Spatial distribution extraction method, terminal equipment and storage medium of Reedia spp.

Abstract

This invention discloses a method, terminal device, and storage medium for extracting the spatial distribution of *Reedia spp.*. Based on the differences in vegetation index characteristics, spectral characteristics, texture characteristics, and spatial characteristics of different land cover features in wetland images at different time phases, the optimal segmentation scale is determined for image segmentation. A wetland land cover classification rule set is constructed based on the spectral values, shape, texture, and other indirect features of the segmented objects, thereby realizing the extraction of *Reedia spp.* spatial distribution information. This invention identifies three optimal segmentation scales of different sizes, analyzes the vertical relationships between land cover types at different scales and the horizontal relationships between land cover types within the same layer resulting from segmentation at the same scale, and establishes three classification levels from large to small, which to some extent solves the influence of smaller image patches and improves classification accuracy.

Description

Technical Field

[0001] This invention relates to the field of agricultural / forestry remote sensing technology, and in particular to a method for extracting the spatial distribution of Reed ferox, a terminal device, and a storage medium. Background Technology

[0002] *Miscanthus sinensis*, a species of Miscanthus endemic to my country, belongs to the subfamily Sorghum of the family Poaceae. It is concentrated in the Dongting Lake wetland and is found only in the middle and lower reaches of the Yangtze River in my country. There is a lack of international research on its ecological conservation, and this unique and valuable species has not received sufficient attention domestically. Traditional monitoring of the spatial distribution and area of ​​*Miscanthus sinensis* relies primarily on manual ground surveys and reporting. With the development of remote sensing technology, scholars have used supervised classification, unsupervised classification, and decision tree classification methods based on spectral differences of ground features to study wetland vegetation classification. Due to the complex internal structure of wetlands, the spectral characteristics of wetland vegetation in remote sensing images are greatly influenced by the wetland environment, resulting in the common phenomena of "same spectrum, different vegetation" and "same vegetation, different spectra." Traditional classification methods relying on spectral differences are no longer suitable for wetland vegetation classification. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide a method, terminal equipment and storage medium for extracting the spatial distribution of Reed ferns, which addresses the shortcomings of the existing technology and improves the accuracy of wetland vegetation classification and identification.

[0004] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a method for extracting the spatial distribution of *Reedia spp.*, comprising the following steps:

[0005] S1. Divide the remote sensing image into study area and non-study area;

[0006] S2. Select the first segmentation scale, segment the image of the study area using the first segmentation scale, determine the optimal threshold range of each feature parameter of each land cover, and extract lakes, rivers, large areas of reeds, and large areas of grassland in sequence according to the different optimal threshold ranges of each feature parameter. The remaining range is classified as unclassified.

[0007] S3. Select the second segmentation scale and use the second segmentation scale to further segment the unclassified image, determine the optimal threshold range of each feature parameter of each land cover, and extract ditches, small patches of reeds, small patches of grassland, woodland, roads, and white mud islands in sequence according to the different optimal threshold ranges of each feature parameter. The remaining range is classified as unclassified.

[0008] S4. Select the third segmentation scale and use the third segmentation scale to further segment the unclassified image obtained in step S3. Determine the optimal threshold range of each feature parameter of each land type. Based on the different optimal threshold ranges of each feature parameter, extract waterlogged areas / ponds and buildings in sequence. The remaining range is classified as unclassified.

[0009] S5. Large areas of reeds and small areas of reeds are classified as reeds; lakes, rivers, ditches, and waterlogged depressions / ponds are classified as water bodies; large areas of grassland and small areas of grassland are classified as grassland; roads and buildings are classified as construction land; and Baini Island and unclassified areas are classified as other land.

[0010] S6. For any land use type, merge the image elements smaller than a set value into other land use types, wherein the land use types are: reeds, water bodies, grasslands, forests, construction land, and other land use.

[0011] S7. Smooth the land types processed in step S6 to obtain the final classification result.

[0012] This invention identifies three optimal segmentation scales of varying sizes, analyzes the vertical relationships between land cover types at different scales and the horizontal relationships between land cover types within the same scale segmentation, establishing three classification hierarchies from largest to smallest. This addresses the impact of smaller patches to some extent, improving classification accuracy. Furthermore, this invention comprehensively considers multi-dimensional remote sensing features of vegetation, including phenological, spectral, spatial, and textural characteristics, determining appropriate classification feature parameters and establishing scientifically sound classification rules, further enhancing classification accuracy. The method of this invention significantly improves the accuracy of wetland vegetation classification and identification.

[0013] In this invention, the characteristic parameters include: Normalized Difference Vegetation Index (NDVI), Normalized Difference Water Index (NDWI), Greenness Index (RG), Red Band Reflectance (R), Near Infrared Band Reflectance (NIR), Brightness, Shape Index (SI), Aspect Ratio (L / W), Rectangularity (RF), Saturation, Intensity, and Homogeneity.

[0014] In this invention, the formulas for calculating the Normalized Difference Vegetation Index (NDVI), Normalized Difference Water Index (NDWI), and Greenness Index (RG) are as follows: NIR is the reflectance in the near-infrared band, and R is the reflectance in the red band. G represents the green band reflectivity; B represents the blue band reflectivity.

[0015] In this invention, the first segmentation scale > the second segmentation scale > the third segmentation scale.

[0016] Furthermore, in this invention, the first segmentation scale is 220, the second segmentation scale is 130, and the third segmentation scale is 60. The first segmentation scale segments the entire image, extracting large-area features such as lakes, rivers, large areas of reeds, and large grasslands. The second segmentation scale further segments objects with blurred or low recognition ranges under the first segmentation scale, extracting relatively small features such as small patches of reeds, small patches of lake grass, flood discharge channels, roads, and woodlands. The third segmentation scale further refines objects with low recognition and small areas, classifying land types such as buildings, waterlogged areas / ponds with higher accuracy. The third segmentation scale focuses on correcting misclassification and omissions through visual interpretation.

[0017] To further improve classification accuracy, the specific implementation process for determining the optimal threshold range of each feature parameter in this invention includes:

[0018] Select a feature parameter, and display the value of that feature parameter in the grayscale image, and use color to indicate the feature range;

[0019] The lower and upper limits of the feature parameter value are continuously adjusted until the remaining coverage area of ​​the feature parameter value in the grayscale image is the target extracted land type;

[0020] Select the minimum and maximum values ​​displayed in the feature parameter view window as the optimal threshold range for extracting the current land cover type.

[0021] In this invention, the specific implementation process of step S6 includes:

[0022] For any land type, set a conditional function, i.e., Number of pixels ≤ a, where Number of pixels represents the number of pixels in the patch and a represents the specific value set.

[0023] The land parcels that satisfy the condition function in the land category are merged into other adjacent land categories.

[0024] In this invention, "adjacent other land types" refers to other land types adjacent to the boundary of the target land type. For example, if a small water feature is surrounded by grassland, woodland, or reeds, then the adjacent land types for the water feature are grassland, woodland, and reeds.

[0025] For reeds, water bodies, grasslands, and woodlands, a = 25; for construction land and other land uses, a = 9.

[0026] As an inventive concept, the present invention also provides a terminal device, including a memory, a processor, and a computer program stored in the memory; the processor executes the computer program to implement the steps of the method described above.

[0027] As an inventive concept, the present invention also provides a computer-readable storage medium having a computer program / instructions stored thereon; when the computer program / instructions are executed by a processor, they implement the steps of the method described above.

[0028] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0029] (1) This invention determines three optimal segmentation scales of different sizes, analyzes the vertical relationship between land types in different scale layers and the horizontal relationship between land types in the same layer generated by segmentation at the same scale, and establishes three classification levels from large to small, which to a certain extent solves the influence of smaller patches and improves classification accuracy.

[0030] (2) In view of the problem that the wetland vegetation is classified in low accuracy due to the diverse types of vegetation, complex topography and landforms and large seasonal water level changes, this invention comprehensively considers the multi-dimensional remote sensing features of vegetation such as phenological features, spectral features, spatial features and texture features, determines appropriate classification feature parameters, establishes a scientific and reasonable set of classification rules, and improves classification accuracy.

[0031] (3) This invention takes into account the changes in biomass and growth environment during the growth period of Dioscorea opposita, as well as the distinguishability of Dioscorea opposita from other vegetation, and makes full use of the growth characteristics of Dioscorea opposita during key periods, so as to reflect the dynamic growth process of Dioscorea opposita with static images. Attached Figure Description

[0032] Figure 1 This is a flowchart of the method in Embodiment 1 of the present invention;

[0033] Figure 2 Experimental diagram for selecting the optimal segmentation scale in embodiments of the present invention;

[0034] Figure 3 This is a ROC plot of an embodiment of the present invention, where A is a small segmentation scale, B is a medium segmentation scale, and C is a large segmentation scale;

[0035] Figure 4 This is a comparison of hyperspectral spectral curves of different ground features at the same time phase in an embodiment of the present invention;

[0036] Figures 5(a) to 5(d) The images are of *Imperata cylindrica*, *Carex spp.*, *Leymus chinensis*, and their measured hyperspectral curves.

[0037] Figure 6 This is a satellite image from April of an embodiment of the present invention;

[0038] Figure 7 Satellite imagery from August, as described in this embodiment of the invention;

[0039] Figure 8 This is a satellite image from December of an embodiment of the present invention;

[0040] Figure 9 This is an NDVI time-series image from an embodiment of the present invention. Detailed Implementation

[0041] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions in the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this invention, not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0042] In this document, the terms "first," "second," and other similar words are not intended to imply any order, quantity, or importance, but are merely used to distinguish different elements. The terms "one," "a," and other similar words are not intended to indicate the existence of only one of the stated things, but rather that the description pertains to only one of the two stated things, which may include one or more. The terms "comprising," "including," and other similar words are intended to indicate a logical relationship, not a spatial relationship. For example, "A includes B" means that logically B belongs to A, not that spatially B is located inside A. Furthermore, the meanings of the terms "comprising," "including," and other similar words should be considered open-ended, not closed. For example, "A includes B" means that B belongs to A, but B does not necessarily constitute all of A; A may also include other elements such as C, D, and E.

[0043] Example 1

[0044] Based on data collection and research into the physiological and biochemical growth characteristics of *Reedia spp.*, this study employs object-oriented techniques for extracting the spatial distribution of *Reedia spp.*. First, a classification system is established based on the land cover type of the study area. Second, normalized difference vegetation index (NDVI) time-series images are reconstructed, and the optimal segmentation scale is determined for image segmentation. Then, based on the differences in vegetation index characteristics, spectral characteristics, texture characteristics, and spatial characteristics of different land cover features across different time-phase images, a wetland land cover classification rule set is constructed to extract the spatial distribution information of *Reedia spp.*. Finally, high-resolution satellite imagery combined with field surveys is used to conduct accuracy evaluation.

[0045] The specific implementation process of this embodiment is as follows: Figure 1 As shown.

[0046] In this embodiment of the invention, satellite data processing includes basic steps such as radiometric calibration, atmospheric correction, orthorectification, image fusion, geometrical fine correction, image mosaicking, and image cropping, which meet the quantitative remote sensing requirements of this invention.

[0047] This embodiment, combining the actual land cover types in the study area with the classifiability of image data, divides the land cover types in the study area into six categories: reeds (including reeds), water bodies (including lakes, rivers, ditches, waterlogged areas / ponds, etc.), grasslands (including sedges, water chestnuts, etc.), woodlands (including poplars, willows), construction land (including buildings, paved roads), and other land uses (including mudflats, bare land, etc.). The distribution of each type exhibits obvious zonation characteristics. Generally, from low-lying areas (water surface) to high-lying areas (near dikes), the distribution is sequentially: lakes (rivers), mudflats, grasslands, reeds, and poplar / willow woodlands. Buildings, roads, ditches, and reed haystacks are randomly mixed among these types. The characteristics of each land cover are shown in Table 1 below:

[0048] Table 1. Characteristics of Major Land Cover Types

[0049]

[0050] Multi-scale segmentation is a fundamental method and tool for object-oriented classification. The ultimate goal of segmentation is to classify neighboring pixels with common spectral characteristics and attributes into small, unified regions, thereby providing an information foundation for further subdivision. The selection of the optimal segmentation scale is crucial to the accuracy of the classification results; both under-segmentation and over-segmentation will affect classification accuracy. This embodiment utilizes the ESP (Estimation of Scale Parameters) tool developed by Dirk Tiede & Shaun R. Levick, employing a controlled variable segmentation experiment method to determine the optimal segmentation scale.

[0051] In this embodiment, using eCognition Developer software, the shape factor and compactness factor were set to 0.1–0.9 and 0.9–0.1, respectively. After nine segmentation experiments, based on the segmentation effects of various land cover types (i.e., the accuracy of segmentation of each land cover boundary) such as reeds, water bodies, grasslands, forests, construction land, and other land uses, the shape factor and compactness factor were determined to be 0.2 and 0.5, respectively. The initial minimum segmentation scale parameter was set to 20, the segmentation scale step size was set to 10, and the iteration count was set to 50. See [link to documentation]. Figure 2 The final ROC (rate of change in local variance between the scale level of interest and the previous one) plot is shown below. Figure 3 .

[0052] By analyzing the obtained ROC (Reactivity Center) map and considering the actual segmentation effects of various surface cover types such as reeds, water bodies, grasslands, woodlands, built-up land, and other land uses, while also taking into account segmentation efficiency (reducing computer processing time), three different segmentation scales—60, 130, and 220—were selected as the optimal segmentation scales. The satellite imagery used for multi-scale segmentation in this embodiment of the invention is Sentinel-2 imagery data with a 10-meter resolution; the three different segmentation scales—60, 130, and 220—only represent the optimal segmentation scale at that resolution.

[0053] A hierarchical classification system refers to the vertical relationships between objects at different scales after segmentation, and the horizontal relationships between objects within the same level after segmentation at the same scale. Using the characteristic information of land cover types at each level generated after segmentation, a set of classification rules is established, and their hierarchical relationships are analyzed to ensure inheritance. This allows classification information to be transferred between levels, with all land cover types inherited at the final level.

[0054] Based on the three determined optimal segmentation scales, as well as the vertical relationships between land types at different scales and the horizontal relationships between land types within the same scale, three classification levels are established from large to small: Level 1 is the first level, with a segmentation scale of 220, extracting lake water, river water, large areas of reeds, large areas of grassland, etc.; Level 2 is the second level, which, based on Level 1, further segments the unclassified areas at a scale of 130, extracting small areas of reeds, small areas of lake grass, flood discharge channels, roads, woodland, etc.; Level 3, based on Level 2, further segments the unclassified areas at a scale of 60, mainly extracting relatively small patches of land features, such as buildings, waterlogged depressions / ponds.

[0055] Table 2 Classification Hierarchy Table

[0056] Classification hierarchy Segmentation scale Extracting land cover types Level 1 220 Lakes, rivers, vast expanses of reeds, and large grasslands, etc. Level 2 130 Small patches of reeds, small patches of grassland, ditches, roads, woodlands, white mudflats, etc. Level 3 60 Relatively small patches of land features such as buildings, puddles / ponds, etc.

[0057] Note: Large areas refer to contiguous areas with a total area of ​​more than 10,000 square meters (approximately 15 mu); small areas refer to areas with a total area of ​​less than 10,000 square meters (approximately 15 mu).

[0058] In this embodiment, sample points were collected from different land cover types in the study area, including reeds, grasslands (sedges), woodlands (poplars), water bodies, buildings, and roads, to create classification training samples. Using GF-5 AHSI data from the study area on June 14, 2019, satellite hyperspectral curves for different land cover types, such as reeds, water bodies, grasslands, woodlands, built-up land, and other land uses, were generated. (See attached data.) Figure 4 The spectral differences among different land cover types were compared and analyzed. Simultaneously, live samples of vegetation such as *Imperata cylindrica*, *Carex spp.*, and *Leymus chinensis* were collected, and their hyperspectral curves were tested. (See attached data.) Figures 5(a) to 5(d) The results were compared and verified with satellite hyperspectral curves.

[0059] Using the aforementioned optimal segmentation scale, satellite imagery was segmented at multiple scales. Training samples were created based on field-collected sample points. The differences in phenological, spectral, spatial, and textural characteristics of segmented objects from various land cover types, such as reeds, water bodies, grasslands, woodlands, built-up land, and other land uses, were compared. After comprehensively considering the spatiotemporal spectral multidimensional remote sensing characteristics, 12 feature parameters for classification were finally determined, including: Normalized Difference Vegetation Index (NDVI), Normalized Difference Water Index (NDWI), Ratio Green (RG), Red (R), Near Infrared (NIR), Brightness, Shape Index (SI), Length / Width (L / W), Rectangular Fit (RF), Saturation, Intensity, and Homogeneity. The meanings of each feature parameter are described in Table 3.

[0060] Table 3. Meaning of each feature parameter

[0061]

[0062]

[0063] In the above table, The sum of the brightness weights of all layers used in the calculation, where K is the number of layers used in the calculation. This is the sum of the brightness weights of all layers used for calculation. The average intensity of layer k for image object v; b v The border length of the image object v, # v The area of ​​the rectangle inscribed in the image object v; Length(v) is the length of the bounding rectangle of the image object v, and Width(v) is the width of the bounding rectangle of the image object v; P v ρ is the total number of pixels contained in the image object v. v (x,y) is the elliptical distance at pixel (x,y); i is the row number, j is the column number, N is the total number of rows or columns of the image object, and P i,jThis is the standardized value of the pixel in the i-th row and j-th column. An image object refers to the patch obtained after segmentation at the corresponding scale.

[0064] The purpose of this embodiment is to obtain the spatial distribution of *Imperata cylindrica* with high precision, and the key is to accurately distinguish *Imperata cylindrica* from other vegetation. *Imperata cylindrica* begins to emerge in mid-to-late March, reaches its maximum height in August, and gradually withers and is even harvested after November. *Imperata cylindrica* exhibits significant phenological differences from grasslands and woodlands, which can serve as a basis for constructing classification rules. Therefore, satellite images from April, August, and December of the current year were selected as key temporal data for extracting *Imperata cylindrica*. Vegetation indices can accurately reflect the growth, biomass, and cover of green vegetation on a large scale. Based on continuous multi-temporal vegetation indices, seasonal changes in vegetation can be reflected, among which NDVI is the most widely used. NDVI data calculated from the three temporal images of April, August, and December were used as bands of RGB images, combined in chronological order, and reconstructed to obtain NDVI time-series images. Using NDVI time-series images in multi-scale segmentation can improve segmentation accuracy. This invention uses 2019 as an example, selecting image data from April, August, and December as the optimal temporal data for monitoring *Dioscorea nipponica*. April (see...) Figure 6 In August, the southern reeds were in their early growth stage, with low plant density. Furthermore, due to the influence of dead reed stalks from the previous year, the saturation, intensity, greenness, and homogeneity of the southern reeds in the images were lower than grassland, and similar to woodland. Figure 7 ), when the southern reeds reach their maximum height, their biomass is higher than that of the grassland. The grassland is largely submerged due to high water levels, while the southern reeds, being emergent plants, are less affected by water levels and exhibit higher saturation and greenness in the images compared to woodland; December (see...) Figure 8 The southern reeds were either dead or harvested, and the vegetation features shown in the images were weak, with saturation and greenness lower than those of grasslands and woodlands. Figure 9 The image shown is an NDVI time-series image, with comparison and verification sample points. Figure 9 The yellow area marked 'a' is the distribution area of ​​Reed ferns, the pink area marked 'b' is the distribution area of ​​grassland, and the white or light yellow area marked 'c' is the distribution area of ​​woodland.

[0065] Based on the aforementioned classification hierarchy, the differences in phenological, spectral, spatial, and textural characteristics of land cover types such as reeds, water bodies, grasslands, forests, construction land, and other land uses under different levels were analyzed. The characteristic parameters required for extracting each land cover type under different levels were determined, and classification rules were established, as shown in Table 4.

[0066] The specific implementation process of this embodiment is as follows:

[0067] Step 1: Chessboard Segmentation. Using the chessboard segmentation method, the satellite imagery is divided into two parts based on the study area bounding file: the study area and the non-study area. The purpose of Step 1 is to remove the influence of the non-study area on classification and improve multi-scale segmentation efficiency.

[0068] Note: If the image data used for classification has been cropped according to the study area, step 1 is unnecessary, and you can start directly from step 2.

[0069] Step 2: Multi-scale segmentation 1 (Level 1). A segmentation scale of 220 was selected to segment the study area imagery (satellite images from April, August, and December, as well as NDVI time-series images were all used for segmentation). The optimal threshold range for each feature parameter of each land cover was determined. Based on the different optimal threshold ranges for each feature parameter, lakes, rivers, large areas of reeds, and large areas of grassland were extracted sequentially. The remaining areas were classified as unclassified.

[0070] Based on the average value of the training sample objects, the optimal threshold range (i.e., the upper and lower limits of the parameter values) for each feature parameter is determined by updating the feature threshold. The specific steps are as follows: First, select a feature parameter and display its value in a grayscale image, using color to indicate the feature range. Then, continuously adjust the lower and upper limits of the feature parameter value until the remaining coverage area of ​​the feature value in the grayscale image represents the target land type. Finally, select the minimum and maximum values ​​displayed in the feature parameter view window as the optimal threshold range for extracting the current land type. This invention uses the 2019 study area distribution as an example to mainly illustrate the classification rule construction method. Due to differences in vegetation growth and water level in different years and time periods within the study area, the spectral, textural, and spatial characteristics of different land cover types in remote sensing images also differ. Therefore, the threshold ranges of each feature parameter fluctuate, and this embodiment does not list the specific feature parameter threshold ranges in detail.

[0071] Step 3: Multi-scale segmentation 2 (Level 2). A segmentation scale of 130 is selected to further segment the unclassified image from Step 3. The optimal threshold range for each feature parameter of each land cover is determined. Based on the different optimal threshold ranges for each feature parameter, ditches, small patches of reeds, small patches of grassland, woodland, roads, and white mud islands are extracted sequentially. The remaining areas are classified as unclassified.

[0072] Step 4: Multi-scale segmentation Level 3. Using a segmentation scale of 60, further segment the unclassified image from Step 4 to determine the optimal threshold range for each feature parameter of each land cover. Based on the different optimal threshold ranges for each feature parameter, extract waterlogged areas / ponds and buildings sequentially. The remaining areas are classified as unclassified.

[0073] Step 5: Categorization and organization. Large areas of reeds and small areas of reeds are categorized together as reeds; lakes, rivers, ditches, and waterlogged depressions / ponds are categorized together as water bodies; large areas of grassland and small areas of grassland are categorized together as grassland; roads and buildings are categorized together as construction land; and white mud islands and uncategorized areas are categorized together as other land uses.

[0074] Step 6: Merge Objects. Merge the following objects separately: reeds, water bodies, grasslands, woodlands, construction land, and other land uses. The aim is to merge adjacent, separate objects of the same land type into a single, complete object.

[0075] Step 7: Remove Small Patches. Run the small patch removal process separately for Reed, Water Bodies, Grasslands, Woodlands, Construction Land, and Other Land Uses. The aim is to merge patches with areas smaller than a certain value into adjacent land use types. Using the RemoveObjects algorithm, select the target land use type from which small patches need to be removed, and set a conditional function (i.e., Number of pixels ≤ a, where Number of pixels represents the number of pixels in the patch, and a represents the specific value set; for Reed, Water Bodies, Grasslands, and Woodlands, this invention recommends setting it to 25; for Construction Land and Other Land Uses, this invention recommends setting it to 9). After running the RemoveObjects algorithm, patches of the target land use type with image elements less than or equal to 'a' will be automatically merged into adjacent land use types.

[0076] Step 8: Smoothing. Smoothing is applied to reeds, water bodies, grasslands, woodlands, construction land, and other land uses separately to improve the naturalness of transitions between adjacent land uses. Using a pixel-based object resizing algorithm, the target land use type to be smoothed is selected, and smoothing parameters are set, including the smoothing method (in this embodiment, the growth method is selected), and the smoothing size (for reeds, water bodies, grasslands, and woodlands, it is recommended to set it to 5×5; for construction land and other land uses, it is recommended to set it to 3×3). After running the pixel-based object resizing algorithm, the boundaries of the target land use type will be smoother, and the transition with adjacent land uses will be more natural.

[0077] Table 4. Details of Classification Rules

[0078]

[0079] Note: Subscripts 04, 08, and 12 represent the time phases of April, August, and December, respectively. The specific threshold range values ​​for each feature parameter at each classification level are not listed in detail here.

[0080] This embodiment utilizes the accuracy evaluation module of eCognition software, employing a confusion matrix to evaluate classification accuracy. The verification samples were selected based on ground-based field survey results, UAV aerial imagery, and satellite imagery with a spatial resolution better than 1 meter. Verification showed that the overall classification accuracy was better than 90%.

[0081] Table 5 Accuracy Evaluation Results

[0082]

[0083]

[0084] Example 2

[0085] Embodiment 2 of the present invention provides a terminal device corresponding to Embodiment 1 above. The terminal device can be a processing device for a client, such as a mobile phone, a laptop, a tablet computer, a desktop computer, etc., to execute the method of the above embodiments.

[0086] The terminal device in this embodiment includes a memory, a processor, and a computer program stored in the memory; the processor executes the computer program in the memory to implement the steps of the method in Embodiment 1 described above.

[0087] In some implementations, the memory may be high-speed random access memory (RAM), and may also include non-volatile memory, such as at least one disk storage device.

[0088] In other implementations, the processor can be any type of general-purpose processor, such as a central processing unit (CPU) or a digital signal processor (DSP), and there is no limitation here.

[0089] Example 3

[0090] Embodiment 3 of the present invention provides a computer-readable storage medium corresponding to Embodiment 1 above, on which a computer program / instructions are stored. When the computer program / instructions are executed by a processor, they implement the steps of the method of Embodiment 1 above.

[0091] A computer-readable storage medium can be a tangible device that holds and stores instructions for use by an instruction execution device. A computer-readable storage medium can be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any combination thereof.

[0092] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of this application can be implemented in various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0093] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0094] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0095] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0096] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for extracting the spatial distribution of *Reedia spp.*, characterized in that, Includes the following steps: S1. Divide the remote sensing image into study area and non-study area; S2. Select the first segmentation scale, segment the image of the study area using the first segmentation scale, determine the optimal threshold range of each feature parameter of each land cover, and extract lakes, rivers, large areas of reeds, and large areas of grassland in sequence according to the different optimal threshold ranges of each feature parameter. The remaining range is classified as unclassified. S3. Select the second segmentation scale and use the second segmentation scale to further segment the unclassified image, determine the optimal threshold range of each feature parameter of each land cover, and extract ditches, small patches of reeds, small patches of grassland, woodland, roads, and white mud islands in sequence according to the different optimal threshold ranges of each feature parameter. The remaining range is classified as unclassified. S4. Select the third segmentation scale and use the third segmentation scale to further segment the unclassified image obtained in step S3. Determine the optimal threshold range of each feature parameter of each land type. Based on the different optimal threshold ranges of each feature parameter, extract waterlogged areas / ponds and buildings in sequence. The remaining range is classified as unclassified. S5. Large areas of reeds and small areas of reeds are classified as reeds; lakes, rivers, ditches, and waterlogged depressions / ponds are classified as water bodies; large areas of grassland and small areas of grassland are classified as grassland; roads and buildings are classified as construction land; and Baini Island and unclassified areas are classified as other land. S6. For any land use type, merge the image elements smaller than a set value into other land use types, wherein the land use types are: reeds, water bodies, grasslands, forests, construction land, and other land use. S7. Smooth the land types processed in step S6 to obtain the final classification result.

2. The method for extracting the spatial distribution of *Reedia spp.* according to claim 1, characterized in that, The characteristic parameters include: Normalized Difference Vegetation Index (NDVI), Normalized Difference Water Index (NDWI), Greenness Index (RG), Red Band Reflectance (R), Near Infrared Band Reflectance (NIR), Brightness, Shape Index (SI), Aspect Ratio (L / W), Rectangularity (RF), Saturation, Intensity, and Homogeneity.

3. The method for extracting the spatial distribution of *Reedia spp.* according to claim 2, characterized in that, The formulas for calculating the Normalized Difference Vegetation Index (NDVI), Normalized Difference Water Index (NDWI), and Greenness Index (RG) are as follows: NIR is the reflectance in the near-infrared band, and R is the reflectance in the red band. G represents the green band reflectivity; B represents the blue band reflectivity.

4. The method for extracting the spatial distribution of *Reedia spp.* according to claim 1, characterized in that, The first segmentation scale > the second segmentation scale > the third segmentation scale.

5. The method for extracting the spatial distribution of *Reedia spp.* according to claim 1, characterized in that, The first segmentation scale is 220, the second segmentation scale is 130, and the third segmentation scale is 60.

6. The method for extracting the spatial distribution of *Reedia spp.* according to claim 1, characterized in that, The specific implementation process for determining the optimal threshold range for each feature parameter includes: Select a feature parameter, and display the value of that feature parameter in the grayscale image, and use color to indicate the feature range; The lower and upper limits of the feature parameter value are continuously adjusted until the remaining coverage area of ​​the feature parameter value in the grayscale image is the target extracted land type; Select the minimum and maximum values ​​displayed in the feature parameter view window as the optimal threshold range for extracting the current land cover type.

7. The method for extracting the spatial distribution of *Reedia spp.* according to claim 1, characterized in that, The specific implementation process of step S6 includes: For any land type, set a conditional function, i.e., Number of pixels ≤ a, where Number of pixels represents the number of pixels in the patch and a represents the specific value set. The land parcels that satisfy the condition function in the land category are merged into other adjacent land categories.

8. The method for extracting the spatial distribution of *Reedia spp.* according to claim 7, characterized in that, For reeds, water bodies, grasslands, and woodlands, a = 25; for construction land and other land uses, a = 9.

9. A terminal device, comprising a memory, a processor, and a computer program stored in the memory; characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program / instructions stored thereon; characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method according to any one of claims 1 to 8.

Images