Forest information extraction method based on remote sensing image

Abstract

The invention relates to the technical field of forest image processing, in particular to a forest information extraction method based on a remote sensing image. The method comprises the following steps: obtaining remote sensing image data of a forest area and a GPS positioning sample place area, and carrying out sample area remote sensing image processing and vegetation index analysis to obtain vegetation index data corresponding to each forest type in the sample area; performing end member component reflectivity analysis on the remote sensing image data of the sample area to obtain spectral reflectivity values corresponding to all end member components of the forest in the sample area; performing mixed pixel decomposition on a corresponding sample area remote sensing image in the sample area remote sensing image data in combination with a linear mixed pixel decomposition model so as to generate an abundance map of each ground feature type of the sample area; and carrying out pixel forest information classification extraction on each ground feature type abundance map of the sample region to obtain each forest type distribution map of the sample region. According to the invention, different forest information in the remote sensing image can be obtained by integrating the surface reflectance and the vegetation index.

Description

Technical Field

[0001] The present invention relates to the technical field of forest image processing, and in particular to a forest information extraction method based on remote sensing images. Background Art

[0002] Currently, although medium- and high-resolution remote sensing images have relatively high measurement accuracy, the data is relatively expensive and difficult to obtain; low-resolution remote sensing images have a wider coverage area and are relatively inexpensive. Since their maximum spatial resolution is 250m and remote sensing images have high spectral radiation accuracy, they can be obtained free of charge. However, there are a large number of mixed pixels, which often cause certain errors in the classification of land objects. It is difficult to obtain the ideal classification accuracy of remote sensing images using traditional remote sensing imagery to obtain forest information, resulting in poor forest information acquisition results. Summary of the Invention

[0003] Based on this, it is necessary for the present invention to provide a forest information extraction method based on remote sensing images to solve at least one of the above technical problems.

[0004] To achieve the above purpose, a forest information extraction method based on remote sensing images includes the following steps: Step S1: acquiring remote sensing image data of a forest area and a GPS-located sample point area, and performing sample area remote sensing image processing on the remote sensing image data based on the GPS-located sample point area to obtain sample area remote sensing image data; Step S2: performing vegetation index analysis on the remote sensing image data of the sample area to obtain vegetation index data corresponding to each forest type in the sample area; performing end-member component reflectance analysis on the remote sensing image data of the sample area based on the vegetation index data corresponding to each forest type in the sample area to obtain spectral reflectance values corresponding to each end-member component of the forest in the sample area; Step S3: performing mixed pixel decomposition on the remote sensing image data of the sample area corresponding to the sample area remote sensing image based on the spectral reflectance values corresponding to each end member component of the forest in the sample area and in combination with a linear mixed pixel decomposition model to generate an abundance map of each feature type in the sample area; Step S4: performing pixel forest information classification extraction on the abundance maps of various species in the sample area to obtain a distribution map of various forest types in the sample area.

[0005] Furthermore, step S1 includes the following steps: Step S11: Acquire remote sensing image data of the forest area; Step S12: using GPS to obtain coordinate information corresponding to typical field sample points of each forest type in the forest area to obtain GPS positioning sample point areas; Step S13: performing projection conversion and atmospheric correction on the corresponding remote sensing image in the remote sensing image data to obtain remote sensing image correction data; Step S14: performing black bar geometric correction processing on the remote sensing image corresponding to the remote sensing image correction data to obtain forest image black bar elimination correction data; Step S15: obtaining the corresponding GPS forest sampling point area range through GPS positioning sampling point area, and dividing the corresponding forest type image in the forest image black bar elimination correction data into sampling areas based on the GPS forest sampling point area range to obtain the sampling area remote sensing image data.

[0006] Furthermore, the remote sensing image data in step S11 includes forest resource second-category survey images, MODIS images, and TM images.

[0007] Furthermore, step S14 includes the following steps: Step S141: performing image space coordinate conversion on the corresponding forest resource category II image, MODIS image, and TM image in the remote sensing image correction data to generate corresponding remote sensing image data under the same image space coordinates; Step S142: performing forest image stitching on the remote sensing image data corresponding to the same image space coordinates to analyze the corresponding forest ecological characteristics in the remote sensing image, including forest community structure, tree crown shape, and water system and terrain distribution, and calculating the geometric relationship and texture similarity between forest feature areas to determine the stitching positions between the remote sensing images, thereby obtaining remote sensing stitched image data; Step S143: performing black stripe removal processing on the black stripes of pixels at corresponding stitching positions in the remote sensing stitching image data to obtain remote sensing image black stripe-removed data; Step S144: geometrically correcting the remote sensing image black bar removal data to obtain forest image black bar removal correction data.

[0008] Furthermore, step S2 includes the following steps: Step S21: performing remote sensing grayscale analysis on each forest feature in the remote sensing image data of the sample area to obtain the grayscale value corresponding to each feature in the sample area on the remote sensing image; Step S22: constructing a corresponding vegetation time series profile curve based on the grayscale values of the objects in the sample area on the remote sensing image, so as to find out the vegetation period corresponding to each forest type; Step S23: performing vegetation index analysis on each forest feature in the remote sensing image data of the sample area based on the vegetation time series profile curve to obtain vegetation index data corresponding to each forest type in the sample area; Step S24: performing end-member component reflectance analysis on the remote sensing image data of the sample area based on the vegetation index data corresponding to each forest type in the sample area to obtain the spectral reflectance value corresponding to each end-member component of the forest in the sample area.

[0009] Furthermore, the forest features described in step S21 specifically include water areas, coniferous forests, broad-leaved forests, shrub forests, construction land, cultivated land, and bamboo forests.

[0010] Furthermore, step S24 includes the following steps: Step S241: using a preset decision tree classification model to classify the forest types of the remote sensing image data of the sample area to obtain remote sensing sub-images corresponding to each forest type in the sample area; Step S242: performing forest phenology difference analysis on the remote sensing sub-images corresponding to each forest type within the sample area to obtain phenology difference characteristics corresponding to each forest type within the sample area; Step S243: Based on the vegetation index data corresponding to each forest type in the sample area and in combination with the decision tree classification model, a forest endmember purification model is constructed for the phenological difference characteristics corresponding to each forest type in the sample area. The vegetation index grayscale value corresponding to each forest feature in the vegetation index data is used as the threshold of the decision tree classification model, and the phenological difference characteristics corresponding to each forest type in the sample area are input into the decision tree classification model for model construction to generate a forest endmember purification model for the sample area. Step S244: Preliminary purification of forest endmember components is performed on the remote sensing image data of the sample area based on the forest endmember purification model of the sample area to obtain endmember component data of each forest type in the sample area; Step S245: performing endmember component reflectance analysis on the endmember component data of each forest type in the sample area to obtain the spectral reflectance value corresponding to each endmember component of the forest in the sample area.

[0011] Furthermore, step S245 includes the following steps: Performing minimum noise separation transformation on the end-member component data of each forest type in the sample area to obtain the noise separation component data of each forest end-member in the sample area; Perform forest endmember purity index analysis on the noise separation component data of each forest endmember in the sample area to obtain the purity index of each forest endmember component in the sample area; Based on the purity index of each forest end-member component in the sample area and using the N-dimensional scatter plot, the end-member component data of each forest type in the sample area are analyzed in N-dimensional component pixels, so as to divide the area of interest corresponding to the forest type according to the purity index of each forest end-member component, and determine the corresponding initial terminal component pixel in the N-dimensional scatter plot, so as to further purify the initial terminal component pixel by using the decision tree classification model, and eliminate or modify some scatter points of the initial terminal component pixels that do not conform to the rules and are erroneous, so as to obtain the corresponding forest pure terminal component pixels in the sample area; The spectral reflectance of each forest feature in each band is obtained through the corresponding forest pure terminal component pixels in the sample area, and the spectral reflectance of each forest feature in each band is corrected by combining the remote sensing image corresponding to the GPS positioning sample site area to obtain the spectral reflectance value corresponding to each end-member component of the regional forest.

[0012] Furthermore, step S3 includes the following steps: Step S31: constructing a spectral response matrix for the spectral reflectance values corresponding to each end-member component of the forest in the sample area to generate an end-member spectral response matrix for the forest in the sample area; Step S32: performing mixed pixel decomposition processing on the sample area remote sensing image corresponding to the sample area remote sensing image data based on the sample area forest endmember spectral response matrix and in combination with the linear mixed pixel decomposition model, so as to classify the spectral responses corresponding to the endmembers of different forest vegetation types according to the sample area forest endmember spectral response matrix and adopt the spectral difference analysis method, and construct a spectral mixed decomposition model for the spectral responses corresponding to the endmembers of different forest vegetation types by adopting the multivariate linear regression analysis method, so as to establish a mixed pixel decomposition model suitable for the sample area remote sensing image, thereby decomposing the corresponding mixed pixels in the output remote sensing image, and obtaining the mixed pixel decomposition results of various types of forest objects in the sample area; Step S33: Based on the mixed pixel decomposition results of various forest features in the sample area, the abundance spatial distribution analysis of each forest feature type in the remote sensing image of the sample area is performed, so as to generate an abundance distribution of each feature type in the forest of the sample area through spatial interpolation and visualization technology, so as to generate an abundance map of each feature type in the sample area.

[0013] Furthermore, step S4 includes the following steps: Step S41: selecting the abundance of each component unit of each feature in the sample area by using a pixel threshold in each component image corresponding to each feature type abundance map in the sample area; Step S42: comparing and judging the abundance of each feature component unit in the corresponding sample area according to the preset component abundance threshold of 0.5. If the abundance of each feature component unit in the sample area in the feature type abundance map is greater than the preset component abundance threshold of 0.5, then the corresponding feature component pixel is classified as belonging to the sample area component, otherwise it is not. Step S43: The pixel images of the ground object components corresponding to the components of the sample area are combined into one image, and the pixel forest information is extracted from the image using the maximum likelihood method to obtain a distribution map of each forest type in the sample area.

[0014] Beneficial effects of the present invention: The forest information extraction method based on remote sensing images proposed in the present invention has the beneficial effect of providing an accurate geographic information basis for the study area by acquiring remote sensing image data of the forest area and combining it with GPS positioning of the sample location area. On this basis, the image data is divided into sample areas and preprocessed to effectively extract relevant information of the study area. The sample area division ensures the matching of the remote sensing image with the actual geographic location, improving the accuracy of data analysis. The preprocessing process can reduce noise, correct image distortion, etc., further improving data quality. The divided areas provide clear sample areas for subsequent analysis, which is helpful for comparative analysis between different forest types and lays the foundation for subsequent vegetation index analysis and reflectance analysis. Secondly, vegetation index (such as NDVI) is a commonly used parameter in remote sensing image processing. It can effectively characterize vegetation coverage and its growth status. By performing vegetation index analysis on remote sensing image data of the sample area, the vegetation characteristics of different types of forests can be identified, helping researchers to distinguish different forest types and their distribution. Next, through endmember analysis, the corresponding spectral reflectance values are extracted based on the vegetation index data of each forest type. This provides a basis for subsequent spectral feature extraction. Endmember reflectance analysis can accurately reflect the spectral characteristics of different land feature types (such as trees, shrubs, and bare land) within the sample area, improving the accuracy of land feature classification. Then, by using a linear mixed pixel decomposition model, the mixed characteristics of pixels in remote sensing data can be effectively processed. In particular, when a pixel contains spectral signals of multiple land feature types, mixed pixel decomposition can identify and quantify the abundance of different land feature types by analyzing spectral reflectance values. By combining the endmember spectral reflectance values and the linear mixed pixel decomposition model, the land feature components contained in each pixel in the sample area can be accurately decomposed, thereby generating an abundance map of each land feature type. This process can provide reliable data support for forest management and land use classification, helping researchers to more clearly understand the spatial distribution of forests and the proportion of each type, thereby improving the classification accuracy of forest land feature endmembers. Finally, by classifying pixel forest information based on the abundance map of each feature type, different types of forests, non-forest areas and other feature categories can be clearly divided by setting classification standards (such as abundance thresholds, clustering algorithms, etc.). This step can provide detailed spatial information on the forest distribution in the study area, further revealing the spatial distribution pattern of forest types and their changing trends. Through pixel forest information classification, researchers can more accurately evaluate key indicators such as the distribution, density and forest coverage of forest resources. This can effectively utilize the excellent spectral radiation accuracy of the data, improve the lack of spatial resolution, and conveniently and quickly use the surface reflectance and vegetation index provided by MODIS images to effectively improve the accuracy of forest cover mapping and forest type identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Other features, objects and advantages of the present invention will become more apparent upon reading the detailed description of non-limiting embodiments thereof made with reference to the following drawings: Figure 1 This is a schematic diagram of the steps of the forest information extraction method based on remote sensing images of the present invention; Figure 2 for Figure 1 Detailed step flow chart of step S1 in FIG. DETAILED DESCRIPTION

[0016] The following is a clear and complete description of the technical method of the present invention in conjunction with the accompanying drawings. It is obvious that the embodiments described are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts are within the scope of protection of the present invention.

[0017] In addition, the accompanying drawings are merely schematic illustrations of the present invention and are not necessarily drawn to scale. Identical reference numerals in the figures denote identical or similar parts, and thus repetitive descriptions thereof will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities that do not necessarily correspond to physically or logically separate entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor and / or microcontroller approaches.

[0018] It should be understood that although the terms "first," "second," and the like may be used herein to describe various elements, these elements should not be limited by these terms. These terms are used solely to distinguish one element from another. For example, a first element may be referred to as a second element, and similarly, a second element may be referred to as a first element, without departing from the scope of the exemplary embodiments. The term "and / or" as used herein includes any and all combinations of one or more of the listed associated items.

[0019] To achieve this, please refer to Figures 1 to 2 The present invention provides a method for extracting forest information based on remote sensing images, which includes the following steps: Step S1: acquiring remote sensing image data of a forest area and a GPS-located sample point area, and performing sample area remote sensing image processing on the remote sensing image data based on the GPS-located sample point area to obtain sample area remote sensing image data; Step S2: performing vegetation index analysis on the remote sensing image data of the sample area to obtain vegetation index data corresponding to each forest type in the sample area; performing end-member component reflectance analysis on the remote sensing image data of the sample area based on the vegetation index data corresponding to each forest type in the sample area to obtain spectral reflectance values corresponding to each end-member component of the forest in the sample area; Step S3: performing mixed pixel decomposition on the remote sensing image data of the sample area corresponding to the sample area remote sensing image based on the spectral reflectance values corresponding to each end member component of the forest in the sample area and in combination with a linear mixed pixel decomposition model to generate an abundance map of each feature type in the sample area; Step S4: performing pixel forest information classification extraction on the abundance maps of various species in the sample area to obtain a distribution map of various forest types in the sample area.

[0020] In the embodiment of the present invention, please refer to Figure 1 FIG. 1 is a flow chart showing the steps of the method for extracting forest information based on remote sensing images according to the present invention. In this example, the method for extracting forest information based on remote sensing images includes the following steps: Step S1: acquiring remote sensing image data of a forest area and a GPS-located sample point area, and performing sample area remote sensing image processing on the remote sensing image data based on the GPS-located sample point area to obtain sample area remote sensing image data; In an embodiment of the present invention, remote sensing image data of the required forest area is downloaded through an official remote sensing data platform, such as NASA's Earthdata website. At the same time, with the help of high-precision GPS equipment, the coordinates of representative sample points in the target area are measured on the spot to form a GPS-located sample point area. The coordinates of the GPS-located sample point area are imported using geographic information system (GIS) software. The spatial analysis function of the software is used to set a circular area with a specific radius (such as 500 meters) as the sample area with each sample point as the center. These sample areas are superimposed with the remote sensing image data. The cropping tool of the software is used to crop the remote sensing image according to the sample area range, and finally a plurality of small blocks of sample area remote sensing image data are obtained, which provide a basis for subsequent refined analysis.

[0021] Step S2: performing vegetation index analysis on the remote sensing image data of the sample area to obtain vegetation index data corresponding to each forest type in the sample area; performing end-member component reflectance analysis on the remote sensing image data of the sample area based on the vegetation index data corresponding to each forest type in the sample area to obtain spectral reflectance values corresponding to each end-member component of the forest in the sample area; In an embodiment of the present invention, a vegetation index analysis is performed on the remote sensing image data of the sample area by using professional remote sensing image processing software, such as ENVI. Taking the calculation of the normalized vegetation index (NDVI) as an example, the reflectance values of the near-infrared band and the red light band are extracted from the image data. According to the formula (near-infrared band reflectance-red light band reflectance) / (near-infrared band reflectance+red light band reflectance), the NDVI value is calculated pixel by pixel to generate an NDVI image, which represents the vegetation index data corresponding to each forest type in the sample area. Then, based on the vegetation index data, a linear spectral mixture model is used to perform end-member component reflectance analysis. Water areas, coniferous forests, broad-leaved forests, etc. are determined as end-member components in the software. The least squares method is used to solve the linear equation group. The spectral reflectance of the pixel and the abundance of each end-member component are combined to finally calculate the spectral reflectance value corresponding to each end-member component.

[0022] Step S3: performing mixed pixel decomposition on the remote sensing image data of the sample area corresponding to the sample area remote sensing image based on the spectral reflectance values corresponding to each end member component of the forest in the sample area and in combination with a linear mixed pixel decomposition model to generate an abundance map of each feature type in the sample area; In an embodiment of the present invention, the spectral reflectance values corresponding to the end-member components of the forest in the sample area previously obtained are input into the remote sensing image processing software, and combined with the linear mixed pixel decomposition model, the spectral reflectance of each pixel in the remote sensing image of the sample area is expressed as a linear combination of the reflectances of each end-member component. The linear equation group is solved by the least squares method to obtain the abundance value of each end-member component (such as coniferous forest, broad-leaved forest, construction land, etc.) in each pixel. For example, for a certain pixel, the abundance of coniferous forest is calculated to be 0.6, the abundance of broad-leaved forest is 0.2, and the abundance of construction land is 0.2. The abundance values of each end-member component of all pixels are arranged according to their positions in the image, and the image generation function of the software is used to represent the abundance of different land feature types with different colors or grayscales to generate an abundance map of each land feature type in the sample area, which intuitively displays the distribution ratio of each land feature type in the sample area.

[0023] Step S4: performing pixel forest information classification extraction on the abundance maps of various species in the sample area to obtain a distribution map of various forest types in the sample area.

[0024] In an embodiment of the present invention, a supervised classification algorithm, such as a maximum likelihood classification method, is used to process the abundance map of various species in the sample area. In the software, training samples of known forest types in the sample area, such as the abundance characteristics of samples such as coniferous forests, broad-leaved forests, and shrub forests, are collected in advance. These training samples are used to calculate the statistical characteristics of each forest type, such as the mean and covariance matrix. For each pixel in the abundance map of various species in the sample area, the probability that the pixel belongs to each forest type is calculated based on its abundance characteristics. For example, the probability that a pixel belongs to a coniferous forest is 0.8, and the probability that it belongs to a broad-leaved forest is 0.2. According to the principle of maximum probability, the pixel is classified as a coniferous forest. After performing such a classification operation on all pixels, the image rendering function of the software is used to display different forest types in different colors to generate a distribution map of each forest type in the sample area, clearly presenting the distribution status of forest types in the sample area.

[0025] Furthermore, step S1 includes the following steps: Step S11: Acquire remote sensing image data of the forest area; Step S12: using GPS to obtain coordinate information corresponding to typical field sample points of each forest type in the forest area to obtain GPS positioning sample point areas; Step S13: performing projection conversion and atmospheric correction on the corresponding remote sensing image in the remote sensing image data to obtain remote sensing image correction data; Step S14: performing black bar geometric correction processing on the remote sensing image corresponding to the remote sensing image correction data to obtain forest image black bar elimination correction data; Step S15: obtaining the corresponding GPS forest sampling point area range through GPS positioning sampling point area, and dividing the corresponding forest type image in the forest image black bar elimination correction data into sampling areas based on the GPS forest sampling point area range to obtain the sampling area remote sensing image data.

[0026] As an embodiment of the present invention, refer to Figure 2 As shown, Figure 1 Detailed step flow diagram of step S1 in FIG. 1 , in this embodiment, step S1 includes the following steps: Step S11: Acquire remote sensing image data of the forest area; In an embodiment of the present invention, remote sensing image data of forest areas is searched and downloaded through a dedicated remote sensing data acquisition platform, such as the Earth Resources Observation and Science Center (EROS) website of the United States Geological Survey (USGS). During the download process, data sets containing Class II forest resource survey images, MODIS images, and TM images are screened out. Class II forest resource survey images can be obtained from the forestry department's database to provide information such as forest type, area, and volume. The data is medium-resolution imaging spectrometer data acquired through sensors, which has a wide coverage area and can provide macroscopic forest information. TM images, or thematic mapper data, can be obtained from relevant remote sensing satellite data providers. They have high spatial resolution and are conducive to detailed analysis of forest characteristics. These data are integrated together to ultimately form complete remote sensing image data.

[0027] Step S12: using GPS to obtain coordinate information corresponding to typical field sample points of each forest type in the forest area to obtain GPS positioning sample point areas; In an embodiment of the present invention, a high-precision GPS receiver is equipped and professionals are organized to travel to the target area. Representative sample points are selected in the field for different forest types, such as coniferous forests, broad-leaved forests, and mixed forests. The selection of each sample point should fully reflect the characteristics of the forest type, such as tree species composition and stand structure. After arriving at the sample point location, the GPS receiver is turned on, and the positioning is completed, and the longitude and latitude coordinate information of the sample point is recorded. For example, in a coniferous forest area, five typical sample points are selected, and their coordinates are recorded as (longitude 110.5°, latitude 25.3°), (longitude 110.6°, latitude 25.4°), etc. The coordinate information of the typical sample points of all forest types in the field is summarized to form a GPS positioning sample point area, providing an accurate geographic coordinate reference for subsequent data analysis.

[0028] Step S13: performing projection conversion and atmospheric correction on the corresponding remote sensing image in the remote sensing image data to obtain remote sensing image correction data; In an embodiment of the present invention, by using professional remote sensing image processing software, such as ENVI or ERDAS, projection conversion is first performed. According to the geographical location and application requirements of the target area, the target projection coordinate system, such as the Gauss-Krüger projection, is determined. The original projection information of each image in the remote sensing image data is read. The projection conversion tool in the software is used to convert the image from the original coordinate system to the target projection coordinate system according to the projection parameter settings to ensure that different images are in the same coordinate frame. Then, atmospheric correction is performed. The 6S (Second Simulation of the Satellite Signal in the Solar Spectrum) model or other atmospheric correction models is used to input parameters such as the sensor type, imaging time, and geographical location of the image. Combined with the atmospheric model and surface reflectivity information, the influence of the atmosphere on the radiation transmission of the image is calculated, and the radiation brightness value of the image is corrected to eliminate the influence of atmospheric scattering and absorption, thereby obtaining remote sensing image correction data that more accurately reflects the real information of the surface.

[0029] Step S14: performing black bar geometric correction processing on the remote sensing image corresponding to the remote sensing image correction data to obtain forest image black bar elimination correction data; In an embodiment of the present invention, if black bars are present in the remote sensing image correction data due to sensor failure or data transmission problems, a special black bar correction algorithm is used. First, the position and range of the black bars are determined by image recognition technology, such as using an edge detection algorithm to identify the black bar boundary. For the black bar area, geometric correction is performed based on the image features of the normal area around it. For example, assuming that the black bar area is located in the middle of the image and has a width of 10 pixels, 20 pixels of normal areas on each side of the black bar are selected. By calculating the grayscale value distribution and texture features of the pixels in these normal areas, the pixels in the black bar area are repaired using an interpolation algorithm. Bilinear interpolation can be used to calculate the grayscale values of the pixels in the black bar area according to a linear relationship based on the grayscale values of the normal pixels on both sides of the black bar. The black bar area is processed row by row and column by column to finally obtain the forest image black bar elimination correction data, so that the image can restore its normal geometric features and information expression.

[0030] Step S15: obtaining the corresponding GPS forest sampling point area range through GPS positioning sampling point area, and dividing the corresponding forest type image in the forest image black bar elimination correction data into sampling areas based on the GPS forest sampling point area range to obtain the sampling area remote sensing image data.

[0031] In an embodiment of the present invention, based on the coordinate information of the GPS positioning sampling point area previously obtained, the spatial analysis function of the GIS software is used to determine the area within a certain range around each sampling point, that is, the GPS forest sampling point area range. For example, a circular area with a radius of 500 meters is set as the sampling point area range with each sampling point as the center, and these area ranges are superimposed on the forest type image in the forest image black bar elimination correction data. The software's cropping tool is used to crop the forest type image according to the GPS forest sampling point area range. For example, for an image covering a large area of forest, multiple small images are cropped according to multiple GPS forest sampling point area ranges. These small images are the sampling area remote sensing image data, and more detailed forest information extraction and analysis can be performed on these sampling areas later.

[0032] Furthermore, the remote sensing image data in step S11 includes forest resource second-category survey images, MODIS images, and TM images.

[0033] Furthermore, step S14 includes the following steps: Step S141: performing image space coordinate conversion on the corresponding forest resource category II image, MODIS image, and TM image in the remote sensing image correction data to generate corresponding remote sensing image data under the same image space coordinates; In an embodiment of the present invention, the spatial coordinate conversion tool in the geographic information system (GIS) software is used to process the forest resource type II image, the MODIS image, and the TM image. First, a unified target coordinate system, such as the WGS-84 coordinate system, is determined. For the forest resource type II image, the original coordinate system parameters and control point information are obtained, and the affine transformation model is used to convert the coordinates of each pixel on the image from the original coordinate system to the target coordinate system. For the MODIS image and the TM image, the same affine transformation method is used for conversion based on their respective coordinate systems and control points. For example, during the conversion process, the conversion parameters are calculated for multiple control points evenly distributed on the image based on their coordinate values in the original coordinate system and the target coordinate system. Then, the coordinates of all pixels on the image are converted according to these parameters, and finally, the corresponding remote sensing image data under the same image space coordinates is generated.

[0034] Step S142: performing forest image stitching on the remote sensing image data corresponding to the same image space coordinates to analyze the corresponding forest ecological characteristics in the remote sensing image, including forest community structure, tree crown shape, and water system and terrain distribution, and calculating the geometric relationship and texture similarity between forest feature areas to determine the stitching positions between the remote sensing images, thereby obtaining remote sensing stitched image data; In an embodiment of the present invention, remote sensing image data under the same image space coordinates are processed by utilizing an image stitching algorithm. First, the forest ecological characteristics in the remote sensing image are analyzed, the boundaries of the forest community structure are identified by an edge detection algorithm, and the crown shape is analyzed by a morphological processing method. The water system and terrain distribution are determined by analyzing different grayscale value areas in the image. Then, the geometric relationship between the forest feature areas, such as distance and angle, is calculated. At the same time, a texture analysis algorithm is used to calculate texture similarity. For example, for two adjacent images, the forest feature areas are selected, and the Euclidean distance and the correlation coefficient of the texture features between them are calculated. Based on these calculation results, the stitching position between the images is determined so that the stitched image remains continuous and consistent in the forest ecological characteristics, and the images are stitched according to the determined stitching position to finally obtain remote sensing stitching image data.

[0035] Step S143: performing black stripe removal processing on the black stripes of pixels at corresponding stitching positions in the remote sensing stitching image data to obtain remote sensing image black stripe-removed data; In an embodiment of the present invention, an image restoration algorithm is used to process black stripes of pixels at the splicing locations in remote sensing mosaic image data. First, a threshold segmentation algorithm is used to determine the location and range of the black stripes. For example, areas with pixel grayscale values below a certain threshold (such as 10) are determined to be black stripes. Then, a neighborhood pixel interpolation method is used to restore the pixels within the black stripes. For each pixel within the black stripe, non-black pixels within a certain neighborhood (such as a 3×3 neighborhood) are selected and a weighted average calculation is performed based on the grayscale values of these pixels. The calculated result is used as the new grayscale value of the black stripe pixel. This method is then used to process all pixels within the black stripe until the black stripe is eliminated, ultimately obtaining remote sensing image black stripe-removed data.

[0036] Step S144: geometrically correcting the remote sensing image black bar removal data to obtain forest image black bar removal correction data.

[0037] In an embodiment of the present invention, a polynomial correction model is used to geometrically correct remote sensing image data for black bar removal. First, a certain number (e.g., at least six) of ground control points are selected on the image. The locations of these control points on the image and in actual geographic space are known. Then, based on the coordinates of these control points, the coefficients of the polynomial correction model are solved using the least squares method. The polynomial correction model can be a linear, quadratic, or higher-order polynomial. The appropriate degree of the polynomial is selected based on the degree of image deformation. For example, a linear polynomial is selected for images with less deformation, and a quadratic polynomial is selected for images with greater deformation. After obtaining the coefficients of the correction model, a coordinate transformation is performed on each pixel in the image, converting it from its original distorted coordinates to its correct geographic coordinates. Simultaneously, the pixel grayscale values are resampled, using methods such as bilinear interpolation or nearest neighbor interpolation to ensure that the grayscale values of the corrected image are reasonable, ultimately obtaining the corrected data for forest image black bar removal.

[0038] Furthermore, step S2 includes the following steps: Step S21: performing remote sensing grayscale analysis on each forest feature in the remote sensing image data of the sample area to obtain the grayscale value corresponding to each feature in the sample area on the remote sensing image; In an embodiment of the present invention, professional remote sensing image processing tools are used to process the remote sensing image data of the sample area. Different forest features in the image, such as water areas, coniferous forests, broad-leaved forests, shrub forests, construction land, cultivated land and bamboo forests, are separated from the image through an image segmentation algorithm. For each feature, the corresponding area is delineated on the image, and then the grayscale values of all pixels in the area are statistically analyzed. Taking coniferous forests as an example, the area where coniferous forests are distributed is identified in the image, the grayscale value of each pixel in the area is read, and the average grayscale value of all pixels in the area is calculated, which is used as the grayscale value corresponding to the coniferous forest in the remote sensing image. According to the same method, the grayscale values of other types of features such as water areas and broad-leaved forests are obtained in turn.

[0039] Step S22: constructing a corresponding vegetation time series profile curve based on the grayscale values of the objects in the sample area on the remote sensing image, so as to find out the vegetation period corresponding to each forest type; In an embodiment of the present invention, remote sensing image data of a sample area at multiple different time points are collected, and the grayscale values of each type of land feature, such as water area, coniferous forest, broad-leaved forest, etc., on the image at each time point are calculated according to the previous method. With time as the horizontal axis and grayscale value as the vertical axis, the grayscale values of each land feature at different time points are connected to construct a corresponding vegetation time series profile curve. For example, for coniferous forest, its grayscale values on the images of different months such as January, March, and May are marked in the coordinate system in turn and connected into a curve. By comparing the vegetation time series profile curves of different forest types, the fluctuation and characteristics of the curves are observed. If the curves of coniferous forest and broad-leaved forest are significantly different within a certain time period, then this time period is the vegetation period that can clearly distinguish the two forest types.

[0040] Step S23: performing vegetation index analysis on each forest feature in the remote sensing image data of the sample area based on the vegetation time series profile curve to obtain vegetation index data corresponding to each forest type in the sample area; In an embodiment of the present invention, remote sensing image data corresponding to vegetation periods that clearly distinguish various forest types are selected based on a previously obtained vegetation time series profile curve. For each forest feature in the MODIS image, such as water areas, coniferous forests, and broad-leaved forests, a specific vegetation index calculation formula is used for calculation. For example, the commonly used normalized difference vegetation index (NDVI) calculation formula is (near-infrared band reflectance - red light band reflectance) / (near-infrared band reflectance + red light band reflectance). The reflectance values of each forest feature in the near-infrared band and the red light band are extracted from the remote sensing image data and substituted into the formula to calculate the NDVI value. For each forest feature, the vegetation index values at multiple time points within the selected vegetation period are calculated, and these values are organized into a data set to obtain vegetation index data corresponding to each forest type in the sample area.

[0041] Step S24: performing end-member component reflectance analysis on the remote sensing image data of the sample area based on the vegetation index data corresponding to each forest type in the sample area to obtain the spectral reflectance value corresponding to each end-member component of the forest in the sample area.

[0042] In an embodiment of the present invention, by combining vegetation index data corresponding to each forest type in the sample area and remote sensing image data of the sample area, a linear spectral mixture model is used to perform end-member component reflectance analysis. First, based on the distribution of different forest objects in the image, water areas, coniferous forests, broad-leaved forests, etc. are determined as end-member components. For each pixel, its spectral reflectance is expressed as a linear combination of the reflectances of each end-member component. The least squares method is used to solve the linear equation group to obtain the abundance of each end-member component in the pixel. Then, based on the spectral reflectance of the pixel and the abundance of each end-member component, the spectral reflectance value corresponding to each end-member component is calculated. For example, for a mixed pixel containing coniferous forest and construction land, the abundance of coniferous forest and construction land is obtained by solving the linear equation group. Then, combined with the spectral reflectance of the pixel, the spectral reflectance values of the coniferous forest and construction land are calculated. Such calculations are performed on all pixels in the image, and finally the spectral reflectance values corresponding to each end-member component of the forest in the sample area are obtained.

[0043] Furthermore, the forest features described in step S21 specifically include water areas, coniferous forests, broad-leaved forests, shrub forests, construction land, cultivated land, and bamboo forests.

[0044] Furthermore, step S24 includes the following steps: Step S241: using a preset decision tree classification model to classify the forest types of the remote sensing image data of the sample area to obtain remote sensing sub-images corresponding to each forest type in the sample area; In an embodiment of the present invention, the collected remote sensing image data of the sample area is input into a preset decision tree classification model. The model has been pre-trained based on a large amount of known forest type sample data and has classification capabilities. The model will make judgments based on the spectral characteristics, texture characteristics and other information of the image data according to the node rules of the decision tree. For example, if the reflectivity of a certain pixel in the near-infrared band is higher than a certain threshold and the reflectivity in the red light band is lower than another threshold, then the pixel is judged to belong to the coniferous forest type. After judging all the pixels in the image one by one, the image is divided into areas corresponding to different forest types. The image data corresponding to each area is the remote sensing sub-image corresponding to each forest type, such as the coniferous forest remote sensing sub-image, the broad-leaved forest remote sensing sub-image, etc., and finally the remote sensing sub-image corresponding to each forest type in the sample area is obtained.

[0045] Step S242: performing forest phenology difference analysis on the remote sensing sub-images corresponding to each forest type within the sample area to obtain phenology difference characteristics corresponding to each forest type within the sample area; In an embodiment of the present invention, by analyzing the changes in remote sensing image characteristics of different forest types at different times for the remote sensing sub-images corresponding to each forest type previously obtained, remote sensing sub-images of multiple phases are selected with a one-year cycle, and for the images of each phase, a vegetation index, such as the normalized vegetation index (NDVI), is calculated. By comparing the NDVI values of different forest types at the same phase, as well as the NDVI change trends of the same forest type at different phases, forest phenological differences are determined. For example, the NDVI value of coniferous forests decreases less in winter, while the NDVI value of broad-leaved forests decreases significantly in autumn. By comprehensively analyzing the changes in vegetation indices of multiple phases, the phenological difference characteristics corresponding to each forest type in the sample area are finally obtained, such as the length of the phenological period, the peak growth period, etc.

[0046] Step S243: Based on the vegetation index data corresponding to each forest type in the sample area and in combination with the decision tree classification model, a forest endmember purification model is constructed for the phenological difference characteristics corresponding to each forest type in the sample area. The vegetation index grayscale value corresponding to each forest feature in the vegetation index data is used as the threshold of the decision tree classification model, and the phenological difference characteristics corresponding to each forest type in the sample area are input into the decision tree classification model for model construction to generate a forest endmember purification model for the sample area. In an embodiment of the present invention, vegetation index data of various forest types in the sample area at different phases are first collected, and vegetation index grayscale values that can effectively distinguish various forest features are screened out. These grayscale values are used as thresholds of the decision tree classification model. Then, the phenological difference characteristics corresponding to each forest type in the sample area, such as the start time and change rate of the phenological period, are input into the decision tree classification model as input data. The model constructs classification rules based on the input phenological difference characteristics and preset thresholds. For example, if the NDVI rising rate of a certain forest type in spring is higher than a certain threshold, and the NDVI peak in summer reaches another threshold, the forest type is determined to be a specific type. By continuously optimizing the classification rules, a forest end member purification model for the sample area is finally generated.

[0047] Step S244: Preliminary purification of forest endmember components is performed on the remote sensing image data of the sample area based on the forest endmember purification model of the sample area to obtain endmember component data of each forest type in the sample area; In an embodiment of the present invention, remote sensing image data of a sample area is input into a forest endmember purification model for the sample area. The model judges each pixel in the image according to the previously constructed classification rules. Pixels that meet the characteristics of a certain forest type are classified into that forest type. During the classification process, some noise pixels and mixed pixels are removed. For example, pixels at the edge of the forest with fuzzy spectral characteristics are eliminated if they do not meet the clear characteristics of any forest type. After processing the entire image data, relatively pure endmember component data corresponding to each forest type in the sample area are obtained. These data can more accurately represent the characteristics of each forest type.

[0048] Step S245: performing endmember component reflectance analysis on the endmember component data of each forest type in the sample area to obtain the spectral reflectance value corresponding to each endmember component of the forest in the sample area.

[0049] In an embodiment of the present invention, spectral analysis technology is used to process the end-member component data of each forest type in the sample area, and the digital quantization value (DN value) of the image data is converted into a radiation brightness value through radiation calibration. Then, according to the atmospheric correction model, the influence of the atmosphere on the spectral reflectance is eliminated, and the radiation brightness value is converted into the surface reflectance. For the end-member component data of each forest type, its average reflectivity in different bands is calculated. For example, for the coniferous forest end-member component data, its average reflectivity in the visible light band (such as blue light, green light, red light) and the near-infrared band is calculated. By analyzing the end-member component data of all forest types, the spectral reflectance values corresponding to each end-member component of the forest in the sample area are obtained. These values can be used for subsequent forest information extraction and classification.

[0050] Furthermore, step S245 includes the following steps: Performing minimum noise separation transformation on the end-member component data of each forest type in the sample area to obtain the noise separation component data of each forest end-member in the sample area; In an embodiment of the present invention, by using professional remote sensing image processing software, the end-member component data of each forest type collected in the sample area are imported into the software, and the software uses the minimum noise separation (MNF) transformation algorithm to process the data. The algorithm first calculates the covariance matrix of the data, obtains the noise eigenvalue and the signal eigenvalue by eigenvalue decomposition, and determines the transformation matrix according to the ratio of the noise eigenvalue to the signal eigenvalue. The transformation matrix is used to perform a linear transformation on the original forest type end-member component data to separate the noise from the effective signal in the data. For example, for a multi-band image containing end-member component data of forest types such as coniferous forests and broad-leaved forests, after MNF transformation, the noise separation component data corresponding to each forest type is obtained. These data remove most of the noise interference and can better reflect the true characteristics of the forest type.

[0051] Preferably, a forest endmember purity index analysis is performed on the noise separation component data of each forest endmember in the sample area to obtain a purity index of each forest endmember component in the sample area; In an embodiment of the present invention, after completing the minimum noise separation transformation, a special index calculation tool or function is used to perform forest endmember purity index analysis on the noise separation component data of each forest endmember, so as to calculate a principle similar to the normalized difference water index (NDWI). The purity index is obtained by operating on data of different bands. For each forest endmember, a specific band combination is selected to calculate its purity index value. For example, for the noise separation component data of a certain forest endmember, the purity index of its forest endmember component is calculated in the near-infrared band and the short-wave infrared band by the formula (near-infrared band value-short-wave infrared band value) / (near-infrared band value+short-wave infrared band value). Such calculation is performed on the noise separation component data of each forest endmember in the sample area, thereby obtaining the purity index corresponding to each forest endmember.

[0052] Preferably, based on the purity index of each forest end-member component in the sample area and using an N-dimensional scatter plot, an N-dimensional component pixel analysis is performed on the end-member component data of each forest type in the sample area, so as to divide the area of interest corresponding to the forest type according to the purity index of each forest end-member component, and determine the corresponding initial terminal component pixel in the N-dimensional scatter plot, so as to further purify the initial terminal component pixel by using a decision tree classification model, and eliminate or modify some scatter points of the initial terminal component pixels that do not conform to the rules and are erroneous, so as to obtain the corresponding forest pure terminal component pixels in the sample area; In an embodiment of the present invention, by utilizing data analysis software, the purity index of each forest end-member component in the sample area is combined with the corresponding forest type end-member component data to draw an N-dimensional scatter plot (N is the number of bands). In the scatter plot, the areas of interest corresponding to different forest types are divided according to the size and distribution pattern of the purity index. For example, an area with a higher purity index and a relatively concentrated distribution corresponds to a certain forest type. In each area of interest, the initial terminal component pixels are determined. These pixels represent the typical characteristics of the forest type. Then, the initial terminal component pixels are further purified using a decision tree classification model. The decision tree model judges each initial terminal component pixel according to the set classification rules. Scatter points that do not conform to the rules, that is, do not conform to the typical characteristics of the forest type, are eliminated; for scatter points with incorrect markings, they are modified according to the characteristics of the surrounding scatter points and the classification rules, and finally the corresponding forest pure terminal component pixels in the sample area are obtained.

[0053] Preferably, the spectral reflectance corresponding to each forest feature in each band is obtained through the corresponding forest pure terminal component pixels in the sample area, and the spectral reflectance corresponding to each forest feature in each band is corrected by combining the remote sensing image corresponding to the GPS positioning sample location area to obtain the spectral reflectance value corresponding to each end-member component of the regional forest.

[0054] In an embodiment of the present invention, a remote sensing image analysis tool is used to extract the digital quantization value (DN value) of each pixel in each band for the pure terminal component pixels of the forest obtained in the sample area, and the DN value is converted into the corresponding spectral reflectance according to the calibration parameters of the remote sensing image. For example, the DN value of a pure terminal component pixel of a forest in the third band is 100, and its spectral reflectance is calculated to be 0.3 through the calibration formula. At the same time, GPS positioning is used to obtain the precise geographical location information of the sample location area, and the high-resolution remote sensing image corresponding to the area is found. The spectral characteristics of the high-resolution image and the same area in the original remote sensing image are compared, and the spectral reflectance calculated previously is corrected. If the high-resolution image shows that the spectral characteristics of the forest objects in a certain area in a certain band deviate from the calculation results of the MODIS image, the spectral reflectance of the MODIS image is adjusted according to the characteristics of the high-resolution image, and finally the accurate spectral reflectance values corresponding to each end-member component of the regional forest are obtained.

[0055] Furthermore, step S3 includes the following steps: Step S31: constructing a spectral response matrix for the spectral reflectance values corresponding to each end-member component of the forest in the sample area to generate an end-member spectral response matrix for the forest in the sample area; In an embodiment of the present invention, by targeting the forest in the sample area, first, each end-member component, such as coniferous forest, broad-leaved forest, grassland, etc., is clarified, and field measurements of different end-member components are performed in the sample area through field spectral measurement equipment to obtain their spectral reflectance values in multiple bands. For example, the spectral reflectance of coniferous forest, broad-leaved forest and grassland is measured in 10 bands of visible light and near-infrared, and the spectral reflectance values of each end-member component in each band are arranged in a certain order to form a matrix. Assuming that three end-member components are measured and each component has reflectance values in 10 bands, the spectral response matrix is a matrix with 3 rows and 10 columns, and each row of the matrix represents the spectral reflectance of an end-member component in each band. The end-member spectral response matrix of the forest in the sample area is generated in this way.

[0056] Step S32: performing mixed pixel decomposition processing on the sample area remote sensing image corresponding to the sample area remote sensing image data based on the sample area forest endmember spectral response matrix and in combination with the linear mixed pixel decomposition model, so as to classify the spectral responses corresponding to the endmembers of different forest vegetation types according to the sample area forest endmember spectral response matrix and adopt the spectral difference analysis method, and construct a spectral mixed decomposition model for the spectral responses corresponding to the endmembers of different forest vegetation types by adopting the multivariate linear regression analysis method, so as to establish a mixed pixel decomposition model suitable for the sample area remote sensing image, thereby decomposing the corresponding mixed pixels in the output remote sensing image, and obtaining the mixed pixel decomposition results of various types of forest objects in the sample area; In an embodiment of the present invention, the remote sensing image of the sample area is processed by utilizing the generated sample area forest end-member spectral response matrix in combination with a linear mixed pixel decomposition model. First, the spectral difference analysis method is used to compare the spectral reflectance differences of the end-members of different forest vegetation types in the sample area forest end-member spectral response matrix in each band, so as to classify the spectral responses corresponding to the end-members of different forest vegetation types. For example, through analysis, it is found that the reflectance of coniferous forests in the near-infrared band is significantly higher than that in the visible light band, while broad-leaved forests have unique reflection characteristics in certain visible light bands. Then, a multivariate linear regression analysis method is used, with the spectral responses corresponding to the end-members of different forest vegetation types as independent variables and the spectral values of the pixels of the sample area remote sensing image as dependent variables, to construct a spectral mixed decomposition model. By solving the coefficients of the model, a mixed pixel decomposition model suitable for the sample area remote sensing image is established. Each pixel in the sample area remote sensing image is substituted into the model to decompose the proportion of each forest feature type in the pixel, and finally the mixed pixel decomposition result of various types of forest features in the sample area is obtained.

[0057] Step S33: Based on the mixed pixel decomposition results of various forest features in the sample area, the abundance spatial distribution analysis of each forest feature type in the remote sensing image of the sample area is performed, so as to generate an abundance distribution of each feature type in the forest of the sample area through spatial interpolation and visualization technology, so as to generate an abundance map of each feature type in the sample area.

[0058] In an embodiment of the present invention, based on the decomposition results of mixed pixels of various types of forest features in the sample area, the abundance spatial distribution of each forest feature type in the remote sensing image of the sample area is analyzed. For each forest feature type, its abundance value in each pixel and the spatial position information of the pixel are obtained. A spatial interpolation technique, such as the Kriging interpolation method, is used to estimate the abundance value of unknown positions in the entire sample area based on the abundance value and spatial position of the known pixels. For example, the coniferous forest abundance value of some pixels is known, and the coniferous forest abundance of other unmeasured positions is predicted by the Kriging interpolation method. Then, using visualization technology, the interpolated abundance values of each feature type are presented in the form of an image, and different colors or grayscales are assigned according to the size of the abundance value, to generate an abundance map of each feature type in the sample area that can reflect the abundance distribution of each feature type in the forest of the sample area, and intuitively display the spatial distribution of each feature type.

[0059] Furthermore, step S4 includes the following steps: Step S41: selecting the abundance of each component unit of each feature in the sample area by using a pixel threshold in each component image corresponding to each feature type abundance map in the sample area; In an embodiment of the present invention, the abundance map of the feature types in the sample area is divided into multiple groups, each group corresponding to a group of component images. For each component image, a pixel threshold is set. The threshold is a fixed value determined based on the previous research and experience on the feature characteristics of the area, for example, 0.2. Each pixel in the component image is traversed. If the value of the pixel is greater than the pixel threshold, the value corresponding to the pixel is used as the abundance of the feature component unit in the sample area. For example, in a certain component image, the value of pixel A (such as coniferous forest) is 0.3, which is greater than the set pixel threshold of 0.2. Then the value of pixel A (such as coniferous forest) 0.3 is determined as the abundance of the corresponding feature component unit. In this way, all component images are processed to finally obtain the abundance of the feature component unit in the sample area.

[0060] Step S42: comparing and judging the abundance of each feature component unit in the corresponding sample area according to the preset component abundance threshold of 0.5. If the abundance of each feature component unit in the sample area in the feature type abundance map is greater than the preset component abundance threshold of 0.5, then the corresponding feature component pixel is classified as belonging to the sample area component, otherwise it is not. In an embodiment of the present invention, the previously obtained abundance of each feature component unit in the sample area is compared one by one with the preset component abundance threshold of 0.5. For each feature component pixel in the feature type abundance map of the sample area, the corresponding feature component unit abundance of the sample area is obtained. If the feature component unit abundance of a certain pixel in the sample area is 0.6, which is greater than 0.5, the pixel is classified into the category of the component in the sample area; if the feature component unit abundance of the pixel in the sample area is 0.4, which is less than 0.5, the pixel does not belong to the component in the sample area. By performing such comparison and division on all feature component pixels, the pixels in the feature type abundance map of the sample area are classified, for example, into forest type, land use type, etc.

[0061] Step S43: The pixel images of the ground object components corresponding to the components of the sample area are combined into one image, and the pixel forest information is extracted from the image using the maximum likelihood method to obtain a distribution map of each forest type in the sample area.

[0062] In an embodiment of the present invention, by integrating the pixel images of the ground object components that were previously classified as belonging to the sample area components, these pixels are recombined into a new image according to their positional relationship in the original image. When using the maximum likelihood method to classify and extract pixel forest information, first, based on the sample data of different forest types in the area collected in the early stage, the statistical characteristics of each forest type, such as the mean, covariance matrix, etc., are calculated. Then, for each pixel in the new image, the probability of the pixel belonging to each forest type is calculated based on its spectral characteristics. The maximum likelihood method will select the forest type with the highest probability as the classification result of the pixel. For example, the probability that a pixel belongs to a coniferous forest is 0.8, and the probability of belonging to a broad-leaved forest is 0.2, then the pixel is classified as a coniferous forest. By performing such classification processing on all pixels in the new image, a distribution map of each forest type in the sample area is finally obtained.

[0063] The present invention is therefore intended to be illustrative and non-restrictive in all respects, with the scope of the invention being defined by the appended claims rather than the foregoing description, and all changes that come within the meaning and range of equivalents of the application documents are intended to be embraced therein.

[0064] The foregoing description is intended only to provide specific embodiments of the present invention, which will enable those skilled in the art to understand and implement the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not intended to be limited to the embodiments shown herein, but is to be construed in the widest possible manner consistent with the principles and novel features disclosed herein.

Claims

1. A forest information extraction method based on remote sensing images, characterized in that: The following steps are involved: Step S1: acquiring remote sensing image data of a forest area and a GPS-located sample point area, and performing sample area remote sensing image processing on the remote sensing image data based on the GPS-located sample point area to obtain sample area remote sensing image data; Step S2: performing vegetation index analysis on the remote sensing image data of the sample area to obtain vegetation index data corresponding to each forest type in the sample area; Based on the vegetation index data corresponding to each forest type in the sample area, the end-member component reflectance analysis of the remote sensing image data of the sample area is performed to obtain the spectral reflectance value corresponding to each end-member component of the forest in the sample area; Step S3: performing mixed pixel decomposition on the remote sensing image data of the sample area corresponding to the sample area remote sensing image based on the spectral reflectance values corresponding to each end member component of the forest in the sample area and in combination with a linear mixed pixel decomposition model to generate an abundance map of each feature type in the sample area; Step S4: performing pixel forest information classification extraction on the abundance maps of various species in the sample area to obtain a distribution map of various forest types in the sample area.

2. The forest information extraction method based on remote sensing images according to claim 1 is characterized in that: Step S1 includes the following steps: Step S11: Acquire remote sensing image data of the forest area; Step S12: using GPS to obtain coordinate information corresponding to typical field sample points of each forest type in the forest area to obtain GPS positioning sample point areas; Step S13: performing projection conversion and atmospheric correction on the corresponding remote sensing image in the remote sensing image data to obtain remote sensing image correction data; Step S14: performing black bar geometric correction processing on the remote sensing image corresponding to the remote sensing image correction data to obtain forest image black bar elimination correction data; Step S15: obtaining the corresponding GPS forest sampling point area range through GPS positioning sampling point area, and dividing the corresponding forest type image in the forest image black bar elimination correction data into sampling areas based on the GPS forest sampling point area range to obtain the sampling area remote sensing image data.

3. The forest information extraction method based on remote sensing images according to claim 2 is characterized in that: The remote sensing image data in step S11 include forest resource second-category survey images, MODIS images, and TM images.

4. The forest information extraction method based on remote sensing images according to claim 2 is characterized in that: Step S14 includes the following steps: Step S141: performing image space coordinate conversion on the corresponding forest resource category II image, MODIS image, and TM image in the remote sensing image correction data to generate corresponding remote sensing image data under the same image space coordinates; Step S142: performing forest image stitching on the remote sensing image data corresponding to the same image space coordinates to analyze the corresponding forest ecological characteristics in the remote sensing image, including forest community structure, tree crown shape, and water system and terrain distribution, and calculating the geometric relationship and texture similarity between forest feature areas to determine the stitching positions between the remote sensing images, thereby obtaining remote sensing stitched image data; Step S143: performing black stripe removal processing on the black stripes of pixels at corresponding stitching positions in the remote sensing stitching image data to obtain remote sensing image black stripe-removed data; Step S144: geometrically correcting the remote sensing image black bar removal data to obtain forest image black bar removal correction data.

5. The method for extracting forest information based on remote sensing images according to claim 1, characterized in that: Step S2 includes the following steps: Step S21: performing remote sensing grayscale analysis on each forest feature in the remote sensing image data of the sample area to obtain the grayscale value corresponding to each feature in the sample area on the remote sensing image; Step S22: constructing a corresponding vegetation time series profile curve based on the grayscale values of the objects in the sample area on the remote sensing image, so as to find out the vegetation period corresponding to each forest type; Step S23: performing vegetation index analysis on each forest feature in the remote sensing image data of the sample area based on the vegetation time series profile curve to obtain vegetation index data corresponding to each forest type in the sample area; Step S24: performing end-member component reflectance analysis on the remote sensing image data of the sample area based on the vegetation index data corresponding to each forest type in the sample area to obtain the spectral reflectance value corresponding to each end-member component of the forest in the sample area.

6. The method for extracting forest information based on remote sensing images according to claim 5, characterized in that: The forest features described in step S21 specifically include water areas, coniferous forests, broad-leaved forests, shrub forests, construction land, cultivated land, and bamboo forests.

7. The method for extracting forest information based on remote sensing images according to claim 5, characterized in that: Step S24 includes the following steps: Step S241: using a preset decision tree classification model to classify the forest types of the remote sensing image data of the sample area to obtain remote sensing sub-images corresponding to each forest type in the sample area; Step S242: performing forest phenology difference analysis on the remote sensing sub-images corresponding to each forest type within the sample area to obtain phenology difference characteristics corresponding to each forest type within the sample area; Step S243: Based on the vegetation index data corresponding to each forest type in the sample area and in combination with the decision tree classification model, a forest endmember purification model is constructed for the phenological difference characteristics corresponding to each forest type in the sample area. The vegetation index grayscale value corresponding to each forest feature in the vegetation index data is used as the threshold of the decision tree classification model, and the phenological difference characteristics corresponding to each forest type in the sample area are input into the decision tree classification model for model construction to generate a forest endmember purification model for the sample area. Step S244: Preliminary purification of forest endmember components is performed on the remote sensing image data of the sample area based on the forest endmember purification model of the sample area to obtain endmember component data of each forest type in the sample area; Step S245: performing endmember component reflectance analysis on the endmember component data of each forest type in the sample area to obtain the spectral reflectance value corresponding to each endmember component of the forest in the sample area.

8. The method for extracting forest information based on remote sensing images according to claim 7, characterized in that: Step S245 includes the following steps: Performing minimum noise separation transformation on the end-member component data of each forest type in the sample area to obtain the noise separation component data of each forest end-member in the sample area; Perform forest endmember purity index analysis on the noise separation component data of each forest endmember in the sample area to obtain the purity index of each forest endmember component in the sample area; Based on the purity index of each forest end-member component in the sample area and using the N-dimensional scatter plot, the end-member component data of each forest type in the sample area are analyzed in N-dimensional component pixels, so as to divide the area of interest corresponding to the forest type according to the purity index of each forest end-member component, and determine the corresponding initial terminal component pixel in the N-dimensional scatter plot, so as to further purify the initial terminal component pixel by using the decision tree classification model, and eliminate or modify some scatter points of the initial terminal component pixels that do not conform to the rules and are erroneous, so as to obtain the corresponding forest pure terminal component pixels in the sample area; The spectral reflectance of each forest feature in each band is obtained through the corresponding forest pure terminal component pixels in the sample area, and the spectral reflectance of each forest feature in each band is corrected by combining the remote sensing image corresponding to the GPS positioning sample site area to obtain the spectral reflectance value corresponding to each end-member component of the regional forest.

9. The method for extracting forest information based on remote sensing images according to claim 1, characterized in that: Step S3 includes the following steps: Step S31: constructing a spectral response matrix for the spectral reflectance values corresponding to each end-member component of the forest in the sample area to generate an end-member spectral response matrix for the forest in the sample area; Step S32: performing mixed pixel decomposition processing on the sample area remote sensing image corresponding to the sample area remote sensing image data based on the sample area forest endmember spectral response matrix and in combination with the linear mixed pixel decomposition model, so as to classify the spectral responses corresponding to the endmembers of different forest vegetation types according to the sample area forest endmember spectral response matrix and adopt the spectral difference analysis method, and construct a spectral mixed decomposition model for the spectral responses corresponding to the endmembers of different forest vegetation types by adopting the multivariate linear regression analysis method, so as to establish a mixed pixel decomposition model suitable for the sample area remote sensing image, thereby decomposing the corresponding mixed pixels in the output remote sensing image, and obtaining the mixed pixel decomposition results of various types of forest objects in the sample area; Step S33: Based on the mixed pixel decomposition results of various forest features in the sample area, the abundance spatial distribution analysis of each forest feature type in the remote sensing image of the sample area is performed, so as to generate an abundance distribution of each feature type in the forest of the sample area through spatial interpolation and visualization technology, so as to generate an abundance map of each feature type in the sample area.

10. The method for extracting forest information based on remote sensing images according to claim 1, characterized in that: Step S4 includes the following steps: Step S41: selecting the abundance of each component unit of each feature in the sample area by using a pixel threshold in each component image corresponding to each feature type abundance map in the sample area; Step S42: comparing and judging the abundance of each feature component unit in the corresponding sample area according to the preset component abundance threshold of 0.

5. If the abundance of each feature component unit in the sample area in the feature type abundance map is greater than the preset component abundance threshold of 0.5, then the corresponding feature component pixel is classified as belonging to the sample area component, otherwise it is not. Step S43: The pixel images of the ground object components corresponding to the components of the sample area are combined into one image, and the pixel forest information is extracted from the image using the maximum likelihood method to obtain a distribution map of each forest type in the sample area.

Images