 1. Figure 2 As shown.

[0086] Table 1. Statistics of absorption peak positions of chlorite after wavelength interpolation.

[0087]

[0088] Combining mining area data and USGS standard spectra, using 2258 nm as the threshold, magnesium-rich chlorite and iron-rich chlorite were distinguished, and the results are as follows: Figure 3 As shown.

[0089] The present invention proposes a band interpolation method to improve the spectral resolution of hyperspectral remote sensing images, which solves the problem in the prior art that it is impossible to accurately identify minerals with similar spectral characteristics, and can accurately identify different minerals or different subspecies of the same mineral.

[0090] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A band interpolation method for improving spectral resolution of a hyperspectral remote sensing image, characterized in that, Includes the following steps: Acquire hyperspectral remote sensing images and read metadata; Radiometric correction is performed on hyperspectral remote sensing images, converting the digital quantization values ​​of the hyperspectral remote sensing images into radiance. Atmospheric correction is performed on the radiometrically corrected hyperspectral remote sensing image. The radiance is input into the atmospheric correction model to eliminate the effects of atmospheric absorption and scattering, and then the surface reflectance is output. Remove redundant bands from hyperspectral remote sensing images; Fourier transform is used to remove periodic stripes from hyperspectral remote sensing images, making the gray levels of uniform ground features in hyperspectral remote sensing images consistent. The hyperspectral remote sensing image after removing periodic stripes is filtered. A fourth-order polynomial is used to perform band interpolation on the filtered hyperspectral remote sensing image to improve the spectral resolution of the hyperspectral remote sensing image. The background spectral trend of the hyperspectral remote sensing image after interpolation is eliminated by using continuum removal. The deepest absorption band between the required interpolation bands is determined by the derivative method, and the absorption peak features are extracted to identify different minerals or different subspecies of the same mineral.

2. The band interpolation method for improving the spectral resolution of a hyperspectral remote sensing image according to claim 1, characterized in that, The process of acquiring hyperspectral remote sensing images and reading metadata includes: Log in to the Natural Resources Satellite Remote Sensing Cloud Service Platform to obtain hyperspectral remote sensing images of the study area and adjust the spectral range to cover the characteristic absorption bands of the target minerals; Obtain the raw data file and read the metadata of the hyperspectral remote sensing image.

3. The band interpolation method for improving the spectral resolution of hyperspectral remote sensing images according to claim 1, characterized in that, Gain and offset parameters are extracted from the metadata of the hyperspectral remote sensing image to convert the digital quantization values ​​of the hyperspectral remote sensing image into radiance.

4. The band interpolation method for improving the spectral resolution of hyperspectral remote sensing images according to claim 1, characterized in that, The process of removing redundant bands from hyperspectral remote sensing images includes: Extract the center wavelength and bandwidth of each band from the metadata of the hyperspectral remote sensing image; Remove the overlapping wavelength ranges caused by channel design or instrument errors and check the spectral curves.

5. The band interpolation method for improving the spectral resolution of hyperspectral remote sensing images according to claim 1, characterized in that, Redundant bands in hyperspectral images include the overlapping bands between the visible and near-infrared wavelength range of 395–1040 nm and the short-wave infrared wavelength range of 1005–2501 nm.

6. The band interpolation method for improving the spectral resolution of hyperspectral remote sensing images according to claim 1, characterized in that, Hyperspectral remote sensing images are filtered using a sliding window least squares filtering method. Within the sliding window, local spectral curves are fitted using a polynomial, and the fitted values ​​are used to replace the original noise values, thus preserving spectral features while suppressing noise.

7. The band interpolation method for improving the spectral resolution of hyperspectral remote sensing images according to claim 1, characterized in that, The fourth-degree polynomial is expressed as: in, Indicates the pixel at wavelength The reflectivity value at that location, , , , , For the coefficients of the function, model each pixel independently and solve for the coefficients using data from neighboring bands. , , , , .

8. The band interpolation method for improving the spectral resolution of hyperspectral remote sensing images according to claim 1, characterized in that, Local maximum points are selected on the spectral curve of the hyperspectral remote sensing image as nodes, and adjacent nodes are connected by straight lines to form a continuum curve, thereby performing continuum removal.

9. The band interpolation method for improving the spectral resolution of hyperspectral remote sensing images according to claim 1, characterized in that, The process of determining the deepest absorption band between the interpolation bands required for hyperspectral remote sensing images using the derivative method includes: Calculate the first derivative of the spectral reflectance function of a hyperspectral remote sensing image; Find the minimum point of the first derivative to determine the center position of the deepest absorption band between the required interpolation bands of the hyperspectral remote sensing image; Set a depth threshold to retain only absorption bands with depths exceeding the threshold to eliminate noise interference.

Images