A hyperspectral image denoising method for aquaculture production scenes

Abstract

The application provides a hyperspectral image denoising method for aquaculture production scenes, and belongs to the technical field of water quality image processing. The hyperspectral image denoising is realized based on non-local statistical similarity and segmented spectrum processing. The method mainly solves the poor adaptability of non-local image block calculation similarity to noise fluctuation and reduces the influence of the difference between the hyperspectral image brightness and darkness on the denoising processing in the prior art. According to different noise levels of different wave bands, the wave band iteration denoising is performed to reduce the problem of missing image texture details caused by global denoising. The implementation steps are as follows: input the hyperspectral image; estimate the initial noise level of each wave band; iterative denoising: project the hyperspectral image to a low-dimensional subspace; calculate the non-local statistical similarity of the image block; denoising based on the low-rank constraint; dimensionality increase reconstruction; update the image and judge whether the termination condition is met; and finally output the denoised hyperspectral image.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of water quality image processing, and in particular, relates to a hyperspectral image denoising method for aquaculture production scenes. BACKGROUND

[0002] With the development of science and technology and the deepening of people's understanding of things, modern agricultural development research increasingly needs a large amount of information. Among them, the hyperspectral image (HIS) has been widely used in the field of agriculture. The hyperspectral image obtained by the hyperspectral imaging system has 3-dimensional information, in which x, y contains spatial information, and λ contains band information. Through the spectral information of dozens or even hundreds of bands, the information of the agricultural production target object is accurately obtained, including the study of water plant growth, pond water quality, etc., to realize water quality judgment, water algae and physiological and biochemical index detection and other applications.

[0003] However, due to the complex agricultural production environment, sensor hardware performance and other factors, there are a large amount of noise in agricultural remote sensing and near-ground hyperspectral images, which brings limitations and difficulties to agricultural application research. At present, a variety of methods have been proposed for hyperspectral image denoising, mainly including the following three types: ① image space-based denoising method, which separately denoises each band of the hyperspectral image as a two-dimensional gray image; ② denoising method based on spectral low rank, which uses the correlation between spectral bands to denoise the image; and ③ denoising method combining image space correlation and spectral space low rank.

[0004] The document "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." studies projecting the hyperspectral image to a low-bit space, finding non-local similar blocks on the average spectral image, and using WNNM denoising. However, the method has the disadvantage that it does not consider the different noise levels of different bands, resulting in the loss of details in some band images after hyperspectral image denoising, and the method needs to provide prior knowledge, and the optimization effect is limited.

[0005] The method proposed by Xi'an University of Electronic Science and Technology in its applied patent document "Hyperspectral remote sensing image denoising method based on four-dimensional block matching filtering" sorts the hyperspectral image according to the spectral signal-to-noise ratio and processes the band images with different noise levels differently. However, this method does not consider the influence of the correlation between the wave spectrum, i.e., the low rank, when grouping the wave bands, resulting in the separation of related wave bands into different groups, affecting the denoising effect. At the same time, the performance error of high-pass filtering is introduced.

[0006] At the same time, various hyperspectral denoising methods including the above two methods mostly use the Euclidean distance similarity formula when searching for similar image blocks. This formula is not sensitive to the size fluctuations of the image pixels within the block, which can easily lead to misjudgment of similarity. SUMMARY

[0007] The present application aims to overcome the shortcomings of the prior art and proposes a hyperspectral image denoising method based on non-local statistical similarity and segmented spectral processing, which can better preserve the image texture information of the denoised hyperspectral image and reduce noise interference, facilitating agricultural scientific research. The present application takes each pixel point in the non-local region of the hyperspectral image as a statistical sample, compares the similarity of different non-local image blocks to find similar blocks, and performs denoising processing based on this. At the same time, different wave bands have different noise levels, so the denoising iteration processing is performed on a wave band basis to achieve the denoising processing of the hyperspectral image data.

[0008] The technical means adopted by the present application are as follows:

[0009] A hyperspectral image denoising method for aquaculture production scenarios, which realizes hyperspectral image denoising based on non-local statistical similarity and segmented spectral processing, includes the following steps:

[0010] Step 1: input the hyperspectral image;

[0011] Step 2: estimate the initial noise level of the hyperspectral image;

[0012] Step 3: iterative denoising;

[0013] Step 4: output the denoised hyperspectral image data.

[0014] Further, the step 3 in the iterative denoising process includes the following steps:

[0015] Step 31: project the low-dimensional subspace; input the hyperspectral image I 0 , the hyperspectral image I 0 is HxWxN, respectively, longxwideband number, construct a two-dimensional tensor T 0 (HxW)XN, where (HxW) represents a row vector and N represents a column vector, and make K << N through singular value decomposition;

[0016] Will I 0 Projecting into the K-dimensional subspace to obtain the reduced dimensionality image I 1 , and get the orthogonal matrix E0, the following relationship exists:

[0017] I 0 =I 1 ×E0;

[0018] sE0 T E0=I;

[0019] Step 32: Calculate the non-local statistical similarity of image blocks: Given a non-local similarity block size of 5×5, I 1 Divide the size into The image block collection {p j}, fill in the boundary points; search the image block p through the K nearest neighbor algorithm i The M image blocks with similar image distances are obtained by calculating the local statistical similarity and obtaining the calculated non-local similar image block set {p l}∈{p j};

[0020] Step 33: Noise reduction based on low-rank constraint; calculate {p j} and p i The two-dimensional tensor Te is obtained by combining the two dimensions, which is (25×K)×L, where L is the number of similar image blocks, that is, {p l} capacity, representing a column vector; 25×K represents all band data of 25 pixels in an image block, representing a row vector; the tensor Te is noise corrected and denoised; all image blocks are processed to complete the denoising;

[0021] Step 34: Dimensional reconstructing: Perform dimensional reconstructing on the denoised image to obtain I 2 , calculate I 2 The noise level and the previous noise change level are compared to confirm whether the band has completed noise reduction;

[0022] Step 35: Update iteration; update the hyperspectral image and enter iteration, and stop when the number of iterations is met.

[0023] Furthermore, the step 2 estimates the initial noise level of the hyperspectral image, and estimates the noise level of each band by the local variance method. The i-th dimension band image has a mean value of u i , with variance σ i 2 Noise n i .

[0024] Further, in the step 32, when the boundary points are supplemented, the boundary point non-image area is filled with 0 in each band, and when the K nearest neighbor algorithm is searched, the Euclidean distance of the image block center point is taken as the distance reference.

[0025] Further, in the step 32, the non-local statistical similarity is calculated as follows:

[0026] Step 321: take the image block p i and the M neighborhood image block set {p j}, j = 1, 2, 3…M; average processing is performed on the K-band image to obtain the average spectral image block For all pixel points, the average spectral intensity value δ is calculated, and all pixel points are subtracted by the average value δ to obtain the average spectral image with a spectral intensity average value of 0

[0027] Step 322: take the image block p j , calculate the average spectral image and perform the same operation as to obtain the average spectral image with a spectral intensity average value of 0

[0028] Step 323: for the average spectral image p i ' and the average spectral image , each pixel point is taken as a sample, and variance ratio analysis, i.e., F test, is performed to evaluate the difference in spectral characteristics of the two image blocks by comparing the variances of the corresponding pixel points in all bands.

[0029] Further, in the step 33, the noise correction denoising process is performed as follows:

[0030] Step 331: SVD decomposition is performed on the tensor Te as follows to obtain the singular value matrix ∑;

[0031] Te = U∑V T ;

[0032] wherein ∑ represents a diagonal matrix, and the elements on the diagonal line are the singular values of Te;

[0033] Step 332: according to the global noise variance of the band, normalization correction is performed and noise is removed according to the soft threshold method:

[0034]

[0035] wherein, represents the jth singular value of the ith band, ​represents the global noise variance of the i-th band, and λ represents a threshold parameter.

[0036] Further, the image in step 34 is subjected to dimensionality reconstruction operation, that is, the noise-reduced tensor is obtained after noise reduction processing:

[0037] Te'=U∑'V T ;

[0038] wherein, Te' represents the tensor after noise reduction processing, and ∑' represents the singular value matrix after noise removal according to the soft threshold method;

[0039] The complete noise reduction of the hyperspectral image is realized by processing all the image blocks and the non-local similar image blocks, and the low-dimensional hyperspectral image I 2 after noise reduction is obtained. 0’ :

[0040] I 0’ =I 2 ×E0.

[0041] Further, the noise level in step 34 is calculated, and the formula is:

[0042]

[0043] wherein, γ represents a scale factor, ε1 and ε2 represent minimum threshold values; σ' represents the noise level after this iteration processing; σ''i represents the noise level after the previous iteration processing, and σ''i=σ for the first iteration; Δσ1 represents the noise level of the image after noise reduction processing, and Δσ2 represents the noise change level in the iteration process; when Δσ1 or Δσ2 approaches 0, it is indicated that the spectral band image has completed the noise reduction processing, and does not enter the next iteration processing. i i

[0044] Further, the hyperspectral image in step 4 is updated as I 0’ (H×W×(N-Q)); Q represents the band that does not perform the next iteration noise reduction.

[0045] Compared with the prior art, the present application has the following advantages:

[0046] 1. The non-local similarity of the image block is calculated by the small sample statistical theory method (F test), which has better adaptability to noise fluctuation than the Euclidean distance. Meanwhile, the method can eliminate the influence of the light and dark difference of similar image blocks on noise reduction.

[0047] ​​2. In combination with the low rank of the spectrum, the noise level is compared with the noise reduction level by dividing the waveband, and the problem of missing image texture details in some wavebands caused by global noise reduction is eliminated.

[0048] 3. According to the characteristics that the noise intensity of different wavebands in non-local similar blocks is different and the distribution is different, the noise reduction processing of different wavebands and different intensities is realized by the adaptive noise intensity threshold setting.

[0049] 4. The method fully considers the different noise intensities of different wavebands, and avoids the situation that the waveband noise cannot be completely removed when the overall noise level of the waveband is high. BRIEF DESCRIPTION OF DRAWINGS

[0050] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or the prior art description will be briefly introduced. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0051] Fig. 1 The overall flowchart of the present application is shown in the figure.

[0052] Fig. 2 The image block division and boundary point supplementing schematic diagram of the present application are shown in the figure. The red and blue boxes are image blocks, the orange box is the center pixel point of the image block, and the green character 0 is the supplementary pixel point.

[0053] Fig. 3 The local images before and after noise reduction for a certain waveband are shown in the figure. (a) is before noise reduction, and (b) is after noise reduction. DETAILED DESCRIPTION

[0054] In order to make the person skilled in the art better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should be within the scope of protection of the present application.

[0055] It is to be understood that the terms "first", "second", and the like, used in the description and the claims of the application, as well as the above-described drawings, are used to distinguish between similar objects and are not necessarily used to describe a particular sequential or chronological order. It is to be understood that the use of data so designated is not to be construed as limiting of the embodiments of the application described herein to only those embodiments absolutely recited in the description and the claims. Further, the terms "include," "includes," and "including" and any variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, system, product, or apparatus that comprises a list of steps or elements does not necessarily comprise only those steps or elements that are literally recited, but can include other steps or elements not expressly listed or inherent to such process, method, system, product, or apparatus.

[0056] As shown in Figs. 1-3 , the application provides a hyperspectral image denoising method for aquaculture production scene, which realizes hyperspectral image denoising based on non-local statistical similarity and segmented spectrum processing, and comprises the following steps:

[0057] Step 1: input the hyperspectral image.

[0058] Step 2: estimate the initial noise level of the hyperspectral image; wherein the mathematical model of the hyperspectral image is:

[0059]

[0060] wherein, denotes a real noisy image, denotes a noise-free original image, denotes noise (usually Gaussian noise). The noise level of each band image of the hyperspectral image is obtained by a local variance method, wherein the i-th dimensional band image has a mean value u i , a variance σ i 2 , and noise n i .

[0061] Step 3: iterative denoising; specifically, comprising the following steps:

[0062] Step 31: project a low-dimensional subspace; input the hyperspectral image I 0 (5460x8192x256, respectively, longxwideband number), convert I 0 and the image of each band to a column vector, and then construct a two-dimensional tensor T 0 ((5460x8192)x256, wherein (5460x8192) is a row vector and 256 is a column vector), perform SVD decomposition on T 0 , and select the first K dimensions (K << N) with a contribution rate of more than 95%. Convert I 0Projecting into K-dimensional subspace to get reduced dimension image I 1 and get orthogonal matrix E0, where

[0063] I 0 = I 1 × E0,

[0064] s.t. E0 T E0= I,

[0065] Step 32: Calculate the non-local statistical similarity of image blocks; given the non-local similarity block size of 5x5, I 1 is divided into image block set with size (H / 5xW / 5xK) i.e. (1092x1638xK). For the boundary points, the data points with 0 intensity in each band are supplemented when calculating the similarity, which is convenient for calculation, as shown below. Fig. 3 Based on the low rank property of image space, the Euclidean distance of the center point of the image block is calculated, and the K nearest neighbor algorithm is used to search for the M nearest neighbor image blocks of each image block, and the non-local statistical similarity is calculated, as follows:

[0066] Step 321: Take image block p i and M neighborhood image block set {p j}, j=1, 2, 3…M; average processing is performed on the K-band image to obtain the average spectral image block For all pixel points, the average spectral intensity value δ is calculated, and all pixel points are subtracted by the mean value δ to obtain the average spectral image with spectral intensity mean value of 0

[0067] Step 322: Take p j , calculate the average spectral image and perform the same operation as to obtain the average spectral image with spectral intensity mean value of 0

[0068] Step 323: For p i ' and , each pixel point is taken as a sample, and F test is used to check whether there is a significant difference between the two image blocks. If not, it is judged that the two image blocks are similar.

[0069] Step 33: Noise reduction based on low rank constraint:

[0070] The non-local similar image blocks {p l}∈{p j} calculated are combined with p i to obtain a two-dimensional tensor T 1, size of ((25 x K) x L), where L is the number of similar image blocks, i.e. the capacity of {p l} represents a column vector; 25 x K represents all the band data of an image block of 25 pixels, representing a row vector.

[0071] The calculated {p j} and p i are combined to obtain a two-dimensional tensor Te, size of ((25 x K) x L), where L is the number of similar image blocks, i.e. the capacity of {p T} represents a column vector; 25 x K represents all the band data of an image block of 25 pixels, representing a row vector. The tensor Te is denoised, and the tensor Te is SVD decomposed as follows,

[0072] Te = U∑V T

[0073] Since the noise intensity of each band image is different, and the noise level of different blocks in the same band image also has some difference. In order to avoid incomplete denoising of high-intensity noise band or loss of details caused by excessive denoising of low-intensity noise when denoising by soft threshold, the noise intensity is modified so as to satisfy the objective function as follows:

[0074]

[0075] Where i represents the i-th band, λ is a soft threshold coefficient, is a regularization term, Tr(∑) is the trace of the singular value matrix Σ, σ i is the global noise variance of the i-th band. Thus, the soft threshold denoising is as follows:

[0076]

[0077] Where, is the j-th singular value of the i-th band in the singular value matrix, i = 1, 2, 3,..., K, j = 1, 2, 3,..., 25. ε0 is a minimum threshold. Update all singular values to obtain the matrix Σ'.

[0078] Step 34: dimensionality lifting reconstruction; the denoised tensor Te' = U∑'V T , the image block p i and its non-local similar image blocks {p j} are denoised. Through processing all image blocks and non-local similar image blocks, a denoised hyperspectral image I 2 is obtained. The orthogonal matrix E0 calculated in step 2.1 is used to back-project I 2 to obtain a high-dimensional hyperspectral image I 0’ , and the formula is as follows:

[0079] I0’ = I 2 × E0;

[0080] obtain the denoised hyperspectral image I 0’ , for I 0’ Each band, re-calculate each noise level to obtain the variance σ i ’2 of the noise n i ', and calculate the iterative noise change, the formula is as follows:

[0081]

[0082] Wherein, γ is the scale factor, ε1 and ε2 are the minimum threshold. σ" i is the noise level of the previous iteration process, the first iteration σ" i = σ i . Δσ1 represents the noise level of the image after denoising processing, and Δσ2 represents the noise change level in the iteration process. When Δσ1 or Δσ2 approaches 0, it means that the spectral band image has completed the denoising processing, and does not enter the next iteration processing.

[0083] Step 35: update iteration; update the hyperspectral image to I 0” (8192×5460×(256-Q)). Q is the band that does not perform the next iteration denoising. Stop when the iteration number is met.

[0084] Step 4: output the denoised hyperspectral image data, as shown in Fig. 3 the local image before and after denoising in a certain band.

[0085] The above-mentioned embodiment numbers of the present application are only for description, and do not represent the advantages and disadvantages of the embodiments. In the above-mentioned embodiments of the present application, the description of each embodiment has its own emphasis, and the parts not described in detail in a certain embodiment can be referred to the related description of other embodiments. In the several embodiments provided in the present application, it should be understood that the disclosed technical contents can be realized by other ways.

[0086] Finally, it should be pointed out that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the above-mentioned embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the above-mentioned embodiments, or make equivalent replacement to part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A hyperspectral image denoising method for aquaculture production scenarios, based on non-local statistical similarity and segmented spectral processing to achieve hyperspectral image denoising, characterized in that, The method comprises the following steps: Step 1: input a hyperspectral image; Step 2: estimate the initial noise level of the hyperspectral image; Step 3: perform iterative noise reduction; The step 3 in the iterative noise reduction process comprises the following steps: Step 31: Projecting low-dimensional subspace; input hyperspectral image , the hyperspectral image is , respectively, long x wide x band number, construct a two-dimensional tensor , where represents a row vector, represents a column vector and is decomposed by singular value to make ; will be projected to onto the lower dimensional subspace to obtain a reduced dimension image , and an orthogonal matrix , there is the following relationship: ； ； Step 32: Calculate non-local statistical similarity of image blocks; given a non-local similar block size of , the image block set is divided into image blocks of size , and the boundary points are supplemented; M image blocks with similar image distance are searched by K- nearest neighbor algorithm, and the non-local statistical similarity is calculated to obtain a non-local similar image block set l j ; the non-local statistical similarity in step 32 is calculated as follows:​​​ Step 321: take image block And M A neighborhood image block collection , ; average processing of images of K Band, get average spectral image block ; for All pixel points, calculate the average spectral intensity value , and subtract the mean value Light all pixel points Get the average spectral image with spectral intensity mean value 0 ; Step 322: Take the image block , calculate the average spectral image , and perform the same operation as , to obtain the average spectral image with the spectral intensity mean value of 0 ; Step 323: For the average spectral image with the average spectral image For each pixel, a variance ratio analysis, i.e. F-test, is performed to evaluate the difference in spectral characteristics of the two image blocks by comparing the variances of the corresponding pixels in all wavebands. Step 33: denoising based on low-rank constraint; combining the calculation {p j} with p i to obtain a two-dimensional tensor Te with a size of , wherein L is the number of similar image blocks, i.e. , and the capacity of the column vector is represented by , wherein all the waveband data of 25 pixels of an image block are represented by a row vector; performing noise correction and denoising processing on the tensor Te; and performing processing on all the image blocks to complete denoising; Step 34: Dimensionality reconstruction; the dimensionality reconstruction is performed on the denoised image to obtain , the noise level of is calculated and the noise change level of the previous time is confirmed to determine whether the wave band is completed denoising; Step 35: update the iteration; update the hyperspectral image and enter the iteration, and stop when the number of iterations is met; Step 4: output the noise-reduced hyperspectral image data.

2. The hyperspectral image denoising method for aquaculture production scene according to claim 1, characterized in that, The step 2 estimates the initial noise level of the hyperspectral image, estimates the noise level of each waveband by local variance method, the first waveband image exists mean , variance of noise ​ 3. The hyperspectral image denoising method for aquaculture production scene according to claim 1, characterized in that, In the step 32, when the boundary points are supplemented, the non-image area of each band of the boundary points is supplemented with 0, and when searching by the K nearest neighbor algorithm, the Euclidean distance of the center point of the image block is taken as the distance reference.

4. The hyperspectral image denoising method for aquaculture production scene according to claim 2, characterized in that, In the step 33, the noise correction noise reduction process is performed as follows: Step 331: Perform SVD decomposition on the tensor Te as follows, to obtain a singular value matrix ; ； wherein denotes a diagonal matrix, the elements on the diagonal being singular values of Step 332: normalize and correct according to the global noise variance of the band, and remove noise according to the soft threshold method: ； wherein, denotes the singular value of the waveband, global noise variance of the waveband, denotes a threshold parameter, denotes a minimal threshold value.

5. The hyperspectral image denoising method for aquaculture production scene according to claim 2, characterized in that, In the step 34, the image is subjected to dimension upgrading reconstruction operation, and the noise-reduced tensor is obtained after noise reduction processing: ； wherein, the denoised tensor, the singular value matrix after removing noise according to the soft threshold method; The complete denoising of the hyperspectral image is realized by processing all the image blocks and the non-local similar image blocks to obtain the low-dimensional hyperspectral image after denoising ; the orthogonal matrix matrix obtained according to step 31 E 0 deprojection to obtain the high-dimensional hyperspectral image : 。 6. The hyperspectral image denoising method for aquaculture production scene according to claim 1 or 5, characterized in that, In the step 34, the noise level is calculated, and the formula is: ； ； wherein, denotes a scale factor, denotes a noise level of the image, denotes a minimum threshold value; denotes a noise level after the current iteration processing, denotes a noise level after the previous iteration processing, and = ; denotes a noise level of the image after the noise reduction processing, denotes a noise level change during the iteration process; when or approaches 0, it indicates that the spectral band image has completed the noise reduction processing and does not enter the next iteration processing.

7. The hyperspectral image denoising method for aquaculture production scene according to claim 1, characterized in that, The step 4 updates the hyperspectral image as ; denotes the waveband for which the next iteration of denoising is not performed.

Images