A hyperspectral remote sensing monitoring method for wetland soil microbial communities

Abstract

The present invention proposes a method for monitoring wetland soil microbial communities by hyperspectral remote sensing. The method comprises the following steps: calculating pixel differences between hyperspectral image data and synthetic aperture radar imaging data, correcting pixel values ​​of hyperspectral image data based on the pixel differences, and obtaining a hyperspectral remote sensing image containing a region to be detected; performing dimension reduction preprocessing on the hyperspectral remote sensing image data, and using the first four principal components to represent the entire hyperspectral remote sensing image of the microbial community; using a separability index M, obtaining the separability of regions with microbial communities and regions without microbial communities in the hyperspectral remote sensing image; separating regions with microbial communities from regions without microbial communities in the hyperspectral remote sensing image, reading regional images with microbial communities, and performing fragment segmentation on the regional images; and using a trained classification model to classify image fragments obtained after fragment segmentation, thereby identifying microbial communities.

Description

Technical Field

[0001] The present invention relates to the field of microbial community detection using hyperspectral remote sensing, and in particular to a wetland soil microbial community hyperspectral remote sensing monitoring method. Background Art

[0002] In the existing technology, ground measurement methods are often used to obtain wetland soil microbial community information, but this method of acquisition is mainly based on manual measurement, which is easily affected by non-objective factors, and has high economic and time costs. It is difficult to quickly obtain data on a large spatial scale, and the accuracy and real-time nature of the data cannot be guaranteed, and the needs of relevant departments cannot be met in a timely manner.

[0003] Compared with traditional monitoring methods, remote sensing technology can observe and identify targets at a long distance, provide various information objectively, accurately and timely, realize large-area synchronous observation, make up for the shortcomings of traditional survey methods, and have strong advantages in ground object monitoring and identification. However, when remote sensing technology uses low and medium spatial resolution remote sensing data for identification, mixed pixel problems are prone to occur, and classification accuracy is low. In addition, the spectral characteristics of wetland soil microbial communities will change over time, making their spectral characteristics highly dynamic and complex. Using single-phase remote sensing data, it is difficult to capture the changes in the spectral characteristics of wetland soil microbial communities, and it is impossible to accurately obtain remote sensing information for wetland soil microbial community identification. Time series remote sensing data can dynamically monitor the changes in wetland soil microbial communities, accurately capture the changing laws of the spectral characteristics of wetland soil microbial communities, and determine the key time window for remote sensing identification of wetland soil microbial communities.

[0004] Hyperspectral remote sensing, with its high spectral resolution and strong band continuity, can obtain fine spectral information that multispectral sensors cannot obtain. Hyperspectral remote sensing data can capture the fine spectral characteristics of soil, which is conducive to quantitatively obtaining the biochemical components of soil. The logarithm of the spectral reciprocal (LR), the spectral first-order differential (FD), the spectral second-order differential (SD), the envelope removal (CR), and the spectral band depth (BD) are the spectral transformation technologies commonly used in hyperspectral remote sensing.

[0005] Using hyperspectral remote sensing data to estimate the physiological and biochemical parameters of wetland soil microbial communities is an urgent need for scientific research such as ecology, agronomy, global change, and application industries such as precision agriculture. This measurement method is fast and convenient, and its measurement area can even be expanded to the scale of the entire earth. And it is a non-destructive measurement method that will not have any impact on vegetation growth. Summary of the invention

[0006] In order to solve the above technical problems, the technical problem to be solved by this application is to provide a hyperspectral remote sensing monitoring method for wetland soil microbial communities, which estimates the spatiotemporal distribution of microbial communities based on the hyperspectral remote sensing detection mechanism.

[0007] A method for monitoring wetland soil microbial communities using hyperspectral remote sensing includes the following steps:

[0008] S1, calculating the pixel difference between the hyperspectral image data and the synthetic aperture radar imaging data, and correcting the pixel value of the hyperspectral image data based on the pixel difference to obtain a hyperspectral remote sensing image containing the area to be detected;

[0009] S2, performing dimension reduction preprocessing on the hyperspectral remote sensing image data, and using the first four principal components to represent the entire hyperspectral remote sensing image of the microbial community;

[0010] S3, using the separability index M, obtaining the separability of the area with microbial community and the area without microbial community in the hyperspectral remote sensing image;

[0011] S4, after separating the area with microbial community from the area without microbial community in the hyperspectral remote sensing image, reading the area image with microbial community, and performing fragment segmentation on the area image;

[0012] S5. Classify the image fragments obtained after fragment segmentation using the trained classification model to identify the microbial community.

[0013] Furthermore, in step S1, each data in the hyperspectral image data and the synthetic aperture radar imaging data is traversed respectively, the maximum value and the minimum value are recorded, and the difference between the maximum value and the minimum value is used as the base to perform data normalization processing:

[0014]

[0015] Where X in formula (1) is the pixel value of the image to be normalized, X max is the maximum value, X min is the minimum pixel value, and S(X') is the normalized pixel value.

[0016] Furthermore, the normalized synthetic aperture radar imaging data and the normalized hyperspectral image data are used to calculate the pixel difference, and the calculation formula is as follows:

[0017] T(X') = S2(X') - S1(X') (2);

[0018] Among them, S2(X') and S1(X') represent the normalized pixel values ​​of hyperspectral image data and synthetic aperture radar imaging data, respectively, and T(X') represents the pixel difference between hyperspectral image data and synthetic aperture radar imaging data;

[0019] Set the pixel difference threshold T h , when T(X')>T h When T(X')≤T h When , the pixel value is retained in the hyperspectral image data, thereby obtaining a hyperspectral remote sensing image containing the area to be detected.

[0020] Furthermore, in step S2, the data of each band of the original hyperspectral image is standardized to obtain a standardized image matrix Xc; the covariance matrix Σc of the standardized matrix Xc is calculated; the eigenvector matrix Ac of the covariance matrix Σc is solved, and the eigenvector matrix Ac is arranged from left to right according to the law of decreasing eigenvalues; the standardized image matrix Xc is linearly transformed using the eigenvector matrix Ac to obtain a transformation matrix X P :

[0021]

[0022] Transformation matrix X P The first row of data represents the first principal component of the hyperspectral remote sensing image, the second row of data represents the second principal component of the hyperspectral remote sensing image, and so on.

[0023] Further, in step S3, the separability index M is:

[0024]

[0025] Where: μ pm is the mean reflectance of wetland soil with microbial communities, u i (i=1,2,…,n) are the mean reflectance values ​​of different types of objects in wetland soil without microbial communities; δ pm is the standard deviation of reflectance of wetland soil with microbial communities, δ i (i=1,2,…,n) are the standard deviations of reflectance of different types of objects in wetland soil without microbial communities.

[0026] Furthermore, the step S4 comprises the following steps:

[0027] S41. Construct a weighted directed graph based on the image to be segmented: Pixels in the image to be segmented are regarded as nodes in the directed graph, and the edges between adjacent pixels are regarded as the edges connecting the nodes. A cost function is defined on the edges. The cost value formula of two adjacent pixels is as shown in formula (5):

[0028] t(p,q)=ω G ×f G (q)+ω Z ×f Z (q)+ω D ×f D (p, q) (5);

[0029] Where t(p,q) is the local cost from pixel p to its adjacent pixel q; G ,ω Z and ω D is the weighting coefficient; f G (q), f Z (q) and f D (q) is the gradient characteristic function, Laplace zero-crossing characteristic function and smoothness constraint function at the corresponding pixel q;

[0030] S42. Use an optimal path search algorithm to find the shortest path between two points in a weighted directed graph as a boundary segment of the object, and perform edge segmentation.

[0031] Furthermore, in step S5, SVM is used to train the training set, and radial basis function is used as the kernel function used for SVM training. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The drawings described herein are used to provide further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute improper limitations on the present application.

[0033] Attached Figure 1 It is a schematic diagram of the process of the hyperspectral remote sensing monitoring method of wetland soil microbial community of the present invention;

[0034] Attached Figure 2 A flow chart of a method for performing fragment segmentation on a regional image having a microbial community according to the present invention;

[0035] Attached Figure 3 It is a schematic diagram of the relationship between the classification accuracy of the present invention and the number of iterations. DETAILED DESCRIPTION

[0036] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0037] The coverage of wetland soil microbial communities is a process that is gradually completed over a certain period of time. At different times, wetland soils with microbial communities and wetland soils without microbial communities have different tones on remote sensing images, and the remote sensing characteristics of microbial communities will change accordingly over time. Selecting remote sensing images with short revisit periods and high spatial resolution can continuously observe changes in microbial communities and accurately capture remote sensing characteristics of microbial communities.

[0038] like Figure 1 FIG. 1 is a schematic diagram of a method for monitoring wetland soil microbial communities by hyperspectral remote sensing according to the present invention, which specifically includes the following steps:

[0039] Step 1: Analyze the pixel difference between the hyperspectral image data and the synthetic aperture radar imaging data, correct the pixel value of the hyperspectral image data based on the pixel difference, eliminate missing or insufficient remote sensing data, and thus obtain a hyperspectral remote sensing image containing the area to be detected.

[0040] A remote sensing detector with little or no clouds is selected to acquire hyperspectral image data. In this embodiment, a single Sentinel-2 satellite is preferably used as the remote sensing detector. The revisit period of a single Sentinel-2 satellite is 10 days, and the revisit period of the A / B double satellite is 5 days. The main payload is a multispectral imager (MSI) with a total of 13 bands and a spectral range of 0.4 to 2.4 μm, covering visible light, near infrared and short-wave infrared.

[0041] The all-day, all-weather synthetic aperture radar imaging system acquires synthetic aperture radar imaging data. The two satellites of the synthetic aperture radar imaging system adopt a sun-synchronous orbit with an orbital altitude of 693km, an inclination of 98.180, an orbital period of 99min, and a single satellite revisit period of 12 days. In this embodiment, single-view complex image data is preferably used as synthetic aperture radar imaging data, the imaging mode adopts dual polarization, and the imaging mode is an interferometric wide-band mode.

[0042] The pixel value of synthetic aperture radar imaging data represents the reflectivity of ground objects, while the pixel value of hyperspectral imaging data represents the backscattering intensity of radar waves from ground targets. In order to correct the pixel value of hyperspectral imaging data, the pixel difference between hyperspectral imaging data and synthetic aperture radar imaging data is analyzed from a numerical perspective, that is, all data in hyperspectral imaging data and synthetic aperture radar imaging data are normalized respectively, and the pixel values ​​of all data are converted to between 0 and 1.

[0043] Specifically, by traversing each data in the hyperspectral image data and the synthetic aperture radar imaging data respectively, the maximum and minimum values ​​are recorded, and the difference between the maximum and minimum values ​​is used as the base to normalize the data. The calculation formula for the normalization processing is shown in formula (1).

[0044]

[0045] Where X in formula (1) is the pixel value of the image to be normalized, X max is the maximum value, X min is the minimum pixel value, and S(X') is the normalized pixel value.

[0046] The pixel difference calculation is performed using the normalized synthetic aperture radar imaging data and the normalized hyperspectral image data. The calculation formula is as follows:

[0047] T(X') = S2(X') - S1(X') (2);

[0048] Among them, S2(X') and S1(X') represent the normalized pixel values ​​of hyperspectral image data and synthetic aperture radar imaging data, respectively, and T(X') represents the pixel difference between hyperspectral image data and synthetic aperture radar imaging data.

[0049] Set the pixel difference threshold T h , when T(X')>T h When T(X')≤T h When , the pixel value is retained in the hyperspectral image data, thereby obtaining a hyperspectral remote sensing image containing the area to be detected.

[0050] Step 2: Perform dimensionality reduction preprocessing on the high-light remote sensing spectral image data.

[0051] The differences in the basic components of microbial communities lead to differences in spectral reflectance curves, which is the physical basis for analyzing and classifying objects based on hyperspectral remote sensing images. Due to the quality of the remote sensing detector itself, images in some bands may have relatively large noise. Therefore, the hyperspectral remote sensing spectral image data is preprocessed for dimensionality reduction.

[0052] In this embodiment, the dimension reduction preprocessing algorithm for the high-light remote sensing spectral image data adopts component analysis, uses KL transformation, and uses the variance of the data to describe the amount of information, attempting to make the transformed data distributed in a decreasing manner according to the amount of information.

[0053] The linear projection method is used to project the original data into a new coordinate space. In the new coordinate space, the first principal component contains the largest amount of information, the second principal component is not related to the first principal component data and has the largest amount of information among the remaining components, and so on. For hyperspectral images with dozens of bands, the first four or five principal components can generally contain more than 90% of the information of the image. Therefore, the first few principal components can be used to replace the entire hyperspectral image to achieve the purpose of dimensionality reduction.

[0054] The specific method of dimensionality reduction preprocessing is as follows:

[0055] Standardize the data of each band of the original hyperspectral remote sensing image to obtain the standardized image matrix Xc;

[0056] Calculate the covariance matrix Σc of the standardized matrix Xc;

[0057] Find the eigenvector matrix Ac of the covariance matrix Σc, where the eigenvector matrix Ac is arranged from left to right according to the law of decreasing eigenvalues;

[0058] The obtained eigenvector matrix Ac is used to linearly transform the standardized image matrix Xc to obtain the transformation matrix X P , the calculation formula is:

[0059]

[0060] Transformation matrix X P The first row of data represents the first principal component of the hyperspectral remote sensing image, the second row of data represents the second principal component of the hyperspectral remote sensing image, and so on. In this embodiment, the first four principal components are preferably used to represent the entire hyperspectral image of the microbial community.

[0061] Step 3: Using the separability index M, by quantitatively analyzing the separation between the regions with microbial communities and the regions without microbial communities, an index that can fully distinguish the regions with microbial communities from the regions without microbial communities is selected.

[0062] The separability index M mainly represents the difference between the mean values ​​of the spectral reflectance of the two regions and the normalized sum of their standard deviations δ. The mathematical expression is:

[0063]

[0064] Where: μ pm is the mean reflectance of wetland soil with microbial communities, u i (i=1,2,…,n) are the mean reflectance values ​​of different types of objects in wetland soil without microbial communities; δ pm is the standard deviation of reflectance of wetland soil with microbial communities, δ i(i=1,2,…,n) are the standard deviations of reflectance of different types of objects in wetland soil without microbial communities.

[0065] If M>1, it means that the separation between the area with microbial community and the area without microbial community is good; if M<1, it means that the separation between the area with microbial community and the area without microbial community is poor. The larger the M value, the better the separability between the area with microbial community and the area without microbial community. Therefore, the separability index M is used to obtain the separability of the area with microbial community and the area without microbial community in the wetland soil.

[0066] Step 4: After separating the area with microbial community from the area without microbial community in the wetland soil, read the hyperspectral image of the area with microbial community in the wetland soil, and perform fragment segmentation on the regional hyperspectral image to obtain a number of segmented image fragments:

[0067] The fragment segmentation algorithm is an interactive image segmentation algorithm that can effectively avoid over-segmentation without losing image information, and has better performance in segmenting complex microbial community images. It specifically includes the following sub-steps:

[0068] S41. Construct a weighted directed graph according to the image to be segmented.

[0069] The pixels in the image to be segmented are regarded as nodes in a directed graph, and the edges between adjacent pixels are regarded as edges connecting nodes. A cost function is defined on the edge, and its value is used as the weight of the edge, and strong edges have smaller cost values, non-edges have larger cost values, and the cost value of the edge connecting non-adjacent pixels is +∞.

[0070] The cost formula of two adjacent pixels is as follows:

[0071] t(p,q)=ω G ×f G (q)+ω Z ×f Z (q)+ω D ×f D (p, q) (5);

[0072] Where t(p,q) is the local cost from pixel p to its adjacent pixel q; G ,ω Z and ω D is the weighting coefficient; f G (q), f Z (q) and f D(q) is the gradient characteristic function, Laplace zero-crossing characteristic function and smoothness constraint function at the corresponding pixel q. The above functions are calculated as shown in equations (6) to (8):

[0073]

[0074]

[0075]

[0076] Where G(q) and L(q) are the amplitude and Laplace value of the gradient at pixel q respectively; D(q) is the unit normal vector at pixel q in the image.

[0077] S42. Use an optimal path search algorithm to find the shortest path between two points in a weighted directed graph as a boundary segment of the object, and perform edge segmentation.

[0078] Step 5: Use the existing confirmed microbial community images as a training set, train the training set with SVM to obtain a microbial community classification model, and use the trained classification model to classify the image fragments obtained in step 2 to identify the microbial community.

[0079] In the preferred embodiment, since the radial basis function has strong locality in SVM training and can show good performance on small sample data, the radial basis function is selected as the kernel function used for SVM training based on the small number of features of image fragments. Figure 3 As shown, the relationship between the classification accuracy and the number of iterations when using existing confirmed microbial community images as training sets.

[0080] Those skilled in the art will readily appreciate other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses or adaptations of the present invention that follow the general principles of the present invention and include common knowledge or customary techniques in the art that are not disclosed by the present invention. The specification and examples are to be considered exemplary only, and the true scope and spirit of the present invention are indicated by the following claims.

[0081] It should be understood that the present invention is not limited to the exact construction that has been described above and shown in the drawings and that various modifications and changes may be made without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.

Claims

1. A method for monitoring wetland soil microbial communities by hyperspectral remote sensing, characterized in that: The steps include: S1. Calculate the pixel difference between the hyperspectral image data and the synthetic aperture radar imaging data, correct the pixel value of the hyperspectral image data based on the pixel difference, and obtain a hyperspectral remote sensing image containing the area to be detected; traverse each data in the hyperspectral image data and the synthetic aperture radar imaging data respectively, record the maximum value and the minimum value, and use the difference between the maximum value and the minimum value as the base to perform data normalization processing: Where X in formula (1) is the pixel value of the image to be normalized, X max is the maximum value, X min is the minimum pixel value, S(X') is the normalized pixel value; The pixel difference calculation is performed using the normalized synthetic aperture radar imaging data and the normalized hyperspectral image data. The calculation formula is as follows: T(X') = S2(X') - S1(X') (2); Among them, S2(X') and S1(X') represent the normalized pixel values ​​of hyperspectral image data and synthetic aperture radar imaging data, respectively, and T(X') represents the pixel difference between hyperspectral image data and synthetic aperture radar imaging data; Set the pixel difference threshold T h , when T(X')>T h When T(X')≤T h When , the pixel value is retained in the hyperspectral image data, thereby obtaining a hyperspectral remote sensing image containing the area to be detected; S2, preprocessing the hyperspectral remote sensing image data by dimensionality reduction, using the first four principal components to represent the entire hyperspectral remote sensing image of the microbial community; standardizing the data of each band of the original hyperspectral image to obtain a standardized image matrix Xc; calculating the covariance matrix Σc of the standardized matrix Xc; solving the eigenvector matrix Ac of the covariance matrix Σc, wherein the eigenvector matrix Ac is arranged from left to right according to the law of decreasing eigenvalues; using the eigenvector matrix Ac to perform a linear transformation on the standardized image matrix Xc to obtain a transformation matrix Xc P : Transformation matrix X P The first row of data represents the first principal component of the hyperspectral remote sensing image, the second row of data represents the second principal component of the hyperspectral remote sensing image, and so on; S3. Using the separability index M, the separability of the area with microbial community and the area without microbial community in the hyperspectral remote sensing image is obtained; the separability index M is: Where: μ pm is the mean reflectance of wetland soil with microbial communities, u i , i = 1, 2, ..., n, are the mean reflectance values ​​of different types of objects in wetland soil without microbial communities; δ pm is the standard deviation of reflectance of wetland soil with microbial communities, δ i ,i=1,2,…,n,are the standard deviations of reflectance of different types of objects in wetland soil without microbial communities; S4, after separating the area with microbial community from the area without microbial community in the hyperspectral remote sensing image, reading the area image with microbial community, and performing fragment segmentation on the area image; S5. Classify the image fragments obtained after fragment segmentation using the trained classification model to identify the microbial community.

2. The method for monitoring wetland soil microbial communities by hyperspectral remote sensing according to claim 1, characterized in that: The step S4 comprises the following steps: S41. Construct a weighted directed graph based on the image to be segmented: Pixels in the image to be segmented are regarded as nodes in the directed graph, and the edges between adjacent pixels are regarded as the edges connecting the nodes. A cost function is defined on the edges. The cost value formula of two adjacent pixels is as shown in formula (5): t(p,q)=ω G ×f G (q)+ω Z ×f Z (q)+ω D ×f D (p,q) (5); Where t(p,q) is the local cost from pixel p to its adjacent pixel q; G ,ω Z and ω D is the weighting coefficient; f G (q), f Z (q) and f D (q) is the gradient characteristic function, Laplace zero-crossing characteristic function and smoothness constraint function at the corresponding pixel q; S42. Use an optimal path search algorithm to find the shortest path between two points in a weighted directed graph as a boundary segment of the object, and perform edge segmentation.

3. The method for monitoring wetland soil microbial communities by hyperspectral remote sensing according to claim 1, characterized in that: In step S5, SVM is used to train the training set, and radial basis function is used as the kernel function used for SVM training.

Images