Suaeda salsa adaptive hyperspectral index and recognition method based on hyperspectral remote sensing image

Abstract

The invention relates to the field of coastal salt marsh wetland vegetation monitoring, and discloses a suaeda salsa adaptive hyperspectral index and identification method based on a hyperspectral remote sensing image. The method mainly aims at solving the problem of difficulty in identifying the suaeda salsa in the intertidal zone caused by serious phenomena of same object and different spectrum and same spectrum and foreign matter in the coastal wetland; the method comprises the following steps of: firstly, selecting representative samples and acquiring an average spectral reflectivity curve of the representative samples by considering different vegetation types, coverage difference and soil water-salt difference of the coastal wetland; then, analyzing the spectrum curve and selecting a hyperspectral image optimal wave band combination which is most suitable for distinguishing the suaeda salsa from other surface features; then, setting an adaptive rule and constructing a suaeda salsa adaptive hyperspectral index (SSAHI); and finally, quickly and accurately identifying and extracting the range of the suaeda salsa in the intertidal zone based on SSAHI. According to the method, the range of the suaeda salsa in the intertidal zone can be rapidly and accurately monitored based on the hyperspectral remote sensing images of different sensors and different time, and the problems that existing suaeda salsa identification based on the hyperspectral remote sensing images is difficult and low in precision are solved.

Description

Technical Field

[0001] The invention relates to the technical field of coastal salt marsh wetland vegetation monitoring, and in particular to a method for self-adapting hyperspectral index and identification of Suaeda salsa based on hyperspectral remote sensing images. Background Art

[0002] Coastal salt marshes play a vital role in shoreline protection, carbon sequestration, biodiversity conservation, and strengthening the resilience of coastal ecosystems. Suaeda salsa is an important pioneer salt marsh species in intertidal wetlands in East Asia and Europe. It has a short plant and often displays a red or purple-red hue, which is caused by increased accumulation of betalains due to high salinity and low temperatures. In the coastal areas of northern China, intertidal communities of Suaeda salsa form a beautiful landscape known as the "Red Beach" and are an important tourist attraction. However, coastal wetland ecosystems are becoming increasingly fragile due to the influence of natural and anthropogenic factors, resulting in the degradation of Suaeda salsa. The area of ​​Suaeda salsa in China has declined by about 60% in the past 30 years, which has attracted national attention, and many regional and local governments have initiated restoration projects aimed at restoring Suaeda salsa. Therefore, timely and accurate monitoring of Suaeda salsa is essential to support in-depth research and provide information for restoration decisions.

[0003] Although remote sensing technology has been widely used in vegetation classification of coastal wetlands, the current monitoring and research on Suaeda salsa mainly relies on multispectral satellite images. Given that the Suaeda salsa community shows considerable differences in coverage and plant height, and due to the fluctuation of soil moisture and salinity, the soil background of intertidal wetlands presents different spectral characteristics, which also makes it difficult to distinguish Suaeda salsa from the soil background. At the same time, compared with other vegetation types in coastal wetlands, the mapping accuracy of Suaeda salsa is much lower, mainly because low-density Suaeda salsa is easily confused with other green vegetation and difficult to distinguish from bare areas. Compared with multispectral data, hyperspectral images provide richer spectral details and can more accurately distinguish various vegetation species through spectral information. Although hyperspectral satellite images have been successfully applied to coastal vegetation classification, existing methods usually use complex classification algorithms and regard Suaeda salsa as one of the land cover categories. There is a lack of direct and rapid methods to identify Suaeda salsa. At the same time, existing methods often miss the identification of objects and have high misclassification rates due to the different spectral reflectances of different pixels in the same image. The above problems bring great difficulties and uncertainties to the identification and extraction of Suaeda salsa. Summary of the invention

[0004] 1. Technical issues to be resolved

[0005] In view of the problems existing in the above-mentioned prior art, the object of the present invention is to provide a method for adaptive hyperspectral index and identification of Suaeda salsa based on hyperspectral remote sensing images to solve the problems raised in the above-mentioned background technology:

[0006] (1) The problem of different spectra of the same objects and the same spectra of different objects, which is caused by the fragmentation of coastal wetland landscapes, large differences in vegetation coverage, and strong dynamic changes in soil water and salt, has been solved, thus restricting the accurate identification of Salicornia heracleifolia.

[0007] (2) It solves the problem of poor spectral index adaptability caused by spectral redundancy and spectral information overlap in hyperspectral data;

[0008] (3) The problem that the spectral index based on fixed bands cannot be applied to the accurate extraction of Salicornia herba under various vegetation coverage and soil background conditions was solved.

[0009] (II) Technical solution

[0010] In order to achieve the above object, the present invention provides the following technical solutions:

[0011] A method for adaptive hyperspectral index and identification of Suaeda salsa based on hyperspectral remote sensing images, comprising the following steps:

[0012] S1. Selection of representative samples considering the differences in vegetation types / coverage and soil water-salinity in coastal wetlands: Based on the hyperspectral remote sensing images of the intertidal zone of coastal salt marsh during the growing season, after pre-processing such as radiation calibration and atmospheric correction, sample points of Suaeda salsa and other green vegetation with different vegetation coverage and sample points of light beaches with different soil water contents were selected with the aid of high-resolution images and field surveys;

[0013] S2. Analysis of hyperspectral reflectance curves of different ground objects: Based on the selected samples, the average reflectance is calculated to obtain the hyperspectral reflectance curves of different ground objects, and the variation law of the spectral reflectance curves of different ground objects is analyzed. The difference between the spectral curves of Suaeda salsa and other green vegetation is analyzed, and the difference between it and the beach is analyzed. At the same time, the difference between the spectral reflectance curves of Suaeda salsa with different vegetation coverage is analyzed.

[0014] S3. Selection of the best band combination for hyperspectral images: In order to avoid band redundancy in hyperspectral images, the best band combination with the largest amount of information, the smallest redundancy and the most suitable for distinguishing Suaeda salsa was selected based on the reflectance curve, band correlation analysis, the best index method and the JM distance method;

[0015] S4, adaptive extreme band selection: Combined with the selected optimal band combination, according to the analysis results of the hyperspectral reflectance curves of different objects, by setting the dynamic functions max and min in the fixed band range respectively, making it traverse the different band ranges of each pixel, select the adaptive extreme band, and extract the band features that are most suitable for distinguishing Salicornia salsa from other objects pixel by pixel;

[0016] S5. Construction of Suaeda salsa adaptive hyperspectral index (SSAHI): Combining the optimal band combination and the change characteristics of the hyperspectral reflectance curve, the Suaeda salsa adaptive index (SSAHI) based on hyperspectral data is constructed by selecting adaptive extreme bands;

[0017] S6, SSAHI threshold method to extract the range of Suaeda salsa: Based on the NDVI index, the water area in the hyperspectral image of the growing season of Suaeda salsa in the intertidal zone is removed, and then the SSAHI index is calculated pixel by pixel in the non-water area. The classification accuracy is analyzed based on the training samples, and the optimal threshold is selected. It is considered that the area where the SSAHI is greater than the optimal threshold is the range of Suaeda salsa;

[0018] S7. Extraction of the range of Suaeda salsa by SSAHI random forest model: Based on the SSAHI calculation results and the sensitive bands of Suaeda salsa identification, the random forest model classification was performed based on the training samples to obtain the range of Suaeda salsa;

[0019] S8. Accuracy assessment and verification of the identification range of Suaeda salsa in the intertidal zone: Based on high-resolution images and field surveys, verification samples of Suaeda salsa and non-Suaeda salsa in the study area are selected to conduct accuracy assessment and verification of the identification and extraction results of the range of Suaeda salsa in the intertidal zone of the study area.

[0020] Preferably, in step S1, the representative sample selection considering the differences in vegetation types / coverage and soil water and salt in coastal wetlands specifically comprises the following steps:

[0021] S1.1. Based on high-resolution remote sensing images and field survey drone images, determine the approximate distribution range of Suaeda salsa and green vegetation, calculate the normalized difference vegetation index NDVI(1) within their respective ranges, and then classify Suaeda salsa and green vegetation into four different vegetation coverage levels based on the NDVI value range, and evenly select sample points;

[0022]

[0023] Among them, ρ 842 Represents the reflectivity value of 842nm in the near infrared band; ρ 670 Indicates the reflectance value of the red band at 670nm.

[0024] S1.2. Based on high-resolution remote sensing images and field survey drone images, determine the approximate distribution range of the light beach, calculate the normalized difference water index NDWI (2) within the range, and then divide the light beach into three different water content levels based on the value range of NDWI, and evenly select sample points;

[0025]

[0026] Among them, ρ534 Represents the reflectivity value of the green band 534nm; ρ 842 Indicates the reflectance value of 842nm in the near-infrared band.

[0027] S1.3. Evenly select sample points in different layers of the above-mentioned Suaeda salsa and green vegetation with different vegetation coverage and light beaches with different soil moisture contents to obtain a representative sample data set;

[0028] Preferably, in step S2, the hyperspectral reflectance curve analysis of different ground object samples comprises the following specific steps:

[0029] S2.1. According to the selected samples, the average spectral reflectance of each level sample point is obtained, that is, the high spectral reflectance spectrum curve of the level sample is obtained;

[0030] S2.2. Analyze the differences between water bodies and other land types: Compared with other land types, the spectral reflectance curves of water bodies in the near-infrared band and short-wave infrared band are much lower than those in the visible light band, and are also lower than the reflectance curves of other land types, so they can be well distinguished;

[0031] S2.3. Analyze the differences in spectral curves of different vegetation coverage levels or different water content levels within each land type: The spectral reflectance curves of different land types obtained all show obvious intra-class changes. With the increase of water content, the spectral reflectance of the beach gradually decreases; with the decrease of vegetation coverage, the visible light band reflectance of Suaeda salsa increases and the near-infrared reflectance decreases; similarly, the reflectance of green vegetation also shows higher visible light reflectance and lower near-infrared reflectance as the vegetation coverage decreases. At the same time, as the vegetation coverage decreases, the green wave peak of the visible light part of green vegetation gradually moves toward the long-wave direction;

[0032] S2.4. Analyze the differences between Suaeda salsa and green vegetation: Both Suaeda salsa and green vegetation show the trend of typical vegetation spectrum curve, but the peak of Suaeda salsa in visible light is in the red band and does not change with vegetation coverage. The red peak is at a central wavelength of 620nm, while the peak of green vegetation in visible light is in the green band (about central wavelength 499-576nm), and will move towards the long wave direction as the vegetation coverage decreases;

[0033] S2.5. Analyze the differences between Suaeda salsa and light beach: the reflectivity of light beach in near-infrared band is higher than that in visible light band, and the reflectivity of light beach in visible light band is much higher than that of Suaeda salsa. At the same time, Suaeda salsa shows the "steep slope" phenomenon of typical vegetation spectrum at around 700nm, while light beach does not have this spectral feature;

[0034] Preferably, in step S3, the optimal band combination of the hyperspectral image is selected by the following specific steps:

[0035] S3.1. Band correlation analysis, that is, based on hyperspectral images, the correlation coefficient is used to analyze the degree of information overlap between bands. The bands are divided into three subspaces with low information overlap and strong independence through the correlation coefficient analysis results;

[0036] S3.2, according to the hyperspectral curve formed in S2 and its analysis, select 620nm in subspace 1 as the most significant band to distinguish Salicornia from other landforms, and then form three combinations according to the subspace band correlation pattern: 620nm and subspace 2, 620nm and subspace 3, and 620nm and subspace 2, subspace 3, and traverse each possible band combination according to the grouping situation, and calculate the combined OIF (3). Compare the calculation results of the three groups of combinations, select the combination of 620nm and subspace 2 with the largest OIF value, and the different band combinations in it are the alternative optimal band combinations.

[0037]

[0038] Among them, S i represents the standard deviation of the reflectivity of the ith band; R ij Represents the correlation coefficient of the reflectivity of the two bands i and j.

[0039] S3.3, based on the sample points selected in S1, calculate the JM distance (Jeffreys-Matusita Distance) of the optimal band combination, and determine the separability of Suaeda salsa from other landforms. The combination with the largest value is selected as the optimal band combination, that is, the combination of 620nm and 860nm;

[0040] Preferably, in step S4, the adaptive extreme value band selection specifically comprises the following steps:

[0041] S4.1, Adaptive local maximum band selection: Based on the analysis of the spectral reflectance curves of Salicornia salsa and green vegetation in S2, they can be distinguished directly by the difference in the peaks within the visible light band. The red peaks of Salicornia salsa are all at a central wavelength of 620nm, and since the green peaks of green vegetation will change with the change of vegetation coverage, it is impossible to set a fixed green peak to represent the green peak position of each green vegetation pixel. Therefore, max(ρ i) function, by traversing the reflectance value of each wavelength in the range from the green band to 620nm (about the central wavelength is 499-611nm), find the position of the maximum reflectance of the visible light band of each pixel, and compare the reflectance with the reflectance value of the pixel at 620nm, so as to distinguish between saltwort and green vegetation. max(ρ i ) setting enables low-density green vegetation to be accurately identified and distinguished from Suaeda salsa;

[0042] S4.2, Adaptive local minimum band selection: Based on the analysis of the spectral reflectance curves of Salicornia salsa and the beach in S2, the two can be distinguished by the difference between the visible light band and the near-infrared band. The reflectance of Salicornia salsa in the visible light part is lower than that of the beach, and there is a local minimum reflectance between the central wavelength of 628nm and 851nm, and then it grows rapidly to the near-infrared range, while the reflectance of the beach changes much more slowly from the visible light to the near-infrared band. Therefore, the introduction of min(ρ j ) function, by traversing the reflectance value of each wavelength in the band with a central wavelength of 628nm to 851nm, the position of the local minimum reflectance of each pixel is found, and the difference comparison is combined with the 860nm selected in S2, so that the salt marsh sedge and the light beach can be distinguished.

[0043] Preferably, in step S5, the Suaeda salsa adaptive hyperspectral index (SSAHI) is constructed, and the SSAHI formula is:

[0044]

[0045] where i∈[499nm, 611nm], j∈[628nm, 851nm], L=0.5; ρ 620 , 860 Indicates the reflectivity value corresponding to the wavelengths of 620nm and 860nm in ZY1-02DAHSI; max(ρ i ) represents the maximum reflectivity in the range of 499nm to 611nm in ZY1-02DAHSI; min(ρ j ) represents the minimum reflectivity in the range of 628nm to 851nm in ZY1-02DAHSI. 620 -max(ρ i )),i∈[499nm,611nm] is used to distinguish Suaeda salsa with respect to green vegetation according to positive and negative; Used to distinguish between salt marsh sedge and light marsh according to the size of the value.

[0046] Preferably, in step S6, the water area in the hyperspectral image of the intertidal zone saltwort growing season is removed based on the NDVI index, that is, the area with a positive NDVI calculation result is taken as the non-water area, and the area with a negative value is taken as the water area, and the area is masked out.

[0047] Preferably, in step S6, the SSAHI threshold method extracts the range of saltwort suaea, specifically: randomly select several training samples of saltwort suaea and non-saltwort suaea in the non-water body area, with a step size of 0.01, and try each threshold in the range of 0.1 to 0.3 to extract the saltwort suaea area, and it is considered that the ones with SSAHI greater than the threshold are saltwort suaea, and the extraction results are subjected to precision analysis based on the selected training samples, and the threshold with the largest overall accuracy is selected as the optimal threshold for extracting saltwort suaea from the hyperspectral image, then the area with SSAHI greater than the threshold is the range of saltwort suaea.

[0048] Preferably, in step S7, the SSAHI random forest model extracts the range of Suaeda salsa, specifically: randomly selecting several training samples of Suaeda salsa and non-Suaeda salsa in non-water body areas, calculating the SSAHI of the hyperspectral image, and combining the sensitive bands of Suaeda salsa identification of 391-598nm, 658-709nm, 881-898nm, 967-984nm, 1001-1044nm and 1972-2002nm to perform random forest classification to obtain the range of Suaeda salsa.

[0049] Preferably, in step S8, the evaluation indicators used in the accuracy evaluation and verification of the intertidal saltwort salsa salsa identification range are overall accuracy, user accuracy, producer accuracy and Kappa coefficient.

[0050] Preferably, the calculation adopts any one of ENVI software, ArcGIS software, and MATLAB software; the ZY1-02D ​​AHSI image in the above steps can be replaced by hyperspectral remote sensing images such as GF5B AHSI and PRISMA, wherein GF5B can adopt the band reflectance of 618nm and 861nm, and PRISMA can adopt the band reflectance of 619nm and 860nm to replace the central wavelength of the band reflectance of 620nm and 860nm in SSAHI, and the remaining hyperspectral images can adopt other central wavelength ranges close to 620nm and 860nm; the above-mentioned high-resolution images include Jilin-1 images and PlanetScope images. At the same time, drone images of the study area were obtained through field surveys using DJI Phantom-4 multispectral drones.

[0051] The present invention has the following beneficial effects:

[0052] (1) The present invention not only enhances the separation between Suaeda salsa and other ground objects by using pixel-by-pixel adaptive band selection, but also reduces the spectral redundancy commonly seen in hyperspectral data by spectral combination selection;

[0053] (2) The Suaeda salsa adaptive hyperspectral index (SSAHI) constructed by the present invention can quickly and accurately identify the range of Suaeda salsa in the intertidal zone based on hyperspectral images;

[0054] (3) The present invention provides two methods, namely, threshold method and non-threshold method (random forest model), to use SSAHI to monitor the range of Suaeda salsa;

[0055] (4) The Suaeda salsa adaptive hyperspectral index (SSAHI) constructed by the present invention fully considers the spectral differences between salt marsh objects under the influence of different factors, sets the rules for pixel-by-pixel adaptive band selection, constructs the hyperspectral index SSAHI that can distinguish Suaeda salsa from green vegetation, beach, water body, etc., and selects the optimal band combination for constructing the index, avoiding data redundancy. The present invention not only solves the problem of lack of a method for quickly and accurately identifying Suaeda salsa in the intertidal zone of coastal salt marsh wetlands based on hyperspectral image data, but also effectively avoids the problem of difficulty in identifying Suaeda salsa in the intertidal zone caused by the serious phenomenon of different spectra of the same object and different objects of the same spectrum in coastal wetlands, provides a new idea for solving the problem of identifying and extracting the range of Suaeda salsa in the intertidal zone of coastal wetlands based on hyperspectral images, provides basic data support for subsequent in-depth research on the coverage and biomass inversion of Suaeda salsa in the intertidal zone, and has far-reaching significance for maintaining the stability of coastal wetland ecosystems, maintaining the habitat of Suaeda salsa, and restoring coastal wetlands based on Suaeda salsa. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. 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 creative work.

[0057] Figure 1 A schematic flow chart of a method for adaptive hyperspectral index and identification of Suaeda salsa L. based on hyperspectral remote sensing images according to the present invention;

[0058] Figure 2 : are the hyperspectral curves of different ground objects in Example 1, where (a) is the average reflectance spectral curve (Mean±SD) of four types of ground object samples; (b) is the spectral curve of different vegetation coverage levels of Suaeda salsa (SS) and green vegetation (GV) and different soil moisture content levels of bare beach (BF);

[0059] Figure 3This is a graph showing the correlation analysis results of the Yellow River Estuary image bands based on the ZY1-02DAHSI data in Example 1;

[0060] Figure 4 Detailed example of the spectrum curve in the range of 400nm to 1000nm in Example 1. The left shadow represents the wavelength range of 499nm to 611nm, and the right shadow represents the wavelength range of 628nm to 851nm. The triangles represent the max (ρ i ) and max(ρ j );

[0061] Figure 5 The optimal threshold selection results of SSAHI images in different study areas and at different times in Example 2. Among them, (a) is the SSAHI threshold selection result and accuracy based on the Resource-1 hyperspectral image of the Yellow River Estuary in 2020, 2022 and 2023, (b) is the SSAHI threshold selection result and accuracy based on the Resource-1 hyperspectral image of the Liaohe River Estuary in 2022 and 2023;

[0062] Figure 6 The feature importance ranking of SSAHI and other Suaeda salsa sensitive band features in Example 3 during random forest classification;

[0063] Figure 7 The extraction results of the Suaeda Salsa Adaptive Hyperspectral Index (SSAHI) based on the Resource No. 1 hyperspectral image of the Yellow River Estuary Study Area in 2020, 2022 and 2023 in Example 3. Wherein Suaeda Salsa represents Suaeda Salsa; (a) is the image of 2020, (b) is the image of 2022, and (c) is the image of 2023; sub-images I, II, III and IV under each image are the corresponding enlarged area sub-images;

[0064] Figure 8 The extraction results of the Suaeda Salsa Adaptive Hyperspectral Index (SSAHI) based on the Resource No. 1 hyperspectral image of the Liaohekou study area in 2022 and 2023 in Example 3. Where Suaeda Salsa represents Suaeda Salsa; (a) is the image in 2022, and (b) is the image in 2023; sub-images I, II, III and IV under each image are the corresponding enlarged area sub-images. DETAILED DESCRIPTION

[0065] In the following description, for the purpose of illustration rather than limitation, specific details such as specific system structures and technologies are provided to provide a thorough understanding of the embodiments of the present invention. However, it should be clear to those skilled in the art that the present invention may also be implemented in other embodiments without these specific details.

[0066] Example 1

[0067] Reference Figure 1-Figure 4 , a method for adaptive hyperspectral index and identification of Suaeda salsa based on hyperspectral remote sensing images, comprising the following steps:

[0068] S1. Selection of representative samples considering the differences in vegetation types / coverage and soil water-salinity in coastal wetlands: Based on the hyperspectral remote sensing images of the intertidal zone of coastal salt marsh during the growing season, after pre-processing such as radiation calibration and atmospheric correction, sample points of Suaeda salsa and other green vegetation with different vegetation coverage and sample points of light beaches with different soil water contents were selected based on high-resolution images and field survey assistance. The specific steps are:

[0069] S1.1. Based on high-resolution remote sensing images and field survey drone images, determine the approximate distribution range of Suaeda salsa and green vegetation, calculate the normalized difference vegetation index NDVI(1) within their respective ranges, and then classify Suaeda salsa and green vegetation into four different vegetation coverage levels based on the NDVI value range, and evenly select sample points;

[0070]

[0071] Among them, ρ 842 Represents the reflectivity value of 842nm in the near infrared band; ρ 670 Indicates the reflectance value of the red band at 670nm.

[0072] S1.2. Based on high-resolution remote sensing images and field survey drone images, determine the approximate distribution range of the light beach, calculate the normalized difference water index NDWI (2) within the range, and then divide the light beach into three different water content levels based on the value range of NDWI, and evenly select sample points;

[0073]

[0074] Among them, ρ 534 Represents the reflectivity value of the green band 534nm; ρ 842 Indicates the reflectance value of 842nm in the near-infrared band.

[0075] S1.3. Evenly select sample points in different layers of the above-mentioned Suaeda salsa and green vegetation with different vegetation coverage and light beaches with different soil moisture contents to obtain a representative sample data set;

[0076] S2. Analysis of hyperspectral reflectance curves of different ground objects: Based on the selected samples, the average reflectance is calculated to obtain the hyperspectral reflectance curves of different ground objects, and the changing rules of the spectral reflectance curves of different ground objects are analyzed. The difference between the spectral curves of Suaeda salsa and other green vegetation is analyzed, and the difference between it and the beach is analyzed. At the same time, the difference between the spectral reflectance curves of Suaeda salsa with different vegetation coverage is analyzed. The specific steps are:

[0077] S2.1. According to the selected samples, the average spectral reflectance of each level sample point is obtained, that is, the high spectral reflectance spectrum curve of the level sample is obtained;

[0078] S2.2. Analyze the differences between water bodies and other land types: Compared with other land types, the spectral reflectance curves of water bodies in the near-infrared band and short-wave infrared band are much lower than those in the visible light band, and are also lower than the reflectance curves of other land types, so they can be well distinguished;

[0079] S2.3. Analyze the differences in spectral curves of different vegetation coverage levels or different water content levels within each land type: The spectral reflectance curves of different land types obtained all show obvious intra-class changes. With the increase of water content, the spectral reflectance of the beach gradually decreases; with the decrease of vegetation coverage, the visible light band reflectance of Suaeda salsa increases and the near-infrared reflectance decreases; similarly, the reflectance of green vegetation also shows higher visible light reflectance and lower near-infrared reflectance as the vegetation coverage decreases. At the same time, as the vegetation coverage decreases, the green wave peak of the visible light part of green vegetation gradually moves toward the long-wave direction;

[0080] S2.4. Analyze the differences between Suaeda salsa and green vegetation: Both Suaeda salsa and green vegetation show the trend of typical vegetation spectrum curve, but the peak of Suaeda salsa in visible light is in the red band and does not change with vegetation coverage. The red peak is at a central wavelength of 620nm, while the peak of green vegetation in visible light is in the green band (about central wavelength 499-576nm), and will move towards the long wave direction as the vegetation coverage decreases;

[0081] S2.5. Analyze the differences between Suaeda salsa and light beach: the reflectivity of light beach in near-infrared band is higher than that in visible light band, and the reflectivity of light beach in visible light band is much higher than that of Suaeda salsa. At the same time, Suaeda salsa shows the "steep slope" phenomenon of typical vegetation spectrum at around 700nm, while light beach does not have this spectral feature;

[0082] S3. Selection of the best band combination for hyperspectral images: In order to avoid band redundancy in hyperspectral images, the best band combination with the largest amount of information, the smallest redundancy and the most suitable for distinguishing Suaeda salsa is selected based on the reflectance curve, band correlation analysis, the best index method and the JM distance method. The specific steps are:

[0083] S3.1. Band correlation analysis, that is, based on hyperspectral images, the correlation coefficient is used to analyze the degree of information overlap between bands. The bands are divided into three subspaces with low information overlap and strong independence through the correlation coefficient analysis results;

[0084] S3.2, according to the hyperspectral curve formed in S2 and its analysis, select 620nm in subspace 1 as the most significant band to distinguish Salicornia from other landforms, and then form three combinations according to the subspace band correlation pattern: 620nm and subspace 2, 620nm and subspace 3, and 620nm and subspace 2, subspace 3, and traverse each possible band combination according to the grouping situation, and calculate the combined OIF (3). Compare the calculation results of the three groups of combinations, select the combination of 620nm and subspace 2 with the largest OIF value, and the different band combinations in it are the alternative optimal band combinations.

[0085]

[0086] Among them, S i represents the standard deviation of the reflectivity of the ith band; R ij Represents the correlation coefficient of the reflectivity of the two bands i and j.

[0087] S3.3, based on the sample points selected in S1, calculate the JM distance (Jeffreys-Matusita Distance) of the optimal band combination, and determine the separability of Suaeda salsa from other landforms. The combination with the largest value is selected as the optimal band combination, that is, the combination of 620nm and 860nm;

[0088] S4, Adaptive extreme band selection: Combined with the selected optimal band combination, according to the analysis results of the hyperspectral reflectance curves of different objects, by setting the dynamic functions max and min in the fixed band range, it traverses the different band ranges of each pixel, selects the adaptive extreme band, and extracts the band features that are most suitable for distinguishing Salicornia salsa from other objects pixel by pixel. The specific steps are:

[0089] S4.1, Adaptive local maximum band selection: Based on the analysis of the spectral reflectance curves of Salicornia salsa and green vegetation in S2, they can be distinguished directly by the difference in the peaks within the visible light band. The red peaks of Salicornia salsa are all at a central wavelength of 620nm, and since the green peaks of green vegetation will change with the change of vegetation coverage, it is impossible to set a fixed green peak to represent the green peak position of each green vegetation pixel. Therefore, max(ρ i ) function, by traversing the reflectance value of each wavelength in the range from the green band to 620nm (about the central wavelength is 499-611nm), find the position of the maximum reflectance of the visible light band of each pixel, and compare the reflectance with the reflectance value of the pixel at 620nm, so as to distinguish between saltwort and green vegetation. max(ρ i ) setting enables low-density green vegetation to be accurately identified and distinguished from Suaeda salsa;

[0090] S4.2, Adaptive local minimum band selection: Based on the analysis of the spectral reflectance curves of Salicornia salsa and the beach in S2, the two can be distinguished by the difference between the visible light band and the near-infrared band. The reflectance of Salicornia salsa in the visible light part is lower than that of the beach, and there is a local minimum reflectance between the central wavelength of 628nm and 851nm, and then it grows rapidly to the near-infrared range, while the reflectance of the beach changes much more slowly from the visible light to the near-infrared band. Therefore, the introduction of min(ρ j ) function, by traversing the reflectance value of each wavelength in the band with a central wavelength of 628nm to 851nm, the position of the local minimum reflectance of each pixel is found, and the difference comparison is made with the 860nm selected in S2, so that the salt marsh sedge and the light beach can be distinguished;

[0091] S5. Construction of Suaeda salsa adaptive hyperspectral index (SSAHI): Combining the optimal band combination and the change characteristics of the hyperspectral reflectance curve, the Suaeda salsa adaptive index (SSAHI) based on hyperspectral data is constructed by selecting adaptive extreme bands. The formula is:

[0092]

[0093] where i∈[499nm, 611nm], j∈[628nm, 851nm], L=0.5; ρ 620 , 860 Indicates the reflectivity value corresponding to the wavelengths of 620nm and 860nm in ZY1-02DAHSI; max(ρ i ) represents the maximum reflectivity in the range of 499nm to 611nm in ZY1-02DAHSI; min(ρj ) represents the minimum reflectivity in the range of 628nm to 851nm in ZY1-02DAHSI. 620 -max(ρ i )), i∈[499nm,611nm] is used to distinguish Suaeda salsa with respect to green vegetation according to positive and negative; Used to distinguish between salt marsh sedge and light marsh according to the size of the value.

[0094] Example 2

[0095] In this embodiment, the Yellow River Delta Nature Reserve (Yellow River Estuary) and the Liaohe River Estuary Nature Reserve are selected as the study areas. Both the Yellow River Estuary and the Liaohe River Estuary belong to the intertidal zone of coastal wetlands, where saltwort salsa grows. We selected the Resource No. 1 hyperspectral images in the growing season in 2020, 2022, and 2023 in the Yellow River Estuary study area and 2022 and 2023 in the Liaohe River Estuary, and used the threshold method to identify the range of saltwort salsa.

[0096] Reference Figure 5 , a method for adaptive hyperspectral index and identification of Suaeda salsa based on hyperspectral remote sensing images, comprising the following steps:

[0097] S1. Remove the water area in the hyperspectral image of the intertidal zone during the growing season of Suaeda salsa based on the NDVI index, that is, use ArcGIS software or ENVI software to take the area with positive NDVI calculation results as the non-water area and the area with negative NDVI calculation results as the water area, and remove the mask of this area;

[0098] S2. Use ENVI software and MATLAB software to calculate the SSAHI value of the image and generate the SSAHI distribution results of the corresponding area;

[0099] S3. Identify and extract the range of the intertidal zone saltwort based on the threshold method, specifically:

[0100] S3.1. Based on high-resolution remote sensing images and field survey drone images, 1,600 samples of Suaeda salsa and non-Suaeda salsa were randomly selected from each image, and 1,000 of them were used as training samples and 600 were used as subsequent accuracy verification samples. The 1,600 samples included 500 samples of Suaeda salsa and 1,100 samples of non-Suaeda salsa.

[0101] S3.2, with a step size of 0.01, in the range of 0.1 to 0.3, try to extract the Suaeda salsa area at each threshold, and consider the area with SSAHI greater than the threshold as Suaeda salsa;

[0102] S3.3. Use the training samples to calculate the overall accuracy of each threshold extraction result, and select the one with the highest accuracy as the optimal threshold of SSAHI for the image, that is, the thresholds for the Yellow River Estuary in 2020, 2022, and 2023 are 0.22, 0.24, and 0.25 respectively, and the thresholds for the Liaohe River Estuary in 2022 and 2023 are both 0.20;

[0103] S3.4, using ENVI software, segment SSAHI with the extracted optimal threshold, and the part larger than the threshold is the range of Suaeda salsa extracted by the threshold method;

[0104] S4. Based on the 600 verification samples in S3.1, the accuracy of the extracted Suaeda salsa range is verified to evaluate the accuracy of the extracted Suaeda salsa range. The accuracy evaluation indicators are overall accuracy (OA), user accuracy (UA), producer accuracy (PA) and Kappa coefficient.

[0105] Example 3

[0106] In this embodiment, the Yellow River Delta Nature Reserve (Yellow River Estuary) and the Liaohe River Estuary Nature Reserve are selected as the study areas. Both the Yellow River Estuary and the Liaohe River Estuary belong to the intertidal zone of coastal wetlands, where Suaeda salsa grows. We selected the Resource No. 1 hyperspectral images in the growing season in 2020, 2022, and 2023 in the Yellow River Estuary study area and 2022 and 2023 in the Liaohe River Estuary, and used the SSAHI-based random forest model to identify the range of Suaeda salsa.

[0107] Reference Figure 6-Figure 8 , a method for adaptive hyperspectral index and identification of Suaeda salsa based on hyperspectral remote sensing images, comprising the following steps:

[0108] S1. Remove the water area in the hyperspectral image of the intertidal zone during the growing season of Suaeda salsa based on the NDVI index, that is, use ArcGIS software or ENVI software to take the area with positive NDVI calculation results as the non-water area and the area with negative NDVI calculation results as the water area, and remove the mask of this area;

[0109] S2. Use ENVI software and MATLAB software to calculate the SSAHI value of the image and generate the SSAHI distribution results of the corresponding area;

[0110] S3. Based on SSAHI and random forest model, the range of Suaeda salsa in intertidal zone is identified and extracted, specifically:

[0111] S3.1. Based on high-resolution remote sensing images and field survey drone images, 1,600 samples of Suaeda salsa and non-Suaeda salsa were randomly selected from each image, and 1,000 of them were used as training samples and 600 were used as subsequent accuracy verification samples. The 1,600 samples included 500 samples of Suaeda salsa and 1,100 samples of non-Suaeda salsa.

[0112] S3.2, combine SSAHI with the sensitive bands of 391-598nm, 658-709nm, 881-898nm, 967-984nm, 1001-1044nm and 1972-2002nm for identifying Suaeda salsa in the intertidal zone in the existing research on Resource-1 hyperspectral image, perform random forest classification based on the selected training samples, and extract the range of Suaeda salsa;

[0113] S4. Based on the 600 verification samples in S3.1, the accuracy of the extracted Suaeda salsa range is verified to evaluate the accuracy of the extracted Suaeda salsa range. The accuracy evaluation indicators are overall accuracy (OA), user accuracy (UA), producer accuracy (PA) and Kappa coefficient.

[0114] S5. The SHapleyAdditive exPlanations (SHAP) value was used to characterize the importance of each feature in the random forest classifier. It was found that SSAHI was the most important feature for identifying Suaeda salsa in any image, indicating that this index made an important contribution to the identification and extraction of the range of Suaeda salsa in the intertidal zone of coastal salt marshes.

[0115] Specifically, the ZY1-02DAHSI image in the above steps can be replaced by hyperspectral remote sensing images such as GF5B AHSI and PRISMA, among which GF5B can use the band reflectance of 618nm and 861nm, PRISMA can use the band reflectance of 619nm and 860nm to replace the central wavelengths of the band reflectance of 620nm and 860nm in SSAHI, and the remaining hyperspectral images can use other central wavelength ranges close to 620nm and 860nm; the above-mentioned high-resolution images include Jilin-1 images and PlanetScope images. At the same time, drone images of the study area were obtained through field surveys using the DJI Phantom-4 multispectral drone.

[0116] The present invention is not limited to the above-mentioned specific implementation modes. Various changes made by ordinary technicians in this field based on the above-mentioned concepts without creative work are all within the protection scope of the present invention.

Claims

1. A method for adaptive hyperspectral index and identification of Suaeda salsa based on hyperspectral remote sensing images, characterized in that: The following steps are involved: S1. Selection of representative samples considering the differences in vegetation types / coverage and soil water-salinity in coastal wetlands: Based on the hyperspectral remote sensing images of the intertidal zone of coastal salt marsh during the growing season, after pre-processing such as radiation calibration and atmospheric correction, sample points of Suaeda salsa and other green vegetation with different vegetation coverage and sample points of light beaches with different soil water contents were selected with the aid of high-resolution images and field surveys; S2. Analysis of hyperspectral reflectance curves of different ground objects: Based on the selected samples, the average reflectance is calculated to obtain the hyperspectral reflectance curves of different ground objects, and the variation law of the spectral reflectance curves of different ground objects is analyzed. The difference between the spectral curves of Suaeda salsa and other green vegetation is analyzed, and the difference between it and the beach is analyzed. At the same time, the difference between the spectral reflectance curves of Suaeda salsa with different vegetation coverage is analyzed. S3. Selection of the best band combination for hyperspectral images: In order to avoid band redundancy in hyperspectral images, the best band combination with the largest amount of information, the smallest redundancy and the most suitable for distinguishing Suaeda salsa was selected based on the reflectance curve, band correlation analysis, the best index method and the JM distance method; S4, adaptive extreme band selection: Combined with the selected optimal band combination, according to the analysis results of the hyperspectral reflectance curves of different objects, by setting the dynamic functions max and min in the fixed band range respectively, making it traverse the different band ranges of each pixel, select the adaptive extreme band, and extract the band features that are most suitable for distinguishing Salicornia salsa from other objects pixel by pixel; S5. Construction of Suaeda salsa adaptive hyperspectral index (SSAHI): Combining the optimal band combination and the change characteristics of the hyperspectral reflectance curve, the Suaeda salsa adaptive index SSAHI based on hyperspectral data is constructed by selecting adaptive extreme bands; S6, SSAHI threshold method to extract the range of Suaeda salsa: Based on the NDVI index, the water area in the hyperspectral image of the growing season of Suaeda salsa in the intertidal zone is removed, and then the SSAHI index is calculated pixel by pixel in the non-water area. The classification accuracy is analyzed based on the training samples, and the optimal threshold is selected. It is considered that the area where the SSAHI is greater than the optimal threshold is the range of Suaeda salsa; S7. Extraction of the range of Suaeda salsa by SSAHI random forest model: Based on the SSAHI calculation results and the sensitive bands of Suaeda salsa identification, the random forest model classification was performed based on the training samples to obtain the range of Suaeda salsa; S8. Accuracy assessment and verification of the identification range of Suaeda salsa in the intertidal zone: Based on high-resolution images and field surveys, verification samples of Suaeda salsa and non-Suaeda salsa in the study area are selected to conduct accuracy assessment and verification of the identification and extraction results of the range of Suaeda salsa in the intertidal zone of the study area.

2. The method for adaptive hyperspectral index and identification of Suaeda salsa based on hyperspectral remote sensing images according to claim 1, characterized in that: In step S1, the representative sample selection considering the differences in vegetation types / coverage and soil water and salt in coastal wetlands specifically includes the following steps: S1.

1. Based on high-resolution remote sensing images and field survey drone images, determine the approximate distribution range of Suaeda salsa and green vegetation, calculate the normalized difference vegetation index NDVI(1) within their respective ranges, and then classify Suaeda salsa and green vegetation into four different vegetation coverage levels based on the NDVI value range, and evenly select sample points; Among them, ρ 842 Represents the reflectivity value of 842nm in the near infrared band; ρ 670 Indicates the reflectance value of the red band at 670nm. S1.

2. Based on high-resolution remote sensing images and field survey drone images, determine the approximate distribution range of the light beach, calculate the normalized difference water index NDWI (2) within the range, and then divide the light beach into three different water content levels based on the value range of NDWI, and evenly select sample points; Among them, ρ 534 Represents the reflectivity value of the green band 534nm; ρ 842 Indicates the reflectance value of 842nm in the near-infrared band. S1.

3. Sample points are evenly selected from different levels of the above-mentioned Salicornia herba and green vegetation with different vegetation coverage and light beaches with different soil moisture contents to obtain a representative sample data set.

3. The method for adaptive hyperspectral index and identification of Suaeda salsa based on hyperspectral remote sensing images according to claim 1, characterized in that: In step S2, the hyperspectral reflectance curve analysis of different ground object samples is carried out in the following specific steps: S2.

1. According to the selected samples, the average spectral reflectance of each level sample point is obtained, that is, the high spectral reflectance spectrum curve of the level sample is obtained; S2.

2. Analyze the differences between water bodies and other land types: Compared with other land types, the spectral reflectance curves of water bodies in the near-infrared band and short-wave infrared band are much lower than those in the visible light band, and are also lower than the reflectance curves of other land types, so they can be well distinguished; S2.

3. Analyze the differences in spectral curves of different vegetation coverage levels or different water content levels within each land type: The spectral reflectance curves of different land types obtained all show obvious intra-class changes. With the increase of water content, the spectral reflectance of the beach gradually decreases; with the decrease of vegetation coverage, the visible light band reflectance of Suaeda salsa increases and the near-infrared reflectance decreases; similarly, the reflectance of green vegetation also shows higher visible light reflectance and lower near-infrared reflectance as the vegetation coverage decreases. At the same time, as the vegetation coverage decreases, the green wave peak of the visible light part of green vegetation gradually moves toward the long-wave direction; S2.

4. Analyze the differences between Suaeda salsa and green vegetation: Both Suaeda salsa and green vegetation show the trend of typical vegetation spectrum curve, but the peak of Suaeda salsa in visible light is in the red band and does not change with vegetation coverage. The red peak is at a central wavelength of 620nm, while the peak of green vegetation in visible light is in the green band (about central wavelength 499-576nm), and will move towards the long wave direction as the vegetation coverage decreases; S2.

5. Analyze the differences between Suaeda salsa and light beach: the reflectivity of light beach in near-infrared band is higher than that in visible light band, and the reflectivity of light beach in visible light band is much higher than that of Suaeda salsa. At the same time, Suaeda salsa shows the "steep slope" phenomenon of typical vegetation spectrum at around 700nm, while light beach does not have this spectral feature.

4. The method for adaptive hyperspectral index and identification of Suaeda salsa based on hyperspectral remote sensing images according to claim 1, characterized in that: In step S3, the optimal band combination of the hyperspectral image is selected, and the specific steps are as follows: S3.

1. Band correlation analysis, that is, based on hyperspectral images, the correlation coefficient is used to analyze the degree of information overlap between bands. The bands are divided into three subspaces with low information overlap and strong independence through the correlation coefficient analysis results; S3.2, according to the hyperspectral curve formed in S2 and its analysis, select 620nm in subspace 1 as the most significant band to distinguish Salicornia from other landforms, and then form three combinations according to the subspace band correlation pattern: 620nm and subspace 2, 620nm and subspace 3, and 620nm and subspace 2, subspace 3, and traverse each possible band combination according to the grouping situation, and calculate the combined OIF (3). Compare the calculation results of the three groups of combinations, select the combination of 620nm and subspace 2 with the largest OIF value, and the different band combinations in it are the alternative optimal band combinations. Among them, S i represents the standard deviation of the reflectivity of the ith band; R ij Represents the correlation coefficient of the reflectance of bands i and j. S3.

3. Based on the sample points selected in S1, calculate the JM distance (Jeffreys-MatusitaDistance) of the optimal alternative band combination to determine the separability of Suaeda salsa from other landforms. The combination with the largest value is selected as the optimal band combination, that is, the combination of 620nm and 860nm.

5. The method for adaptive hyperspectral index and identification of Suaeda salsa based on hyperspectral remote sensing images according to claim 1, characterized in that: In step S4, the adaptive extreme value band selection specifically includes the following steps: S4.1, Adaptive local maximum band selection: Based on the analysis of the spectral reflectance curves of Salicornia salsa and green vegetation in S2, they can be distinguished directly by the difference in the peaks within the visible light band. The red peaks of Salicornia salsa are all at a central wavelength of 620nm, and since the green peaks of green vegetation will change with the change of vegetation coverage, it is impossible to set a fixed green peak to represent the green peak position of each green vegetation pixel. Therefore, max(ρ i ) function, by traversing the reflectance value of each wavelength in the range from the green band to 620nm (about the central wavelength is 499-611nm), find the position of the maximum reflectance of the visible light band of each pixel, and compare the reflectance with the reflectance value of the pixel at 620nm, so as to distinguish between saltwort and green vegetation. max(ρ i ) setting enables low-density green vegetation to be accurately identified and distinguished from Suaeda salsa; S4.2, Adaptive local minimum band selection: Based on the analysis of the spectral reflectance curves of Salicornia salsa and the beach in S2, the two can be distinguished by the difference between the visible light band and the near-infrared band. The reflectance of Salicornia salsa in the visible light part is lower than that of the beach, and there is a local minimum reflectance between the central wavelength of 628nm and 851nm, and then it grows rapidly to the near-infrared range, while the reflectance of the beach changes much more slowly from the visible light to the near-infrared band. Therefore, the introduction of min(ρ j ) function, by traversing the reflectance value of each wavelength in the band with a central wavelength of 628nm to 851nm, the position of the local minimum reflectance of each pixel is found, and the difference comparison is combined with the 860nm selected in S2, so that the salt marsh sedge and the light beach can be distinguished.

6. The method for adaptive hyperspectral index and identification of Suaeda salsa based on hyperspectral remote sensing images according to claim 1, characterized in that: In step S5, the Suaeda salsa adaptive hyperspectral index (SSAHI) is constructed, and the SSAHI formula is: where i∈[499nm, 611nm], j∈[628nm, 851nm], L=0.5; ρ 620 , 860 Indicates the reflectivity value corresponding to the wavelengths of 620nm and 860nm in ZY1-02DAHSI; max(ρ i ) represents the maximum reflectivity in the range of 499nm to 611nm in ZY1-02DAHSI; min(ρ j ) represents the minimum reflectivity in the range of 628nm to 851nm in ZY1-02DAHSI. 620 -max(ρ i )), i∈[499nm,611nm] is used to distinguish Suaeda salsa with respect to green vegetation according to positive and negative; Used to distinguish between salt marsh sedge and light marsh according to the size of the value.

7. The method for adaptive hyperspectral index and identification of Suaeda salsa based on hyperspectral remote sensing images according to claim 1, characterized in that: In step S6, the water area in the hyperspectral image of the intertidal zone saltwort growing season is removed based on the NDVI index, that is, the area with a positive NDVI calculation result is taken as the non-water area, and the area with a negative NDVI calculation result is taken as the water area, and the area is masked and removed.

8. The method for adaptive hyperspectral index and identification of Suaeda salsa based on hyperspectral remote sensing images according to claim 1, characterized in that: In step S6, the SSAHI threshold method extracts the range of saltwort salsa, specifically: randomly select several training samples of saltwort salsa and non-salsa salsa in the non-water area, with a step size of 0.01, in the range of 0.1 to 0.3, try each threshold to extract the saltwort salsa area, it is considered that the area with SSAHI greater than the threshold is saltwort salsa, and the extraction result is subjected to precision analysis according to the selected training samples, and the threshold with the largest overall accuracy is selected as the optimal threshold for extracting saltwort salsa from the hyperspectral image, then the area with SSAHI greater than the threshold is the saltwort salsa range.

9. The method for adaptive hyperspectral index and identification of Suaeda salsa based on hyperspectral remote sensing images according to claim 1, characterized in that: In step S7, the SSAHI random forest model extracts the range of Suaeda salsa, specifically: randomly selecting several training samples of Suaeda salsa and non-Suaeda salsa in non-water areas, calculating the SSAHI of the hyperspectral image, and combining the sensitive bands of 391-598nm, 658-709nm, 881-898nm, 967-984nm, 1001-1044nm and 1972-2002nm identified by Suaeda salsa for random forest classification to obtain the range of Suaeda salsa.

10. The method for adaptive hyperspectral index and identification of Suaeda salsa based on hyperspectral remote sensing images according to claim 1, characterized in that: In step S8, the evaluation indicators used in the accuracy evaluation and verification of the identification range of the intertidal zone saltwort salsa include overall accuracy, user accuracy, producer accuracy and Kappa coefficient.

11. The method for adaptive hyperspectral index and identification of Suaeda salsa based on hyperspectral remote sensing images according to claim 1, characterized in that: The calculation adopts any one of ENVI software, ArcGIS software, and MATLAB software; the ZY1-02D ​​AHSI image in the above steps can be replaced by GF5B AHSI, PRISMA and other hyperspectral remote sensing images, among which GF5B can adopt the band reflectance of 618nm and 861nm, PRISMA can adopt the band reflectance of 619nm and 860nm to replace the central wavelength of the band reflectance of 620nm and 860nm in SSAHI, and the remaining hyperspectral images can adopt other central wavelength ranges close to 620nm and 860nm; the above high-resolution images include Jilin-1 images and PlanetScope images. At the same time, drone images of the study area were obtained through field surveys using DJI Phantom-4 multispectral drones.

Images