Unmanned aerial vehicle push-broom hyperspectral automatic correction method and device, equipment and medium
By combining reflectance and atmospheric correction with principal component analysis and feature point matching, the spatial misalignment and spectral discontinuity problems of UAV hyperspectral data were solved, achieving high-precision data correction and improving the quantitative inversion and interpretation of UAV hyperspectral data.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HEBEI SAILHERO ENVIRONMENTAL PROTECTION HIGH TECH
- Filing Date
- 2026-03-20
- Publication Date
- 2026-06-12
AI Technical Summary
During flight, the attitude of the drone platform is unstable due to wind speed and airflow disturbances, which causes problems such as spatial misalignment of hyperspectral data, discontinuity of spectral curves and color overflow in RGB composite images. Existing correction methods cannot guarantee accuracy.
The true reflectance of ground objects is restored by reflectance correction and atmospheric correction. The first principal component feature map is obtained by principal component analysis. Hyperspectral data is corrected by combining feature point matching and geometric transformation fitting.
It improves the accuracy of hyperspectral data, solves problems such as spatial misalignment between bands, discontinuity of spectral curves, and RGB synthesis artifacts, and enhances the accuracy of quantitative inversion and interpretation of the data.
Smart Images
Figure CN122193164A_ABST
Abstract
Description
Technical Field
[0001] This application belongs to the field of UAV pushbroom hyperspectral automatic correction technology, and more specifically, it relates to a UAV pushbroom hyperspectral automatic correction method, device, equipment, and medium. Background Technology
[0002] Pushbroom hyperspectral cameras are characterized by their lightweight design, high spatial resolution, and rich spectral information. When mounted on highly flexible unmanned aerial vehicles (UAVs), they can be applied to fields such as agricultural monitoring, environmental assessment, and resource exploration. However, UAV platforms are susceptible to problems such as unstable flight attitude and uneven speed caused by wind speed and airflow disturbances during flight, resulting in differences in exposure time and significant defects in the acquired hyperspectral data: spatial misalignment between different bands (band non-correspondence), discontinuous and distorted spectral curves, and color overflow and artifacts in RGB composite images. These issues severely reduce the accuracy of quantitative inversion and interpretation of hyperspectral data.
[0003] Methods for hyperspectral data registration (correction) mainly include: First, geometric inversion based on inertial measurement unit (IMU) attitude data and extrinsic parameter models, which has strict physical deduction and high accuracy. However, in UAV platforms, the sampling frequency of hyperspectral cameras is high, while the sampling frequency of IMUs (especially low-cost IMUs) is low, resulting in the inability to obtain complete geometric registration parameters and the inability to guarantee correction accuracy. Second, feature matching-based methods (such as SIFT and SURF algorithms) extract image feature points and perform matching to calculate the transformation matrix, which can handle complex geometric deformations. However, they have poor adaptability to hyperspectral bands with low contrast and poor texture, and are prone to registration failure due to insufficient feature points.
[0004] Therefore, it is urgent to improve existing hyperspectral data correction methods to enhance the accuracy of hyperspectral data. Summary of the Invention
[0005] The purpose of this application is to provide a UAV pushbroom-based automatic hyperspectral correction method, apparatus, device, and medium that can improve the accuracy of hyperspectral data. To achieve the above objective, the technical solution provided by this application is as follows: Firstly, a pushbroom-based automatic hyperspectral correction method for unmanned aerial vehicles (UAVs) is provided, including: Acquire hyperspectral data; the hyperspectral data includes radiance values in multiple bands; Reflectance correction and atmospheric correction are performed on the radiance values of each band to obtain the surface reflectance data for each band; Principal component analysis was performed on surface reflectance data from multiple bands to obtain the first principal component feature map; Using the first principal component feature map as a reference feature, the surface reflectance data of each band is matched with the first principal component feature map to obtain multiple matching point pairs for each band. If the number of matching point pairs in each band is greater than or equal to a preset threshold, data fitting is performed based on multiple matching point pairs in each band to obtain a transformation matrix for each band. The surface reflectance data of each band is then converted based on the transformation matrix to obtain corrected hyperspectral data.
[0006] Secondly, a drone-based pushbroom-type hyperspectral automatic correction device is provided, comprising: The data acquisition module is used to acquire hyperspectral data; the hyperspectral data includes radiance values of multiple bands. The radiation correction module is used to perform reflectivity correction and atmospheric correction on the radiance values of each band to obtain the surface reflectivity data for each band. The principal component extraction module is used to perform principal component analysis on surface reflectance data of multiple bands to obtain the first principal component feature map. The feature point matching module is used to use the first principal component feature map as a reference feature to match the surface reflectance data of each band with the first principal component feature map to obtain multiple matching point pairs for each band. The data correction module is used to perform data fitting based on multiple matching point pairs in each band when the number of matching point pairs in each band is greater than or equal to a preset threshold, to obtain a transformation matrix for each band, and to convert the surface reflectance data of each band based on the transformation matrix to obtain corrected hyperspectral data.
[0007] Thirdly, embodiments of this application also provide an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the UAV pushbroom-type hyperspectral automatic correction method provided in any possible implementation of the first aspect.
[0008] Fourthly, embodiments of this application also provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the UAV pushbroom-based hyperspectral automatic correction method provided in any possible implementation of the first aspect.
[0009] The beneficial effects of the technical solution provided in this application are as follows: The UAV pushbroom hyperspectral automatic correction method, apparatus, equipment, and medium provided in this application, compared with related technologies, firstly restore the true reflectance of ground objects through reflectance correction and atmospheric correction, avoiding interference from non-ground object factors on feature point extraction, and ensuring the accuracy of feature point matching from the data source; then, based on principal component analysis, a first principal component feature map is obtained. Through the global reference characteristics of the first principal component feature map, the spatial feature information of hyperspectral data is maximized, allowing sufficient effective feature points to be extracted even in texture-deficient bands, avoiding the registration failure caused by low contrast and insufficient feature points in texture-deficient bands in traditional single feature matching methods; finally, based on feature point matching and geometric transformation fitting, complex geometric deformations such as translation, rotation, scaling, and shearing caused by UAV pushbroom imaging can be accurately quantified and corrected, fundamentally solving problems such as spatial misalignment between bands, discontinuous spectral curves, and RGB synthesis artifacts, thereby improving the accuracy of hyperspectral data. Attached Figure Description
[0010] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments of this application will be briefly introduced below.
[0011] Figure 1 A flowchart illustrating the UAV pushbroom-based automatic hyperspectral correction method provided in this application embodiment; Figure 2 A flowchart illustrating another embodiment of the UAV pushbroom-based hyperspectral automatic correction method provided in this application; Figure 3 A structural block diagram of the UAV pushbroom-type hyperspectral automatic correction device provided in the embodiments of this application; Figure 4 A schematic block diagram of an electronic device provided in an embodiment of this application. Detailed Implementation
[0012] The embodiments of this application are described below with reference to the accompanying drawings. It should be understood that the embodiments described below with reference to the accompanying drawings are exemplary descriptions for explaining the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions of the embodiments of this application.
[0013] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the terms “comprising” and “including” as used in embodiments of this application mean that the corresponding feature can be implemented as the presented feature, information, data, step, operation, element, and / or component, but do not exclude implementation as other features, information, data, step, operation, element, component, and / or combinations thereof supported by the art. It should be understood that when we say that an element is “connected” or “coupled” to another element, the one element can be directly connected or coupled to the other element, or it can mean that the one element and the other element establish a connection relationship through an intermediate element. Furthermore, “connected” or “coupled” as used herein can include wireless connection or wireless coupling. The term “and / or” as used herein indicates at least one of the items defined by the term; for example, “A and / or B” can be implemented as “A,” or as “B,” or as “A and B.” When describing multiple (two or more) items, if the relationship between the multiple items is not explicitly defined, the multiple items can refer to one, several or all of the multiple items. For example, the description of "parameter A includes A1, A2, A3" can be implemented as parameter A includes A1 or A2 or A3, or it can be implemented as parameter A includes at least two of the three items A1, A2 and A3.
[0014] This application provides an unmanned aerial vehicle (UAV) pushbroom-based automatic hyperspectral calibration method, which can be executed by electronic devices, such as... Figures 1-2 As shown, the method may include: S101: Acquire hyperspectral data; hyperspectral data includes radiance values in multiple bands.
[0015] In this embodiment, the hyperspectral data comes from a pushbroom hyperspectral camera developed by Company A, with a spectral range of 400-1000nm. The raw data is stored in a raw format using Digital Number (DN) values. The camera is mounted on a drone, flying at an altitude of 300m and a speed of 3.3m / s. After data conversion algorithms such as dark current removal and spectral line alignment, the raw format data is converted into typical hyperspectral cube data (C×H×W) using the Geospatial Data Abstraction Library (GDAL), where C represents the number of bands, H represents the number of rows, and W represents the number of columns.
[0016] S102: Perform reflectivity correction and atmospheric correction on the radiance values of each band to obtain the surface reflectivity data for each band.
[0017] In this embodiment, reflectance correction and atmospheric correction are performed sequentially on the original radiance values of each band, and the surface reflectance data of each band is output. This can eliminate the influence of camera hardware response and light intensity, as well as the spectral distortion caused by atmospheric scattering, absorption and refraction, restore the true reflectance characteristics of ground objects, and lay an accurate data foundation for subsequent geometric correction.
[0018] Reflectance correction was performed using a whiteboard and blackboard, while atmospheric correction was achieved using a gray reflectance blanket. The reflectance correction process is as follows: Use a whiteboard (with known reflectivity) ) and blackboard (with known reflectivity) The corresponding radiance value was measured. and Assuming reflectivity With radiance value The two sides approximately satisfy a linear relationship:
[0019] Given two groups ( , )and( , ), and obtain the coefficients a and b:
[0020]
[0021] For any radiance value From the pixels, the corresponding reflectance can be obtained. for:
[0022] Before the drone takes flight, a flight path is pre-set. A 1m x 1m gray reflective mat with a reflectivity of 0.5 is placed in suitable areas along the flight path. Drone flights are generally conducted in clear weather, so we can assume that the atmospheric path radiance is 0. An empirical formula is used in the gray mat area:
[0023] in The reflectance of the gray cloth reflectivity blanket is 0.5. The gain coefficient is calculated from the reflectivity data after reflectivity correction. :
[0024] Therefore, the atmospherically corrected surface reflectance is:
[0025] in, This represents the atmospherically corrected surface reflectance. This represents the radiance value before atmospheric correction.
[0026] S103: Principal component analysis was performed on surface reflectance data of multiple bands to obtain the first principal component feature map.
[0027] In this embodiment, the hyperspectral data contains 240 bands, exhibiting high information redundancy. A transpose dimension transformation can be performed on the surface reflectance data across multiple bands to obtain a two-dimensional matrix in (H×W)×C format, conforming to the pixel×band input structure required for PCA principal component analysis. The transformed (H×W)×C format data is then subjected to PCA principal component analysis. PCA principal component analysis is the most effective algorithm for dimensionality reduction, noise reduction, and feature enhancement; after dimensionality reduction, the first 3-6 principal components can contain over 99% of the information, with the first principal component feature map being particularly valuable. The first principal component feature map contains the most information, accounting for 90%-97% of the total. The extraction steps are as follows: (1) Data standardization: Zero mean standardization is performed on the data of each band column of the two-dimensional matrix to eliminate the differences in the dimensions and values of reflectance of different bands and avoid high value bands dominating the principal component calculation; (2) Calculate the covariance matrix: Based on the standardized two-dimensional matrix, solve the covariance matrix between the bands to quantify the linear correlation of reflectance data of each band; (3) Eigenvalue and eigenvector solution: Perform eigenvalue decomposition on the covariance matrix to obtain a set of eigenvalues and corresponding eigenvectors. The magnitude of the eigenvalues represents the information contribution rate of the corresponding principal components. (4) Selecting principal components: Sort the eigenvalues from largest to smallest and select the eigenvector corresponding to the largest eigenvalue as the first principal component transformation matrix; (5) Generation of the first principal component feature map: Perform a linear projection transformation on the standardized two-dimensional matrix and the first principal component transformation matrix to obtain the flattened first principal component feature data. The specific calculation formula is as follows:
[0028] in, This represents the flattened first principal component feature data. Represents the standardized two-dimensional matrix. Let represent the transformation matrix of the first principal component.
[0029] The flattened first principal component feature data is reconstructed into a two-dimensional image with the same spatial dimension as the original data, thus obtaining the first principal component feature map: ; in, This represents the feature map of the first principal component. For dimension reshaping functions.
[0030] S104: Using the first principal component feature map as a reference feature, perform feature point matching between the surface reflectance data of each band and the first principal component feature map to obtain multiple matching point pairs for each band.
[0031] In this embodiment, a scale-invariant feature transform (SIFT), a binary feature extraction algorithm, or a corner feature extraction algorithm can be used to match the surface reflectance data of each band with the first principal component feature map for feature points. The binary feature extraction algorithms include Oriented Fast and Rotated BRIEF (ORB algorithm) and Fast Retina Keypoint (FREAK algorithm). The corner feature extraction algorithms include Harris Corner Detection (Harris) and Shi-Tomasi Corner Detection (Shi-Tomasi algorithm), among others.
[0032] Taking the scale-invariant feature transform algorithm as an example, OpenCV's SIFT algorithm can be used. To adapt to the extremely low contrast of the original hyperspectral bands, the contrast threshold of SIFT is reduced from the default 0.04 to 0.01, thereby increasing the number of feature point detections by 5-15 times and ensuring that sufficient matching point pairs can still be obtained on noisy and texture-weak UAV hyperspectral data.
[0033] S105: If the number of matching point pairs in each band is greater than or equal to the preset number threshold, perform data fitting based on multiple matching point pairs in each band to obtain the transformation matrix of each band. Based on the transformation matrix, transform the surface reflectance data of each band to obtain the corrected hyperspectral data.
[0034] Specifically, each matching point pair includes a reference feature point and a corresponding query feature point. The reference feature point is the coordinates of the feature point in the first principal component feature map, and the query feature point is the coordinates of the feature point in the surface reflectance data of each band.
[0035] In this embodiment, after obtaining multiple matching point pairs for each band (1500–4000 SIFT feature points are usually extracted for a single band), multiple optimal matching point pairs can be selected from the multiple matching point pairs for data fitting.
[0036] For example, data fitting based on multiple matching point pairs for each band includes: S1051: Construct an index structure based on multiple reference feature points; S1052: For each query feature point, search the index structure for the nearest candidate matching point and the second nearest candidate matching point; calculate the ratio between the nearest neighbor distance and the second nearest neighbor distance, and select the candidate matching point whose ratio is less than a preset ratio threshold from multiple nearest candidate matching points as the optimal matching point, and take the query feature point and the corresponding optimal matching point as the optimal matching point pair; where the nearest neighbor distance is the distance between the nearest candidate matching point and the query feature point, and the second nearest neighbor distance is the distance between the second nearest candidate matching point and the query feature point; S1053: Data fitting based on multiple optimal matching point pairs for each band.
[0037] Specifically, a fast matching method based on the FLANN index, which is an approximate nearest neighbor, can be used. This method can reduce the matching time from seconds to milliseconds while maintaining a matching accuracy of over 99%. The matching process includes: First, construct an index structure containing 5 random KD-Trees (index_params = dict(algorithm=1,trees=5)). Then, for each query descriptor, a finite backtracking priority search is performed in the index (search_params = dict(checks=50)), returning the two nearest and second nearest candidate matching points; Finally, the Lowe ratio test was applied, and only those meeting the requirements were retained. The matching point pair is taken as the optimal matching point.
[0038] After obtaining multiple optimal matching point pairs, data fitting can be performed using methods such as least squares, robust estimation algorithm (RandomSample Consensus, RANSAC), or Least Median of Squares (LMEDS). Taking LMEDS as an example, this embodiment converts the coordinates of multiple optimal matching point pairs into N×1×2 floating-point arrays src_pts and ref_pts. Then, the OpenCV estimateAffine2D function is called to estimate the 2×3 affine transformation matrix M. Based on the affine transformation matrix M, the surface reflectance data of each band is transformed to obtain the corrected hyperspectral data.
[0039] As can be seen from the above, this embodiment first restores the true reflectance of ground objects through reflectance correction and atmospheric correction, avoiding interference from non-ground object factors on feature point extraction, and ensuring the accuracy of feature point matching from the data source. Then, based on principal component analysis, the first principal component feature map is obtained. Through the global reference characteristics of the first principal component feature map, the spatial feature information of hyperspectral data is maximized, allowing sufficient effective feature points to be extracted even in texture-deficient bands. This avoids the registration failure caused by low contrast and insufficient feature points in texture-deficient bands that is common in traditional single feature matching methods. Finally, based on feature point matching and geometric transformation fitting, complex geometric deformations such as translation, rotation, scaling, and shearing caused by UAV pushbroom imaging can be accurately quantified and corrected. This fundamentally solves problems such as spatial misalignment between bands, discontinuous spectral curves, and RGB synthesis artifacts, thereby improving the accuracy of hyperspectral data.
[0040] In one embodiment of this application, for each band, using the first principal component feature map as a reference feature, feature point matching is performed between the surface reflectance data of that band and the first principal component feature map, including: The surface reflectance data of this band were normalized to obtain the hyperspectral image of this band. The hyperspectral image of this band is divided into multiple local blocks; For each local block, calculate the gray-level histogram of that local block, and calculate the mapping function of that local block based on the number of pixels at each gray level in the gray-level histogram; For each pixel in the hyperspectral image, four adjacent target local blocks corresponding to that pixel are identified; grayscale enhancement is performed on the pixel based on the mapping function of each target local block to obtain the grayscale enhancement result corresponding to each local block; the grayscale enhancement results corresponding to each local block are fused to obtain the grayscale enhancement result of the pixel. The enhanced hyperspectral image is obtained based on the grayscale enhancement results of each pixel in the hyperspectral image. Feature point matching is performed between the enhanced hyperspectral image of this band and the first principal component feature map.
[0041] In this embodiment, the hyperspectral data can be normalized band by band to 0-255 to obtain a hyperspectral image for each band. Adaptive histogram equalization is then applied to the hyperspectral image of each band to enhance features. Specifically, Min-Max normalization can be used, and the specific calculation formula is as follows:
[0042] in, This represents the normalized surface reflectance data. This represents the surface reflectance data before normalization. This represents the minimum value of the surface reflectance data. This represents the maximum value of the surface reflectance data.
[0043] Adaptive histogram equalization divides a hyperspectral image into several non-overlapping local tiles, and performs histogram equalization on each tile individually, which can greatly enhance local contrast. Assuming the original single-band image is... Divided into There are 8x8 tiles (default size). The specific steps for histogram equalization are as follows: (1) Calculate the grayscale histogram of the tile. :
[0044] in, Indicates the grayscale level.
[0045] (2) Calculate the transformation function (CDF) of the tile:
[0046] Where CDF(k) represents the proportion of all pixels with gray values ≤ k in the total pixels of the image of this tile. First, the frequency of pixels with gray values equal to i in the image is counted using hist[i], and then... Summing up the total number of pixels with grayscale values ≤ k, then dividing by the summation. The result is normalized to the interval [0,1] to obtain the probability value. When k=0, CDF(0) is the proportion of pixels with a gray value of 0. When k equals the maximum gray value (e.g., 255), CDF(k) must equal 1, because all pixels are included in the count. The entire CDF curve is monotonically non-decreasing because as k increases, the accumulated number of pixels will only increase or remain unchanged.
[0047] The mapping function is then obtained as follows:
[0048] Here, round() represents the rounding operation, which rounds down the integer part of the integer. Multiplying by 255 and then rounding down yields the target gray value of gray level k after histogram equalization enhancement. Essentially, it converts the cumulative distribution ratio of the original grayscale into the actual grayscale values for display / calculation of an 8-bit grayscale image.
[0049] (3) For each pixel, the grayscale enhancement results of the four adjacent tiles are fused using bilinear interpolation to obtain the grayscale enhancement result of that pixel. To eliminate the block effect:
[0050] in , The results show the grayscale enhancement of four adjacent tiles: top left, top right, bottom left, and bottom right.
[0051] Specifically, assuming the size of each local block is For any pixel (x, y) in a hyperspectral image, it must fall at the intersection of four adjacent local blocks: the local block at row i, column j (top left), the local block at row i, column j+1 (top right), the local block at row i+1, column j (bottom left), and the local block at row i+1, column j+1 (bottom right). These four adjacent local blocks are taken as the target local block. , , (Indicates rounding down), calculate the relative coordinates of the pixel in the grid cell composed of four adjacent local blocks ( , )for:
[0052]
[0053] relative coordinates ( , Normalization yields normalized coordinates. )for:
[0054]
[0055] Therefore, the weights of the four target local blocks are obtained:
[0056]
[0057]
[0058]
[0059] The grayscale value of the pixel is enhanced by the mapping function of the four target local blocks respectively, resulting in four grayscale enhancement results. The four grayscale enhancement results are then weighted and fused to obtain the final grayscale enhancement result for the pixel. :
[0060] For each band, the grayscale enhancement results of each pixel in the hyperspectral image of that band form the enhanced hyperspectral image. Image enhancement can improve the texture and contrast of the hyperspectral image and highlight features such as the edges and corners of ground objects in the low-texture band. By matching the enhanced hyperspectral image of that band with the first principal component feature map, feature point extraction can be more complete, thereby improving the accuracy of the matching point pair.
[0061] In one embodiment of this application, for each local block, a mapping function for that local block is calculated based on the number of pixels at each gray level in the gray-scale histogram, including: If there is no gray level among multiple gray levels with a number of pixels greater than the preset clipping threshold, then the mapping function of the local block is calculated based on the number of pixels of each gray level in the gray level histogram. If, among multiple gray levels, there is a gray level with a number of pixels greater than a preset clipping threshold, then the number of pixels exceeding the clipping threshold in each gray level is accumulated to obtain a clipped pixel accumulation value. The clipped pixel accumulation value is then redistributed to each gray level to obtain the adjusted number of pixels in each gray level. Based on the adjusted number of pixels in each gray level, the mapping function of the local block is calculated.
[0062] In this embodiment, performing histogram equalization on each local tile individually can enhance local contrast, but it produces severe blocking artifacts. To address this issue, this embodiment employs contrast-limited adaptive histogram equalization (CLAHE). By introducing a clipping threshold, excessive noise amplification can be prevented. Specifically, the calculation formula for the clipping threshold (clipLimit) is as follows:
[0063] in, =2.0, tileGridSize=(8,8), meaning that the total number of pixels for each tile is 8×8, which is 1 / 8 of the image's width and height.
[0064] First, a baseline value is calculated by dividing the total number of pixels in the tiles by 256. This represents the average number of pixels allocated to each gray level under an ideal uniform distribution. Then, the clipLimit parameter (here, 2.0) is multiplied by the baseline value to obtain the maximum allowed pixel threshold, representing the number of pixels per gray level. This threshold cannot exceed twice the number of pixels per gray level under an ideal uniform distribution. Finally, max(1, …) is used to ensure that clipLimit is at least 1, avoiding unreasonable thresholds less than 1 (a threshold less than 1 implies a stricter restriction than an ideal uniform distribution, leading to excessive image flattening). This formula limits the maximum pixel percentage of a single gray level during histogram equalization, preventing excessive pixel concentration at a particular gray level and avoiding overexposure, artifacts, or excessive amplification of noise in the image.
[0065] The portion exceeding the clipping threshold (clipLimit) is uniformly redistributed across all gray levels. The specific steps are as follows: (1) First, cut and count the excess part for each gray level. have:
[0066]
[0067] The total number of pixels that were cropped for:
[0068] in, Represents grayscale level The corresponding cut portion, Indicates the grayscale level after cropping The corresponding excess portion.
[0069] (2) Then calculate the average distribution, that is, the minimum number of pixels that each gray level can be allocated. :
[0070] Next, all gray levels are added back evenly to obtain the number of pixels per gray level after average distribution. :
[0071] Finally, process the remaining parts that are not divisible and calculate the number of remaining pixels. :
[0072] This Each pixel is allocated individually to obtain the adjusted number of pixels for each gray level. :
[0073] Finally, the mapping function of the local block is calculated based on the number of pixels at each gray level after adjustment. The adjusted gray-level histogram has a more uniform distribution. When performing gray-level mapping based on the mapping function it generates, the gray-level transition within the block is smoother, with no drastic brightness changes. The brightness and contrast differences between adjacent local blocks are greatly reduced, thereby weakening the "visual tortuosity" of the block boundary and eliminating block artifacts.
[0074] In one embodiment of this application, the surface reflectance data for each band includes surface reflectance data of multiple original coordinate points in that band; Specifically, for each band, the surface reflectance data of that band is transformed based on the transformation matrix to obtain the corrected hyperspectral data for that band, including: Calculate the inverse of the transformation matrix; Obtain multiple grid coordinate points in the preset correction grid; For each grid coordinate point, the original coordinate point corresponding to that grid coordinate point is calculated based on the inverse of the transformation matrix. If the original coordinate point is an integer coordinate point, the surface reflectance data of the original coordinate point is used as the surface reflectance data of the grid coordinate point. If the original coordinate point is a non-integer coordinate point, the surface reflectance data corresponding to the original coordinate point is estimated using an interpolation method, and the estimation result is used as the surface reflectance data of the grid coordinate point. If the original coordinate point exceeds the boundary range of the surface reflectance data for this band, the surface reflectance data of the nearest effective boundary coordinate point to the original coordinate point is used as the surface reflectance data of the grid coordinate point. By integrating the surface reflectance data of all grid coordinate points, the corrected hyperspectral data for this band is obtained.
[0075] In this embodiment, the first principal component feature map can be used as a reference image. The size of the corrected hyperspectral data (image) is preset to be exactly the same as that of the reference image, i.e., dst_width=ref_width and dst_height=ref_height. Based on the preset image size, the grid is divided to obtain multiple grid coordinate points in the preset correction grid, where the column coordinate x' ranges from 0 to dst_width-1 and the row coordinate y' ranges from 0 to dst_height-1.
[0076] Based on this, a reverse mapping can be used to traverse each grid coordinate point. The original coordinates of each grid point in the original surface reflectance data are calculated using the inverse of the transformation matrix. If the original coordinate point is an integer, its surface reflectance data is used as the surface reflectance data for that grid point. If the original coordinate point is a non-integer point, interpolation is used to estimate its corresponding surface reflectance data, and the estimated result is used as the surface reflectance data for that grid point. If the original coordinate point exceeds the boundary range of the surface reflectance data for that band, it is filled in using the nearest valid boundary coordinate point. Integrating the assigned values of all grid coordinate points according to their spatial location generates a two-dimensional data matrix that matches the correction grid, resulting in the corrected hyperspectral data for that band.
[0077] Specifically, in this embodiment, OpenCV's warpAffine is used in conjunction with bilinear interpolation to apply the estimated affine transformation matrix M to the original unenhanced surface reflectance data. The boundary is filled using OpenCV's BORDER_REPLICATE copy mode, thereby achieving pixel-level precision correction while preserving the original spectral radiance quantization characteristics.
[0078] The `warpAffine` function takes the original, unenhanced surface reflectance data, the affine transformation matrix M, and the dimensions of the corrected image. It automatically performs the inverse mapping calculation and outputs the corrected image data. The `warpAffine` function automatically handles the mapping relationship between the original image coordinates and the corrected image coordinates, eliminating the need to manually write a loop to iterate through the pixels. For each non-integer original coordinate point (... , The specific steps of bilinear interpolation are as follows: (1) Find the area surrounding ( , The four adjacent integer coordinate pixels of ) (2) Calculate the weighted average of these 4 pixels in the horizontal direction (x-axis) and vertical direction (y-axis) respectively, and finally obtain ( , The pixel value at ( ); where the weighting coefficient is determined by the original coordinates ( ) , The weight is determined by the distance between the value and the four integer coordinates; the closer the distance, the greater the weight.
[0079] As can be seen from the above, this embodiment resamples the original distorted data to the standardized correction grid by using the inverse matrix of the transformation matrix, which can effectively correct complex geometric deformations such as translation, rotation, scaling or shearing caused by UAV push-broom imaging, and fundamentally solve the problem of spatial misalignment between bands.
[0080] In one embodiment of this application, for each band, if the number of matching point pairs in each band is less than a preset threshold, the following operations are performed: Fourier transforms are performed on the enhanced hyperspectral image of this band and the first principal component feature map, respectively, to obtain the first spectrum and the second spectrum; Calculate the normalized cross-power spectrum based on the first and second spectra; The correlation function is obtained by performing an inverse Fourier transform on the normalized cross-power spectrum. Use the column coordinate corresponding to the peak of the absolute value of the relevant function as the column offset; The surface reflectance data for each band is shifted and corrected based on the column offset to obtain the corrected hyperspectral data.
[0081] In this embodiment, to avoid SIFT failure due to insufficient feature points (<5), the phase correlation method is used as a degradation scheme, which can quickly calculate the global translation offset. The specific process includes: (1) Perform 2D Fourier transform on the enhanced hyperspectral image and the first principal component feature map (reference image) respectively to obtain the spectrum F(u,v) and G(u,v); (2) Calculate the normalized cross-power spectrum:
[0082] in, Represents the normalized cross-power spectrum. for The complex conjugate form.
[0083] Cross-power spectrum can characterize the phase correlation between the spectra of hyperspectral images and the first principal component feature map. The phase directly corresponds to the translation offset of the two images, and the amplitude is always 1 (only the phase information is retained).
[0084] (3) To Performing the inverse Fourier transform yields the correlation function. ,right Taking the absolute value yields the absolute value image. The coordinates corresponding to the point with the largest pixel value (peak) in the absolute value image are the global translation offset of the hyperspectral image relative to the reference image, i.e., the hyperspectral image along... Axis translation Translation along the y-axis After y, it can overlap with the reference image to the greatest extent.
[0085] (4) Since the Fourier transform assumes the image is an infinitely repeating periodic array, the correlation function obtained by the inverse Fourier transform is... It also exhibits periodicity; if the actual translational offset... x represents a negative offset (a hyperspectral image shifted left relative to a reference image). Under periodicity, this will be displayed as a large positive offset. For example, if the image width is 500, a leftward shift of 10 pixels will be displayed as a rightward shift of 490 pixels, leading to an error in offset calculation.
[0086] To avoid the above problems, this embodiment adds a correction step: If ,but This is used to correct the offset. For example, if the image width is 500, the calculated value is... After correction That is, the hyperspectral image is shifted 10 column pixels to the left relative to the reference image to obtain the true offset.
[0087] (4) Since hyperspectral pushbroom imaging is a column scan, the column direction is the main direction of global translational distortion. In this embodiment, only the column offset is applied. Translation correction of hyperspectral images involves shifting each column of pixels in the hyperspectral image along the x-axis. Each pixel is aligned with the column position of the reference image, while the pixel position in the row direction remains unchanged. After translation, gaps will appear at the column boundaries (e.g., shifting to the left will result in a missing column on the right, and shifting to the right will result in a missing column on the left). The nearest neighbor interpolation mode is used to fill the gaps. That is, for the gap position, the nearest valid column pixel value is directly copied as the pixel value of the gap position.
[0088] In summary, this embodiment utilizes a fusion strategy of "CLAHE + PCA + SIFT / phase correlation method." By enhancing band features through CLAHE preprocessing and adding principal component analysis (PCA), it addresses the issue of insufficient feature points in low-contrast bands. The fusion of SIFT and phase correlation methods supports correction of complex geometric deformations while ensuring the reliability of translation error correction, avoiding the failure risk of single methods. This significantly improves registration robustness, eliminates spectral discontinuity distortion, and produces RGB composite images free of color spillover and artifacts, enhancing visual interpretation and quantitative analysis accuracy. Furthermore, this embodiment optimizes the offset calculation direction of the phase correlation method for the characteristics of UAV pushbroom hyperspectral data, adapting to the main misalignment types caused by UAV flight disturbances. It is suitable for hyperspectral data correction in various scenarios, automating the entire correction process from preprocessing to registration without manual intervention, thus improving correction efficiency and reducing operational complexity.
[0089] This application also provides an unmanned aerial vehicle (UAV) pushbroom-type hyperspectral automatic correction device, such as... Figure 3 As shown, the UAV pushbroom hyperspectral automatic correction device 20 may specifically include: a data acquisition module 21, a radiation correction module 22, a principal component extraction module 23, a feature point matching module 24, and a data correction module 25.
[0090] The data acquisition module 21 is used to acquire hyperspectral data; the hyperspectral data includes radiance values of multiple bands. The radiation correction module 22 is used to perform reflectivity correction and atmospheric correction on the radiance value of each band to obtain the surface reflectivity data of each band. Principal component extraction module 23 is used to perform principal component analysis on surface reflectance data of multiple bands to obtain the first principal component feature map; The feature point matching module 24 is used to use the first principal component feature map as a reference feature to match the surface reflectance data of each band with the first principal component feature map to obtain multiple matching point pairs for each band. The data correction module 25 is used to perform data fitting based on multiple matching point pairs in each band when the number of matching point pairs in each band is greater than or equal to a preset threshold, to obtain the transformation matrix of each band, and to convert the surface reflectance data of each band based on the transformation matrix to obtain the corrected hyperspectral data.
[0091] In one embodiment of this application, for each band, the feature point matching module 24 is specifically used for: The surface reflectance data of this band were normalized to obtain the hyperspectral image of this band. The hyperspectral image of this band is divided into multiple local blocks; For each local block, calculate the gray-level histogram of that local block, and calculate the mapping function of that local block based on the number of pixels at each gray level in the gray-level histogram; For each pixel in the hyperspectral image, four adjacent target local blocks corresponding to that pixel are identified; grayscale enhancement is performed on the pixel based on the mapping function of each target local block to obtain the grayscale enhancement result corresponding to each local block; the grayscale enhancement results corresponding to each local block are fused to obtain the grayscale enhancement result of the pixel. The enhanced hyperspectral image is obtained based on the grayscale enhancement results of each pixel in the hyperspectral image. Feature point matching is performed between the enhanced hyperspectral image of this band and the first principal component feature map.
[0092] In one embodiment of this application, for each local block, the feature point matching module 24 is further configured to: if there is no gray level with a number of pixels greater than a preset clipping threshold among the multiple gray levels, then calculate the mapping function of the local block based on the number of pixels of each gray level in the gray level histogram. If, among multiple gray levels, there is a gray level with a number of pixels greater than a preset clipping threshold, then the number of pixels exceeding the clipping threshold in each gray level is accumulated to obtain a clipped pixel accumulation value. The clipped pixel accumulation value is then redistributed to each gray level to obtain the adjusted number of pixels in each gray level. Based on the adjusted number of pixels in each gray level, the mapping function of the local block is calculated.
[0093] In one embodiment of this application, each matching point pair includes a reference feature point and a corresponding query feature point. The reference feature point is the coordinates of a feature point in the first principal component feature map, and the query feature point is the coordinates of a feature point in the surface reflectance data for each band. The feature point matching module 24 is further configured to: An index structure is constructed based on multiple reference feature points; For each query feature point, the nearest and second nearest candidate matching points are searched in the index structure. The ratio between the nearest neighbor distance and the second nearest neighbor distance is calculated. From multiple nearest candidate matching points, the candidate matching points whose ratio is less than a preset threshold are selected as the optimal matching points. The query feature point and the corresponding optimal matching point are then considered as the optimal matching point pair. Here, the nearest neighbor distance is the distance between the nearest candidate matching point and the query feature point, and the second nearest neighbor distance is the distance between the second nearest candidate matching point and the query feature point. Data fitting is performed based on multiple optimal matching point pairs for each band.
[0094] In one embodiment of this application, the feature point matching module 24 is specifically used for: The surface reflectance data of each band is matched with the first principal component feature map using scale-invariant feature transformation algorithm, binary feature extraction algorithm or corner feature extraction algorithm.
[0095] In one embodiment of this application, the surface reflectance data for each band includes surface reflectance data of multiple original coordinate points in that band; the data correction module 25 is specifically used for: Calculate the inverse of the transformation matrix; Obtain multiple grid coordinate points in the preset correction grid; For each grid coordinate point, the original coordinate point corresponding to that grid coordinate point is calculated based on the inverse of the transformation matrix. If the original coordinate point is an integer coordinate point, the surface reflectance data of the original coordinate point is used as the surface reflectance data of the grid coordinate point. If the original coordinate point is a non-integer coordinate point, the surface reflectance data corresponding to the original coordinate point is estimated using an interpolation method, and the estimation result is used as the surface reflectance data of the grid coordinate point. If the original coordinate point exceeds the boundary range of the surface reflectance data for this band, the surface reflectance data of the nearest effective boundary coordinate point to the original coordinate point is used as the surface reflectance data of the grid coordinate point. By integrating the surface reflectance data of all grid coordinate points, the corrected hyperspectral data for this band is obtained.
[0096] In one embodiment of this application, for each band, if the number of matching point pairs in each band is less than a preset threshold, the data correction module 25 is further configured to: Fourier transforms are performed on the enhanced hyperspectral image of this band and the first principal component feature map, respectively, to obtain the first spectrum and the second spectrum; Calculate the normalized cross-power spectrum based on the first and second spectra; The correlation function is obtained by performing an inverse Fourier transform on the normalized cross-power spectrum. Use the column coordinate corresponding to the peak of the absolute value of the relevant function as the column offset; The surface reflectance data for each band is shifted and corrected based on the column offset to obtain the corrected hyperspectral data.
[0097] The apparatus in this application embodiment can execute the method provided in this application embodiment, and the implementation principle is similar. The actions performed by each module in the apparatus of each embodiment of this application correspond to the steps in the method of each embodiment of this application. For detailed functional descriptions of each module of the apparatus, please refer to the descriptions in the corresponding methods shown above, which will not be repeated here.
[0098] Figure 4 A schematic diagram of the structure of an electronic device to which this application embodiment applies is shown, such as... Figure 4 As shown, the electronic device can be used to implement the methods provided in any embodiment of this application.
[0099] like Figure 4 As shown, the electronic device 300 may primarily include at least one processor 301. Figure 4 The diagram shows components such as a memory 302, a communication module 303, and an input / output interface 304. Optionally, these components can be connected and communicate with each other via a bus 305. It should be noted that... Figure 4The structure of the electronic device 300 shown is merely illustrative and does not constitute a limitation on the electronic devices to which the methods provided in the embodiments of this application are applicable.
[0100] The memory 302 can be used to store operating systems and applications, etc. The applications can include computer programs that implement the methods shown in the embodiments of this application when invoked by the processor 301, and can also include programs for implementing other functions or services. The memory 302 can be ROM (Read Only Memory) or other types of static storage devices that can store static information and instructions, RAM (Random Access Memory) or other types of dynamic storage devices that can store information and computer programs, or it can be EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer, but is not limited thereto.
[0101] Processor 301 is connected to memory 302 via bus 305 and implements corresponding functions by calling the application programs stored in memory 302. Processor 301 can be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 301 can also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc.
[0102] Electronic device 300 can connect to a network via communication module 303 (which may include, but is not limited to, components such as a network interface) to communicate with other devices (such as user terminals or servers) through the network and achieve data interaction, such as sending data to or receiving data from other devices. Communication module 303 may include wired network interfaces and / or wireless network interfaces, meaning the communication module may include at least one of wired or wireless communication modules.
[0103] The electronic device 300 can connect to necessary input / output devices, such as a keyboard or display device, via the input / output interface 304. The electronic device 300 itself may have a display device, and other display devices can also be connected externally via the interface 304. Optionally, a storage device, such as a hard drive, can also be connected via the interface 304, allowing data from the electronic device 300 to be stored, read, or transferred to the memory 302. It is understood that the input / output interface 304 can be a wired or wireless interface. Depending on the specific application scenario, the device connected to the input / output interface 304 can be an integral part of the electronic device 300 or an external device connected to the electronic device 300 when needed.
[0104] The bus 305 used to connect the components may include a path for transmitting information between the components. The bus 305 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Depending on its function, the bus 305 may be divided into an address bus, a data bus, a control bus, etc.
[0105] Optionally, for the solution provided in the embodiments of this application, the memory 302 can be used to store a computer program that executes the solution of this application, and the processor 301 runs the computer program. When the processor 301 runs the computer program, it implements the operation of the method or apparatus provided in the embodiments of this application.
[0106] Based on the same principle as the method provided in the embodiments of this application, the embodiments of this application provide a computer-readable storage medium storing a computer program, which, when executed by a processor, can implement the corresponding content of the aforementioned method embodiments.
[0107] This application also provides a computer program product, which includes a computer program that, when executed by a processor, can implement the corresponding content of the aforementioned method embodiments.
[0108] It should be noted that the terms "first," "second," "third," "fourth," "1," "2," etc. (if present) in the specification, claims, and accompanying drawings of this application 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 this application described herein can be implemented in a sequence other than that shown in the figures or text.
[0109] In the embodiments of this application, the terms "module" or "unit" refer to a computer program or part of a computer program that has a predetermined function and works with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more modules or units. Furthermore, each module or unit can be part of an overall module or unit that includes the functionality of that module or unit.
[0110] It should be understood that although arrows indicate various operation steps in the flowcharts of this application's embodiments, the order in which these steps are implemented is not limited to the order indicated by the arrows. Unless explicitly stated herein, in some implementation scenarios of this application's embodiments, the implementation steps in each flowchart can be executed in other orders as required. Furthermore, some or all steps in each flowchart, based on the actual implementation scenario, may include multiple sub-steps or multiple stages. Some or all of these sub-steps or stages can be executed at the same time, and each sub-step or stage can also be executed at different times. In scenarios where execution times differ, the execution order of these sub-steps or stages can be flexibly configured according to requirements, and this application's embodiments do not limit this.
[0111] The above description is only an optional implementation method for some implementation scenarios of this application. It should be noted that for those skilled in the art, other similar implementation methods based on the technical concept of this application without departing from the technical concept of this application also fall within the protection scope of the embodiments of this application.
Claims
1. A pushbroom-based automatic hyperspectral calibration method for unmanned aerial vehicles (UAVs), characterized in that, include: Acquire hyperspectral data; the hyperspectral data includes radiance values in multiple bands; Reflectance correction and atmospheric correction are performed on the radiance values of each band to obtain the surface reflectance data for each band; Principal component analysis was performed on surface reflectance data from multiple bands to obtain the first principal component feature map; Using the first principal component feature map as a reference feature, the surface reflectance data of each band is matched with the first principal component feature map to obtain multiple matching point pairs for each band. If the number of matching point pairs in each band is greater than or equal to a preset threshold, data fitting is performed based on multiple matching point pairs in each band to obtain a transformation matrix for each band. The surface reflectance data of each band is then converted based on the transformation matrix to obtain corrected hyperspectral data.
2. The UAV pushbroom-type hyperspectral automatic correction method as described in claim 1, characterized in that, For each band, using the first principal component feature map as a reference feature, feature point matching is performed between the surface reflectance data of that band and the first principal component feature map, including: The surface reflectance data of this band were normalized to obtain the hyperspectral image of this band. The hyperspectral image of this band is divided into multiple local blocks; For each local block, calculate the gray-level histogram of the local block, and calculate the mapping function of the local block based on the number of pixels at each gray level in the gray-level histogram; For each pixel in the hyperspectral image, four adjacent target local blocks corresponding to that pixel are identified; grayscale enhancement is performed on the pixel based on the mapping function of each target local block to obtain the grayscale enhancement result corresponding to each local block; the grayscale enhancement results corresponding to each local block are fused to obtain the grayscale enhancement result of the pixel. The enhanced hyperspectral image is obtained based on the grayscale enhancement results of each pixel in the hyperspectral image. The enhanced hyperspectral image of this band is matched with the feature points of the first principal component feature map.
3. The UAV pushbroom-type hyperspectral automatic correction method as described in claim 2, characterized in that, For each local block, a mapping function for that local block is calculated based on the number of pixels at each gray level in the gray-level histogram, including: If there is no gray level among multiple gray levels with a number of pixels greater than a preset clipping threshold, then the mapping function of the local block is calculated based on the number of pixels of each gray level in the gray level histogram. If, among multiple gray levels, there is a gray level with a number of pixels greater than a preset clipping threshold, then the number of pixels exceeding the clipping threshold in each gray level is accumulated to obtain a clipped pixel accumulation value. The clipped pixel accumulation value is then redistributed to each gray level to obtain the adjusted number of pixels in each gray level. Based on the adjusted number of pixels in each gray level, the mapping function of the local block is calculated.
4. The UAV pushbroom-type hyperspectral automatic correction method as described in claim 2, characterized in that, Each pair of matching points includes a reference feature point and a corresponding query feature point. The reference feature point is the coordinates of a feature point in the first principal component feature map, and the query feature point is the coordinates of a feature point in the surface reflectance data for each band. The data fitting based on multiple matching point pairs for each band includes: An index structure is constructed based on multiple reference feature points; For each query feature point, the nearest candidate matching point and the second nearest candidate matching point are searched in the index structure; the ratio between the nearest neighbor distance and the second nearest neighbor distance is calculated, and from multiple nearest candidate matching points, the candidate matching points whose corresponding ratio is less than a preset ratio threshold are selected as the optimal matching points, and the query feature point and the corresponding optimal matching point are selected as the optimal matching point pair; wherein, the nearest neighbor distance is the distance between the nearest candidate matching point and the query feature point, and the second nearest neighbor distance is the distance between the second nearest candidate matching point and the query feature point; Data fitting is performed based on multiple optimal matching point pairs for each band.
5. The UAV pushbroom-type hyperspectral automatic correction method as described in claim 1, characterized in that, The surface reflectance data of each band is matched with the first principal component feature map using a scale-invariant feature transformation algorithm, a binary feature extraction algorithm, or a corner feature extraction algorithm.
6. The UAV pushbroom-type hyperspectral automatic correction method as described in claim 1, characterized in that, The surface reflectance data for each band includes the surface reflectance data of multiple raw coordinate points in that band; Specifically, for each band, the surface reflectance data of that band is transformed based on the transformation matrix to obtain the corrected hyperspectral data for that band, including: Calculate the inverse of the transformation matrix; Obtain multiple grid coordinate points in the preset correction grid; For each grid coordinate point, based on the inverse of the transformation matrix, the original coordinate point corresponding to that grid coordinate point is calculated. If the original coordinate point is an integer coordinate point, the surface reflectance data of the original coordinate point is used as the surface reflectance data of the grid coordinate point. If the original coordinate point is a non-integer coordinate point, the surface reflectance data corresponding to the original coordinate point is estimated using an interpolation method, and the estimation result is used as the surface reflectance data of the grid coordinate point. If the original coordinate point exceeds the boundary range of the surface reflectance data for that band, the surface reflectance data of the nearest effective boundary coordinate point to the original coordinate point is used as the surface reflectance data of the grid coordinate point. By integrating the surface reflectance data of all grid coordinate points, the corrected hyperspectral data for this band is obtained.
7. The UAV pushbroom-type hyperspectral automatic correction method as described in claim 2, characterized in that, For each band, if the number of matching point pairs in each band is less than a preset threshold, then perform the following operations: The enhanced hyperspectral image of this band and the first principal component feature map are subjected to Fourier transform to obtain the first spectrum and the second spectrum, respectively. Calculate the normalized cross-power spectrum based on the first spectrum and the second spectrum; The normalized cross-power spectrum is subjected to inverse Fourier transform to obtain the correlation function; The column coordinate corresponding to the peak value of the absolute value of the relevant function is used as the column offset; Based on the column offset, the surface reflectance data of each band are shifted and corrected to obtain the corrected hyperspectral data.
8. A drone pushbroom-type hyperspectral automatic correction device, characterized in that, include: The data acquisition module is used to acquire hyperspectral data; the hyperspectral data includes radiance values of multiple bands. The radiation correction module is used to perform reflectivity correction and atmospheric correction on the radiance values of each band to obtain the surface reflectivity data for each band. The principal component extraction module is used to perform principal component analysis on surface reflectance data of multiple bands to obtain the first principal component feature map. The feature point matching module is used to use the first principal component feature map as a reference feature to match the surface reflectance data of each band with the first principal component feature map to obtain multiple matching point pairs for each band. The data correction module is used to perform data fitting based on multiple matching point pairs in each band when the number of matching point pairs in each band is greater than or equal to a preset threshold, to obtain a transformation matrix for each band, and to convert the surface reflectance data of each band based on the transformation matrix to obtain corrected hyperspectral data.
9. An electronic device, characterized in that, The electronic device includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the UAV pushbroom hyperspectral automatic correction method according to any one of claims 1 to 7 when running the computer program.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the UAV pushbroom-type hyperspectral automatic correction method according to any one of claims 1 to 7.