Single-flight UAV multispectral image correction method, system, equipment and medium

Abstract

The present invention discloses a method, system, device, and medium for correcting a multispectral image of a single UAV flight, relating to the technical field of multispectral correction for UAVs. The method comprises: estimating the initial noise level of each band image in the multispectral image of a single UAV flight; performing noise reduction processing on the multispectral image of the single UAV flight in an iterative manner based on the low-rank property of the image space and a small sample statistical theory method and the initial noise level of each band image, thereby obtaining a noise-reduced multispectral image of the single UAV flight; and correcting the noise-reduced multispectral image of the single UAV flight using a bilinear interpolation method based on solar irradiance, thereby obtaining a corrected noise-reduced multispectral image of the single UAV flight. The present invention can solve the problem of radiation inconsistency in multispectral imaging of a single UAV flight.

Description

Technical Field

[0001] The present invention relates to the technical field of multispectral correction of unmanned aerial vehicles (UAVs), and in particular to a method, system, equipment and medium for correcting multispectral images of a single-flight UAV based on solar irradiance. Background Art

[0002] Multispectral imagery is currently widely used in agriculture, forestry, ecology, environmental protection, and other fields, effectively accelerating crop breeding and becoming a key research area in near-earth remote sensing. Using drone multispectral sensors to obtain radiometric information about ground objects, and using radiometric and spectral correction methods to determine the reflectivity of each band, quantitative remote sensing is achieved. Images captured by drone multispectral sensors are below cloud cover and unobstructed by clouds, but are susceptible to radiometric influences. With the continuous improvement of the endurance of industrial drones, single drone missions can now last up to 40 minutes. Even in clear, cloudless weather, radiometric fluctuations can still occur.

[0003] Multispectral calibration for drones often uses an alternative calibration method. This involves measuring the reflectance spectral curves of three diffuse reflectance reference objects with a ground spectrometer before the experiment. This establishes a linear relationship between the reflectance of the ground objects and the DN (digital number) value of the image. The spectral correction coefficients established for the diffuse reflectance reference objects are then applied to all images from a single drone flight, resulting in a corrected multispectral image. This converts the DN value of a single drone multispectral image into reflectance. In recent years, with the continuous improvement of drone endurance, the radiometric conditions during a single drone flight have changed. Establishing a multispectral correction curve using exposure parameters set before takeoff is unable to address this radiometric variation.

[0004] However, due to factors such as the complex farmland experimental environment and sensor hardware performance, remote sensing of farmland and near-surface multispectral images contain significant noise, which limits and hinders agricultural application research. Currently, various methods have been proposed for multispectral image denoising, mainly falling into the following three categories: 1. Image-space-based denoising methods treat each band of a multispectral image as a two-dimensional grayscale image and perform independent denoising; 2. Spectral low-rank-based denoising methods exploit inter-band correlation to reduce image noise; and 3. Denoising methods combine image spatial correlation with spectral low-rank.

[0005] The paper "He W, Yao Q, Li C, et al. Non-local Meets Global: An Integrated Paradigm for Hyperspectral Image Restoration [J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020" projects multispectral images into a low-level space, searches for non-locally similar blocks on the average spectral image, and employs WNNM for denoising. However, this method suffers from the fact that it fails to account for the varying noise levels across different bands, resulting in loss of detail in some bands after denoising. Furthermore, this method requires prior knowledge, resulting in limited optimization effectiveness. Xidian University, in its patent application for "Multispectral Remote Sensing Image Denoising Method Based on Four-Dimensional Block Matching Filtering," proposes a method that sorts multispectral images based on their spectral signal-to-noise ratio and processes bands with different noise levels differently. However, this method fails to consider the influence of inter-spectral correlation (i.e., low rank) when grouping bands, resulting in the separation of related bands into different groups, which compromises the denoising effect.

[0006] At present, many methods have been proposed for solar radiation correction of multispectral images, mainly in the following three categories: ① Based on the automatic exposure mode setting of the object being measured, the camera aperture and sensitivity are adjusted in real time; ② Based on the auxiliary radiometric sensor in fixed aperture mode, the exposure time is adjusted in a timely manner; ③ Combining the ground reflectivity of the ground diffuse reflector with the empirical formula of the image DN to correct the image spatial correlation and spectral space.

[0007] The papers "Sungji B, Jaehyung Y, Lei W, et al. Experimental analysis of sandgrain size mapping using UAV remote sensing [J]. Remote Sensing Letters, 2019." and "Liang W, Haiyan C, Jiangpeng Z, et al. Grain yield prediction of rice using multi-temporal UAV-based RGB and multispectral images and model transfer - a case study of small farmlands in the South of China [J]. Agricultural and Forest Meteorology, 2020." study calibrate imaging spectral reflectance data generated after spectral reconstruction using an atmospheric radiometric transfer model to reduce errors. However, this method has the disadvantage of not considering the radiometric inconsistencies during a single UAV flight. Furthermore, the method requires prior knowledge, resulting in limited optimization effects. In its patent application, Hohai University aligns single-flight UAV images with reference images to eliminate image distortion caused by strong reflectivity from ground objects and drastic changes in radiometric intensity. However, the shortcoming of this method is that the correlation between spectra, that is, low rank, is not considered when grouping the multispectral bands of drones, which causes the related bands to be separated into different groups, affecting the correction effect, and is highly dependent on the reference image.

[0008] Various multispectral image processing methods, including the two methods mentioned above, often use empirical formulas to correct the multispectral images of a single UAV during spectral correction. However, these methods are insensitive to changes in radiometric intensity and cannot solve the radiation inconsistency problem in the multispectral imaging of a single UAV. Summary of the Invention

[0009] The purpose of the present invention is to provide a single-flight UAV multispectral image correction method, system, equipment and medium, which can solve the radiation inconsistency problem in single-flight UAV multispectral imaging.

[0010] To achieve the above object, the present invention provides the following solutions:

[0011] In a first aspect, the present invention provides a method for correcting multispectral images of a single UAV, comprising:

[0012] Estimate the initial noise level of each band image in a single UAV multispectral image;

[0013] Based on the low-rank property of image space, the denoising method of single-flight UAV multispectral image is performed using a small sample statistical theory method and the initial noise level of each band image. The denoised single-flight UAV multispectral image is obtained.

[0014] Based on solar radiation, the bilinear interpolation method is used to correct the denoised single-flight UAV multispectral image to obtain the corrected denoised single-flight UAV multispectral image.

[0015] In a second aspect, the present invention provides a single-flight UAV multispectral image correction system, comprising:

[0016] The initial noise level estimation module is used to estimate the initial noise level of each band image in a single UAV multispectral image;

[0017] The denoising module is used to perform denoising on the multispectral image of a single UAV flight in an iterative manner based on the low rank of the image space, the small sample statistical theory method and the initial noise level of each band image, and obtain the denoised multispectral image of the single UAV flight;

[0018] The correction module is used to correct the noise-reduced single-flight UAV multispectral image based on solar radiation and adopt a bilinear interpolation method to obtain a corrected noise-reduced single-flight UAV multispectral image.

[0019] In a third aspect, the present invention provides an electronic device comprising a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to perform the single-flight UAV multispectral image correction method according to the first aspect.

[0020] In a fourth aspect, the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the single-flight UAV multispectral image correction method described in the first aspect.

[0021] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0022] ① Using a small sample statistical theory method (F test), the non-local similarity of image blocks is calculated. Compared with the Euclidean distance, this method is more adaptable to noise fluctuations and can eliminate the impact of brightness differences between similar image blocks on noise reduction. ② In combination with the low rank property of the spectrum, the noise level and noise reduction change level are compared in each band to eliminate the problem of missing texture details in some bands caused by global noise reduction. ③ Based on solar irradiance, the relative position of each band image is corrected to address the radiation inconsistency problem in single-flight multispectral imaging. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0024] Figure 1 A schematic diagram of a flow chart of a single-flight UAV multispectral image correction method provided by an embodiment of the present invention;

[0025] Figure 2 A schematic diagram of a flow chart of a single-flight UAV multispectral image correction method based on solar irradiance provided by an embodiment of the present invention;

[0026] Figure 3 A schematic diagram illustrating the calculation principle of the bilinear spatial radiation correction coefficient provided by an embodiment of the present invention;

[0027] Figure 4 A schematic diagram of the solar radiation variation and radiation consistency correction effect of a single UAV flight provided by an embodiment of the present invention;

[0028] Figure 5 A schematic diagram of the structure of a single-flight UAV multispectral image correction system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0029] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0030] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0031] Example 1

[0032] like Figure 1 and Figure 2 As shown, an embodiment of the present invention provides a single-flight UAV multispectral image correction method based on solar irradiance, which includes the following steps.

[0033] Step 100: Estimate the initial noise level of each band image in the multispectral image of a single UAV flight.

[0034] The mathematical model of a single UAV multispectral image is:

[0035] in, is a real noisy image, is the original image without noise, is noise (usually Gaussian noise).

[0036] In this embodiment of the present invention, step 100 specifically includes:

[0037] Obtain a single-flight UAV multispectral image, and then use the local variance method to obtain the initial noise level of each band image in the single-flight UAV multispectral image, where the i-th dimension band image has a mean value of u i , with variance σ i 2 The noise n i .

[0038] Step 200: Based on the low rank of the image space, according to the small sample statistical theory method and the initial noise level of each band image, an iterative method is used to perform noise reduction processing on the single-flight UAV multispectral image to obtain a noise-reduced single-flight UAV multispectral image.

[0039] In this embodiment of the present invention, step 200 specifically includes:

[0040] (1) Perform dimensionality reduction on the multispectral image of a single UAV flight to obtain a reduced-dimensional image.

[0041] (2) The dimensionality-reduced image is divided into several blocks to obtain a set of image blocks containing multiple blocks. Based on the low-rank property of the image space and the small sample statistical theory method, the non-local similar image blocks corresponding to each image block are determined.

[0042] (3) A second two-dimensional tensor is obtained according to each image block and the corresponding non-local similar image block, and the second two-dimensional tensor is denoised using the WNNM method to obtain a denoised multispectral image.

[0043] (4) Calculate the current noise level of each band image in the denoised multispectral image.

[0044] (5) According to the current noise level and initial noise level of each band image, an iterative method is used to determine the noise-reduced single-flight UAV multispectral image.

[0045] Furthermore, the multispectral image of a single UAV is subjected to dimensionality reduction to obtain a reduced-dimensional image, specifically comprising: converting each band image in the multispectral image of the single UAV into a column vector to obtain a first two-dimensional tensor; performing SVD decomposition on the first two-dimensional tensor, selecting the column vectors of the first K dimensions whose contribution rate exceeds a first set value to obtain a K-dimensional subspace; and projecting the multispectral image of the single UAV into the K-dimensional subspace to obtain a reduced-dimensional image.

[0046] Furthermore, based on the low rank of the image space, according to the small sample statistical theory method, the non-local similar image block corresponding to each image block is determined, specifically including: based on the low rank of the image space, calculating the Euclidean distance of the center points of adjacent image blocks, and according to the Euclidean distance of the center points of adjacent image blocks, using the K nearest neighbor algorithm to search and obtain the M nearest neighboring image blocks of each image block; according to the small sample statistical theory method, calculating the non-local statistical similarity between the target image block and any neighboring image block corresponding to the target image block; the target image block is any image block in the image block collection; the neighboring image blocks that meet the set conditions are determined as non-local similar image blocks of the target image block, and then the non-local similar image blocks corresponding to each image block are obtained; the set condition is the condition that the non-local statistical similarity is less than a second set value.

[0047] Furthermore, based on the current noise level and initial noise level of each band image, an iterative method is used to determine the denoised single-flight UAV multispectral image, specifically including: performing the first operation on any band image to obtain the output result corresponding to each band image, and obtaining the denoised single-flight UAV multispectral image based on the output result.

[0048] The first operation is: according to the current noise level and the initial noise level of the band image, the noise change value of the current iteration number is calculated; the current noise level is the noise level corresponding to the current iteration number; the initial noise level is the noise level corresponding to the previous iteration number; when the set constraints are met, the denoised band image corresponding to the current iteration number is output; the set constraints are that the noise change value is less than the minimum threshold or the current iteration number is the total iteration number; when the set constraints are not met, the band image that does not meet the set constraints is determined as a single-flight UAV multispectral image, the number of iterations is increased by 1, and the step of reducing the dimensionality of the single-flight UAV multispectral image is returned to obtain the reduced-dimensionality image.

[0049] In an embodiment of the present invention, the above-mentioned iterative noise reduction process is:

[0050] 1.1 Projection into low-dimensional subspace.

[0051] First, input a single-flight UAV multispectral image I 0(H×W×N, where H is the length, W is the width, and N is the number of bands. In the embodiments of the present invention, N = 25); secondly, convert each band image in the multi-spectral image I0 of a single flight of the unmanned aerial vehicle into a column vector, and construct a two-dimensional tensor T 0 ((H×W)×N, where (H×W) is a row vector and N is a column vector); then perform SVD decomposition on the two-dimensional tensor T 0 , select the column vectors of the first K dimensions with a contribution rate exceeding 95% (K << N) to obtain a K-dimensional subspace; then project the multi-spectral image I of a single flight of the unmanned aerial vehicle 0 into the K-dimensional subspace to obtain a dimensionality-reduced image I 1 , and obtain an orthogonal matrix E0. The dimensionality reduction formula and the orthogonal matrix calculation formula are as follows:

[0052] I 0 = I 1 ×E0.

[0053] s.t. E0 T [[ID=2,0]]E0 = I.

[0054] 1.2 Calculate the non-local statistical similarity of image patches.

[0055] First, determine that the non-local similarity patch size is 5×5; secondly, according to the given non-local similarity patch size, divide the dimensionality-reduced image I 1 into an image patch collection with a size of (H / 5×W / 5×K); for boundary points, when calculating the non-local statistical similarity, supplement data points with the intensity of each band being 0 for easy calculation; then, based on the low-rank property of the image space, calculate the Euclidean distance between the central points of adjacent image patches, and search for the M nearest neighbor image patches of each image patch using the K-nearest neighbor algorithm; then calculate the non-local statistical similarity, specifically as follows:

[0056] ① Take the image patch p i and the collection of M neighbor image patches {p j}, j = 1, 2, 3... M; perform average processing on the images of K bands to obtain an average spectral image patch For all pixel points of the average spectral image patch<, calculate the average spectral intensity value δ, and subtract the average spectral intensity value δ from all pixel points of the average spectral image patch to obtain an average spectral image with a mean spectral intensity of 0 ② Take the neighbor image patch p j , calculate the average spectral image and perform the same operation as to obtain an average spectral image with a mean spectral intensity of 0 ③ For the average spectral image and the average spectral image Each pixel is taken as a sample, and the small sample statistical theory method (F test) is used to check whether there is a significant difference between the two image blocks. If not, the two image blocks are judged to be similar.

[0057] 1.3 Noise reduction based on low-rank constraints.

[0058] First, the calculated non-local similarity image blocks {p l}∈{p j} and image block p i Combine and get the second two-dimensional tensor T 1 , the size is ((25×K)×L), where L is the number of similar image blocks, that is, {p l}, representing the column vector; 25×K represents all the band data of 25 pixels in an image block, representing the row vector. Then for the two-dimensional tensor T 1 The WNNM method is used to reduce noise and obtain the multispectral image I after noise reduction. 2 .

[0059] 1.4 Dimensional reconstruction.

[0060] For the denoised multispectral image I 2 For each band, recalculate each noise level to get the variance σ i ’2 Noise n i ’ , and calculate the iterative noise change, the formula is as follows:

[0061]

[0062] Among them, γ is the scale factor, ε1 and ε2 are the minimum thresholds. ′ i ′ is the noise level of the previous iteration process, σ″ in the first iteration i =σ i Δσ1 represents the noise level of the image after denoising, and Δσ2 represents the change in noise level during the previous and subsequent iterations. When Δσ1 or Δσ2 approaches 0, it indicates that the spectral band image has completed denoising and will not enter the next iteration.

[0063] 1.5 update iteration.

[0064] Update the multispectral image to I 0’ (H×W×(NQ)). Q is the band that will not be subjected to the next iteration of noise reduction. Stop when the number of iterations is met.

[0065] Step 300: Based on the solar radiation, a bilinear interpolation method is used to correct the noise-reduced single-flight UAV multispectral image to obtain a corrected noise-reduced single-flight UAV multispectral image.

[0066] In this embodiment of the present invention, step 300 specifically includes:

[0067] Firstly, based on solar irradiance, the bilinear spatial radiation correction coefficient is calculated according to the bilinear interpolation method and the relative position between the single-flight UAV multispectral image and the starting point of the UAV flight. Secondly, the image bands in the denoised single-flight UAV multispectral image are corrected according to the bilinear spatial radiation correction coefficient to obtain the corrected and denoised single-flight UAV multispectral image.

[0068] An example is: Figure 3 As shown, the correction process is:

[0069] (1) Input the denoised single-flight UAV multispectral image I.

[0070] (2) Based on the four known calibration plates, calculate the initial calibration coefficients, which are β 00 , β 01 , β 10 , β 11 .

[0071] (3) According to the relative positions of the single-flight UAV multispectral image and the starting point of the UAV flight sortie as tx and ty, the correction coefficient Δβ of the UAV multispectral image at position P is obtained by the bilinear interpolation method. (x,y) . And the correction coefficient Δβ (x,y) and β (x,y) Radiation correction coefficient β of nearby adjacent points (x-1,y) , β (x+1,y) , β (x,y-1) and β (x,y+1) Perform weighted averaging to obtain the bilinear spatial radiation correction coefficient of the multispectral image of the UAV at position P. The calculation formula of the bilinear spatial radiation correction coefficient is:

[0072] Δβ (x,y) =β 00 (1-tx)(1-ty)+β 10 tx(1-ty)+β 01 (1-tx)ty+

[0073] β 11 txty.

[0074]

[0075] (4) Using the bilinear spatial radiation correction coefficient β (x,y)As the reference image, calculate the bilinear spatial radiation correction coefficients corresponding to the adjacent band images according to the method described in step (3), and use the calculated bilinear spatial radiation correction coefficients to correct the adjacent band images. This is repeated until all band images are corrected. When the illumination difference between the two band images is less than 1%, the solar radiation difference between the two band images is considered negligible and no correction is performed.

[0076] An example:

[0077] The image was collected in the evening when the sun's radiation changed dramatically and the illumination dropped rapidly. The illumination changes are as follows: Figure 4 The illumination shows an overall downward trend, with illumination troughs appearing near the 60th, 120th, and 200th image acquisition points. This is due to clouds blocking the sun, especially at the 100th image acquisition point where the illumination drops sharply.

[0078] This example uses a small sample statistical theory method (F test) to calculate the non-local similarity of image blocks. Compared with the Euclidean distance, this method is more adaptable to noise fluctuations. It also eliminates the influence of brightness differences between similar image blocks on noise reduction.

[0079] This example combines spectral low rank with the noise level and noise reduction change level by band, eliminating the problem of missing texture details in some bands of images caused by global noise reduction.

[0080] This example corrects the relative positions of images in each band based on solar radiation to address the radiation inconsistency problem in multispectral imaging of a single drone flight.

[0081] After the above processing, we get Figure 4 As shown in (b), the single-flight UAV multispectral image after correction and noise reduction is obtained.

[0082] Example 2

[0083] In order to execute the method corresponding to the above-mentioned embodiment 1 and achieve the corresponding functions and technical effects, a single-flight UAV multispectral image correction system is provided below.

[0084] like Figure 5 As shown in FIG, the single-flight UAV multispectral image correction system specifically includes:

[0085] The initial noise level estimation module 1 is used to estimate the initial noise level of each band image in the multispectral image of a single UAV flight.

[0086] Denoising module 2 is used to perform denoising on the multispectral image of a single UAV flight in an iterative manner based on the low rank of the image space, according to the small sample statistical theory method and the initial noise level of each band image, to obtain the denoised multispectral image of a single UAV flight.

[0087] The correction module 3 is used to correct the noise-reduced single-flight UAV multispectral image based on solar radiation using a bilinear interpolation method to obtain a corrected noise-reduced single-flight UAV multispectral image.

[0088] Example 3

[0089] An embodiment of the present invention provides an electronic device including a memory and a processor. The memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to perform the single-flight UAV multispectral image correction method of embodiment 1.

[0090] Optionally, the above-mentioned electronic device may be a server.

[0091] In addition, an embodiment of the present invention further provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the single-flight UAV multispectral image correction method of embodiment 1.

[0092] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0093] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only intended to help understand the method and core concept of the present invention. At the same time, those skilled in the art will find that the specific implementation methods and application scopes may vary based on the concept of the present invention. In summary, the contents of this specification should not be construed as limiting the present invention.

Claims

1. A single-flight UAV multispectral image correction method, characterized in that: include: Estimate the initial noise level of each band image in a single UAV multispectral image; Based on the low rank of the image space, according to the small sample statistical theory method and the initial noise level of each band image, an iterative method is used to perform denoising on the multispectral image of a single flight UAV to obtain a denoised multispectral image of a single flight UAV, specifically including: reducing the dimension of the multispectral image of a single flight UAV to obtain a reduced dimension image; dividing the reduced dimension image to obtain a set of image blocks containing multiple image blocks, and based on the low rank of the image space, according to the small sample statistical theory method, determining the non-local similar image block corresponding to each image block; obtaining a second two-dimensional tensor based on each image block and the corresponding non-local similar image block, and using the WNNM method to denoise the second two-dimensional tensor to obtain a denoised multispectral image; calculating the current noise level of each band image in the denoised multispectral image; and determining the denoised multispectral image of a single flight UAV based on the current noise level and the initial noise level of each band image in an iterative method; Based on solar irradiance, a bilinear interpolation method is used to correct the denoised single-flight UAV multispectral image to obtain a corrected denoised single-flight UAV multispectral image. Specifically, the method includes: calculating a bilinear spatial radiation correction coefficient based on solar irradiance, according to the bilinear interpolation method, and the relative position between the single-flight UAV multispectral image and the starting point of the UAV flight sortie; and correcting each band image in the denoised single-flight UAV multispectral image according to the bilinear spatial radiation correction coefficient to obtain a corrected denoised single-flight UAV multispectral image.

2. A single-flight UAV multispectral image correction method according to claim 1, characterized in that: The estimating of the initial noise level of each band image in the multispectral image of a single UAV flight specifically includes: Acquire multispectral images from a single UAV flight; The local variance method is used to obtain the initial noise level of each band image in the multispectral image of a single UAV flight.

3. The method for correcting multispectral images of a single-flight UAV according to claim 1, characterized in that: Perform dimensionality reduction on the multispectral image of a single UAV flight to obtain a reduced-dimensional image, specifically including: Convert each band image in a single UAV multispectral image into a column vector to obtain the first two-dimensional tensor; Perform SVD decomposition on the first two-dimensional tensor, select the column vectors of the first K dimensions whose contribution rate exceeds the first set value, and obtain a K-dimensional subspace; The multispectral image of a single UAV flight is projected into the K-dimensional subspace to obtain a reduced-dimensional image.

4. The method for correcting multispectral images of a single-flight UAV according to claim 1, characterized in that: Based on the low rank of the image space and the small sample statistical theory method, the non-local similar image blocks corresponding to each image block are determined, specifically including: Based on the low rank of the image space, the Euclidean distance between the center points of adjacent image blocks is calculated, and the K nearest neighbor algorithm is used to search for the M nearest neighbor image blocks of each image block according to the Euclidean distance between the center points of adjacent image blocks. Calculating the non-local statistical similarity between a target image block and any neighboring image block corresponding to the target image block according to a small sample statistical theory method; the target image block is any image block in the image block collection; The neighborhood image blocks that meet the set conditions are determined as non-local similar image blocks of the target image block, and then the non-local similar image blocks corresponding to each image block are obtained; the set conditions are that the non-local statistical similarity is less than a second set value.

5. The method for correcting multispectral images of a single-flight UAV according to claim 1, characterized in that: According to the current noise level and initial noise level of each band image, an iterative method is used to determine the noise reduction of a single-flight UAV multispectral image, including: The first operation is performed on any band image to obtain an output result corresponding to each band image, and a noise-reduced single-flight UAV multispectral image is obtained according to the output result; The first operation is: Calculate the noise change value of the current iteration number according to the current noise level and initial noise level of the band image; the current noise level is the noise level corresponding to the current iteration number; the initial noise level is the noise level corresponding to the previous iteration number; When the set constraints are met, the denoised band image corresponding to the current number of iterations is output; the set constraints are that the noise change value is less than the minimum threshold or the current number of iterations is the total number of iterations; When the set constraints are not met, the band image that does not meet the set constraints is determined to be a single-flight UAV multispectral image, the number of iterations is increased by 1, and the process returns to the step of reducing the dimensionality of the single-flight UAV multispectral image to obtain a reduced-dimensional image.

6. A single-flight UAV multispectral image correction system, characterized in that: include: The initial noise level estimation module is used to estimate the initial noise level of each band image in a single UAV multispectral image; The denoising module is used to perform denoising on the multispectral image of a single UAV based on the low rank of the image space, according to the small sample statistical theory method and the initial noise level of each band image, in an iterative manner to obtain the denoised multispectral image of the single UAV, specifically including: reducing the dimension of the multispectral image of the single UAV to obtain a reduced dimension image; dividing the reduced dimension image to obtain a set of image blocks containing multiple image blocks, and based on the low rank of the image space, according to the small sample statistical theory method, determining the non-local similar image block corresponding to each image block; obtaining a second two-dimensional tensor based on each image block and the corresponding non-local similar image block, and using the WNNM method to denoise the second two-dimensional tensor to obtain the denoised multispectral image; calculating the current noise level of each band image in the denoised multispectral image; and determining the denoised multispectral image of the single UAV based on the current noise level and initial noise level of each band image in an iterative manner; The correction module is used to correct the noise-reduced single-flight UAV multispectral image based on solar irradiance using a bilinear interpolation method to obtain a corrected noise-reduced single-flight UAV multispectral image. Specifically, the correction module includes: calculating a bilinear spatial radiation correction coefficient based on solar irradiance, according to the bilinear interpolation method, and the relative position of the single-flight UAV multispectral image and the starting point of the UAV flight sortie; and correcting each band image in the noise-reduced single-flight UAV multispectral image according to the bilinear spatial radiation correction coefficient to obtain a corrected noise-reduced single-flight UAV multispectral image.

7. An electronic device, characterized in that: The electronic device includes a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to perform the single-flight UAV multispectral image correction method according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that It stores a computer program, which, when executed by a processor, implements the single-flight UAV multispectral image correction method according to any one of claims 1 to 5.

Images