Sparse forest crown coverage rate monitoring method and device, electronic equipment and storage medium

Abstract

The invention provides a sparse forest crown coverage rate monitoring method and device, electronic equipment and a storage medium, and relates to the technical field of forestry remote sensing monitoring, and the method comprises the steps: extracting target features of a to-be-monitored sparse forest from a growing season remote sensing image, a non-growing season remote sensing image and SRTM data of the to-be-monitored sparse forest; the target features comprise spectral information, vegetation indexes and salinity indexes of the growing season and the non-growing season, difference values of the features in the growing season and the non-growing season, and texture features and topographic features of the growing season; inputting the target features of the to-be-monitored sparse forest into the sparse forest crown coverage rate estimation model, and outputting a crown coverage rate spatial distribution diagram of the to-be-monitored sparse forest; the sparse forest crown coverage rate estimation model is constructed based on the target features of the sparse forest sample plot and the crown coverage rate sample of the sparse forest sample plot. According to the method, the multi-temporal information of the sparse forest is increased, so that the coverage rate of the canopy of the sparse forest is accurately monitored.

Description

Technical Field

[0001] This invention relates to the field of forestry remote sensing monitoring technology, and in particular to a method, device, electronic equipment and storage medium for monitoring the canopy coverage of sparse forests. Background Technology

[0002] Canopy cover is a key indicator in forest resource surveys. With the development of remote sensing technology, remote sensing imagery has been widely used for canopy cover estimation. Existing studies mostly use the following four methods for estimation: hybrid pixel decomposition, physical model method, empirical model method, and machine learning method.

[0003] Existing methods mainly rely on remote sensing data from the growing season to estimate canopy cover, which does not make sufficient use of multi-temporal features. As a result, current canopy cover estimation schemes are significantly inadequate in terms of accurate monitoring of sparse forests. Summary of the Invention

[0004] This invention provides a method, device, electronic equipment, and storage medium for monitoring the canopy coverage of sparse forests, in order to address the shortcomings of insufficient accuracy in monitoring the canopy coverage of sparse forests in the prior art, and to improve the accuracy of monitoring the canopy coverage of sparse forests.

[0005] This invention provides a method for monitoring the canopy cover of sparse forests, comprising the following steps: The target features of the sparse forest to be monitored are extracted from remote sensing images of the growing season, remote sensing images of the non-growing season, and SRTM data. The target features of the sparse forest to be monitored are input into the sparse forest canopy coverage estimation model, and the spatial distribution map of the canopy coverage of the sparse forest to be monitored is output. The sparse forest canopy coverage estimation model is constructed based on the target features of the sparse forest plots and the canopy coverage sample data of the sparse forest plots.

[0006] In some embodiments, the method further includes: Based on the growing season remote sensing images, non-growing season remote sensing images, and SRTM data of the sparse forest plots, spectral information, vegetation index, salinity index, texture features, and topographic features are extracted to construct a feature set. Regression analysis was performed on the canopy coverage sample data of the sparse forest plots and the feature set; The target features are selected from the feature set based on the feature contribution evaluation.

[0007] In some embodiments, the method further includes: The point cloud data of the sparse forest plots are processed to obtain DSM data and DEM data. The DSM data and DEM data are subtracted to obtain CHM data. Pixels corresponding to CHM data with grass height greater than the average grass height are classified as woody vegetation, and pixels corresponding to CHM data with grass height less than or equal to the average grass height are used as background points to obtain a binary image; the average grass height is obtained by field sampling of the sparse forest plots; The binary image is divided into multiple samples, with each pixel forming a grid; each grid is aligned with the pixel edges of the remote sensing image. The proportion of pixels classified as woody vegetation within each grid is used as the canopy coverage rate for each sample.

[0008] In some embodiments, the method further includes: The multiple samples are screened using one or more of the following methods: The canopy coverage of the sparse forest plots was divided into zones, and an equal number of samples were drawn from each zone. The samples were screened using the visible light images of the sparse forest plots as a reference. Calculate the mean and standard deviation of the coefficients of variation for multiple bands in each grid, and discard samples whose coefficients of variation are greater than the mean plus one standard deviation.

[0009] In some embodiments, the method further includes: The accuracy of the sparse forest canopy coverage estimation model was evaluated using the coefficient of determination, root mean square error, and mean absolute error.

[0010] In some embodiments, the spectral information includes information on eight bands—coastal blue, blue, green, second green, yellow, red, red edge, and near-infrared—during the growing season and non-growing season, as well as the difference between the same band information during the growing season and non-growing season. The vegetation indices include normalized vegetation index (NDI) for both growing season and non-growing season, normalized green light NDI, normalized difference water vegetation index, soil-regulating vegetation index, improved soil-regulating vegetation index, atmospheric resistance vegetation index, differential vegetation index, difference vegetation index, enhanced vegetation index, total vegetation index, ratio vegetation index, improved triangular vegetation index, MERIS terrestrial chlorophyll index, plant senescence reflectance index, red-edged chlorophyll index, wide dynamic range vegetation index, visible light atmospheric resistance index, improved photochemical reflectance index, and the difference between the same vegetation index for both growing season and non-growing season. The salinity index includes salinity indices 1 to 13 for the growing season and non-growing season, as well as the difference between the same salinity index for the growing season and non-growing season. The texture features include the second moment of the first principal component and the second principal component, contrast, correlation, variance, inverse moment, mean sum, variance sum, entropy sum, entropy sum, and difference entropy; the first principal component and the second principal component are obtained by principal component analysis of the growing season remote sensing image. The terrain features include slope, aspect, and elevation.

[0011] In some embodiments, each of the grids is 3×3 pixels in size.

[0012] The present invention also provides a sparse forest canopy coverage monitoring device, comprising the following modules: The extraction module is used to extract the target features of the sparse forest to be monitored from remote sensing images of the growing season, remote sensing images of the non-growing season, and SRTM data. The estimation module is used to input the target features of the sparse forest to be monitored into the sparse forest canopy coverage estimation model and output the spatial distribution map of the canopy coverage of the sparse forest to be monitored; the sparse forest canopy coverage estimation model is constructed based on the target features of the sparse forest plots and the canopy coverage sample data of the sparse forest plots.

[0013] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the sparse forest canopy coverage monitoring device method as described above.

[0014] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the sparse forest canopy coverage monitoring device method as described above.

[0015] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the sparse forest canopy coverage monitoring device method as described above.

[0016] The present invention provides a method, device, electronic equipment, and storage medium for monitoring the canopy cover of sparse forests. It extracts target features from remote sensing images of the sparse forest during the growing season, the non-growing season, and SRTM data. These target features include spectral information, vegetation indices, salinity indices, and the differences between these features in the growing and non-growing seasons, as well as textural and topographic features in the growing season. The target features are input into a sparse forest canopy cover estimation model, which outputs a spatial distribution map of the canopy cover of the sparse forest. This invention improves the accuracy of canopy cover monitoring by adding multi-temporal information about the sparse forest. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 This is one of the flowcharts of the sparse forest canopy coverage monitoring method provided by the present invention.

[0019] Figure 2 This is a schematic diagram of the binary image provided by the present invention.

[0020] Figure 3 This is a statistical accuracy map of canopy coverage rate for drones provided by the present invention.

[0021] Figure 4 This is a schematic diagram of the sampling method provided by the present invention.

[0022] Figure 5 This is a schematic diagram of the visible light image of the UAV provided by the present invention.

[0023] Figure 6 This is a verification diagram of the accuracy of the canopy coverage estimation model provided by this invention.

[0024] Figure 7 This is the second flowchart of the sparse forest canopy coverage monitoring method provided by the present invention.

[0025] Figure 8 This is a map showing the distribution of forest canopy coverage provided by the present invention.

[0026] Figure 9 This is a schematic diagram of the structure of the sparse forest canopy coverage monitoring device provided by the present invention.

[0027] Figure 10 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0028] Canopy cover is a key indicator in forest resource surveys, and is mainly measured through two methods: traditional ground measurement and remote sensing estimation. Traditional methods for measuring canopy cover primarily rely on field measurements, including manual methods such as subjective sampling and visual estimation, as well as data collection using instruments such as Cajanus tubes and spherical densitometers.

[0029] These methods each have their own characteristics. Photo analysis and visual estimation have the advantages of lower cost and faster implementation, but they are easily affected by subjective factors and environmental conditions (such as weather changes), and often have large estimation biases under complex forest stand conditions. In contrast, the Cajanus tube and line intersection sampling method is widely recognized for its high accuracy and unbiasedness, but it is complex to operate, time-consuming, and requires a large investment of human resources, thus making it difficult to meet the need for rapid acquisition of large-area canopy cover data.

[0030] Overall, traditional methods are of great value in ensuring data reliability, but their complexity and time-consuming nature limits their efficiency in large-scale applications.

[0031] With the development of remote sensing technology, remote sensing imagery has been widely used for canopy cover estimation. Existing studies mostly employ four methods: mixed pixel decomposition, physical modeling, empirical modeling, and machine learning. Mixed pixel decomposition assumes that the spectrum of each pixel is composed of a finite number of endmember spectra in a linearly mixed manner. These endmember spectra represent the spectral characteristics of typical "pure" land cover types in the area (e.g., soil, woody vegetation, and rock). By comparing the measured spectrum of a single pixel with the mixed spectrum of the endmembers, the coverage proportion of each land cover type in the pixel can be estimated. However, due to the significant temporal and spatial variations in the spectral characteristics of vegetation and soil, determining endmembers in sparse forest areas remains a major challenge.

[0032] Empirical models estimate canopy cover by establishing a statistical relationship between canopy cover and remote sensing spectral information. However, empirical models are generally only applicable to specific regions and vegetation types and are difficult to use for large-scale canopy cover estimation.

[0033] Physical models simulate canopy spectral characteristics by establishing a physical relationship between vegetation canopy spectral reflectance and a set of parameters (including leaf and canopy properties, soil properties, and observational geometric parameters). However, due to the high complexity of these models and the need to define a large number of parameters, direct inversion of such models is often difficult when the required input data is limited. Furthermore, to simplify the model, physical models require certain assumptions. For example, the SAIL model assumes a horizontally uniform canopy distribution, but at higher spatial resolutions (such as Landsat and Sentinel-2 imagery), most sparse forests exhibit significant spatial variations, making it difficult to satisfy this assumption. Generalization of inversion parameters is easier to achieve with coarse-resolution data, which is one reason why current physical models primarily rely on low-resolution data for canopy cover estimation.

[0034] Machine learning, which typically uses multi-source remote sensing data as input, has been widely applied. However, existing research has mainly focused on the assessment of forests with relatively high canopy closure, while research on the precise monitoring of sparse forests with low canopy closure is severely lacking.

[0035] Current canopy cover estimation schemes have significant shortcomings in the accurate monitoring of sparse forests, specifically: (1) Insufficient reliability of sample acquisition: Existing technologies mostly use fixed threshold segmentation methods based on UAV lidar point cloud data to distinguish woody vegetation from the background, which fails to fully consider the actual distribution and growth characteristics of small woody vegetation in the region, resulting in reduced sample reliability.

[0036] (2) Limitations of sample statistical methods: Existing methods usually use a grid with the same pixel size as the remote sensing image to count the canopy coverage of the sample, ignoring the problem of vegetation uniformity in the neighborhood, which can easily introduce statistical bias and reduce sample quality.

[0037] (3) Insufficient utilization of multi-temporal features: Existing methods mainly rely on remote sensing data during the growing season to estimate canopy coverage, without verifying the effectiveness of remote sensing data during the non-growing season, and ignoring the potential contribution of sparse forest area features in non-growing season images to canopy coverage estimation.

[0038] To address the aforementioned shortcomings, this invention provides a method for estimating canopy cover in sparse forests, thereby overcoming the deficiencies of existing methods in estimating canopy cover in sparse forest areas.

[0039] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0040] Figure 1 This is one of the flowcharts illustrating the sparse forest canopy coverage monitoring method provided by the present invention, such as... Figure 1 As shown, the method includes the following: Step 101: Extract the target features of the sparse forest to be monitored from the growing season remote sensing images, non-growing season remote sensing images, and SRTM data.

[0041] Step 102: Input the target features of the sparse forest to be monitored into the sparse forest canopy coverage estimation model and output the spatial distribution map of the canopy coverage of the sparse forest to be monitored; the sparse forest canopy coverage estimation model is constructed based on the target features of the sparse forest plots and the canopy coverage sample data of the sparse forest plots.

[0042] Specifically, in order to improve the estimation of canopy coverage in sparse forests, in addition to considering data from the growing season, non-growing season data and the differences between the growing and non-growing seasons for each feature are also considered, thereby increasing the multi-temporal information of sparse forests.

[0043] Based on the target features of sparse forest plots and sample data of canopy cover in these plots, a machine learning model was constructed to estimate canopy cover in sparse forests. The machine learning model is not limited to gradient boosting tree models; regression models such as random forest, LightGBM, and XGBoost were also used. Target features included spectral information, vegetation indices, and salinity indices during both the growing and non-growing seasons, as well as the differences between these features in the growing and non-growing seasons, and textural and topographic features during the growing season.

[0044] Acquire remote sensing images of the sparse forest to be monitored during the growing season, the non-growing season, and data from the Shuttle Radar Topography Mission (SRTM). Extract target features of the sparse forest to be monitored from the growing season, non-growing season, and SRTM data. Input the target features of the sparse forest to be monitored into the sparse forest canopy cover estimation model, and output a spatial distribution map of the canopy cover of the sparse forest to be monitored.

[0045] The present invention provides a method for monitoring the canopy cover of sparse forests. This method extracts target features from remote sensing images of the sparse forest during the growing season, the non-growing season, and SRTM data. These target features include spectral information, vegetation indices, salinity indices, and the differences between these features in the growing and non-growing seasons, as well as textural and topographic features in the growing season. The target features are input into a canopy cover estimation model, which outputs a spatial distribution map of the canopy cover of the sparse forest. This invention improves the accuracy of canopy cover monitoring by adding multi-temporal information about the sparse forest.

[0046] The following section will introduce what needs to be done before constructing a sparse forest canopy coverage estimation model, namely, obtaining canopy coverage data of sparse forest sample plots based on drones.

[0047] In some embodiments, the average grass height of the sparse forest plots is obtained by field sampling. The point cloud data of the sparse forest plots were processed to obtain DSM data and DEM data. The DSM data and DEM data were subtracted to obtain CHM data. Pixels corresponding to CHM data with grass height greater than the average grass height were classified as woody vegetation, while pixels corresponding to CHM data with grass height less than or equal to the average grass height were used as background points to obtain a binary image; the average grass height was obtained through field sampling of sparse forest plots. The binary image is divided into multiple samples, with each grid consisting of multiple pixels; each grid is aligned with the pixel edges of the remote sensing image. The proportion of pixels classified as woody vegetation within each grid is used as the canopy coverage rate for each sample.

[0048] The method involves setting up sparse forest sample plots in the drone's flight area, and using a drone equipped with a LiDAR (Light Detection and Ranging) sensor to jointly collect point cloud data and visible light data of the sparse forest sample plots.

[0049] For example, the drone model is DJI M300 RTK, equipped with Zenmuse L1 LiDAR, which performs joint acquisition of point cloud and visible light data. The drone is equipped with a real-time kinematics (RTK) system. The drone flight parameter settings are shown in Table 1.

[0050] Table 1. Flight parameters for UAV aerial photography

[0051] Drone data is processed using software (such as DJI Terra). A 2D reconstruction module processes visible light imagery, automatically generating and mosaicking orthophotos through aerial triangulation and feature point matching, with a resampling resolution of 0.02 meters. A 3D reconstruction module processes point cloud data, including noise removal, point cloud merging, and classification. A Digital Surface Model (DSM) is generated from the classified ground points, and a Digital Elevation Model (DEM) is generated from the highest elevation echo. The DSM and DEM data are then resampled to 0.05 meters. Subtracting the DSM and DEM data yields a Canopy Height Model (CHM) with a spatial resolution of 0.05 m.

[0052] During the sample production process, different data processing software and algorithms can be used to process drone data. For example, in addition to DJI Terra, other data processing platforms such as Pix4D and Agisoft Metashape can be used, or self-developed data processing algorithms can be used to process point cloud data and generate canopy height models, which can also meet the sample production requirements and achieve the same technical effect.

[0053] Field sampling was conducted on sparse forest plots to obtain the average grass height. Using the average grass height as a threshold, the CHM data was masked to generate binary images. Specifically, pixels corresponding to CHM data with grass heights greater than the average were classified as woody vegetation, while pixels corresponding to CHM data with grass heights less than or equal to the average were designated as background points, thus obtaining the binary image. Figure 2 This is a schematic diagram of the binary image provided by the present invention, as shown below. Figure 2As shown, the white areas represent pixels with a height greater than the average grass height, and the black areas represent pixels with a height less than or equal to the average grass height. Figure 2 The square in the diagram represents a grid, and the dot in the center of the square represents the center point of the grid.

[0054] It should be noted that existing threshold segmentation methods based on UAV lidar point cloud data often select a fixed threshold to distinguish woody vegetation from the background. This invention uses the average grass height measured in the field as the threshold, taking into account small woody vegetation with a height of less than 2m, thereby improving the reliability of the canopy coverage sample.

[0055] Using multiple pixels (e.g., 3×3 pixels) of remote sensing imagery (such as PlanetScope imagery, which includes both growing season and non-growing season images) as a grid, with each grid precisely aligned to the pixel edges of the remote sensing image, the binary image is divided into multiple samples. One grid corresponds to one sample, and the center point of a grid serves as the location information of a sample. The proportion of pixels classified as woody vegetation within each grid is calculated, and this proportion is used as the canopy cover rate for each sample.

[0056] The canopy cover of the sample is expressed as: In the formula, The canopy cover of the sample. This represents the number of pixels within the grid whose CHM (grass height) is greater than the average grass height. This represents the total number of pixels within the grid.

[0057] It should be noted that existing methods often use grids with the same pixel size as remote sensing images to statistically measure the canopy coverage of samples, which makes it difficult to guarantee the uniformity of vegetation distribution in the neighborhood. This invention uses a 3×3 pixel grid as the statistical range of sample canopy coverage, and the grid boundary completely coincides with the pixels of the remote sensing image, thereby improving the reliability of the sample.

[0058] The sample locations were obtained from the field based on the field measurement records. The drone lidar data was cropped according to the sample size, and the canopy coverage value of each sample was calculated. The statistical value was then fitted with the canopy coverage measured in the field.

[0059] Figure 3 This is a statistical accuracy map of canopy coverage provided by the UAV according to the present invention. Figure 3 It can be seen that the correlation coefficient r between the statistical value and the field measured canopy coverage is 0.851, and the coefficient of determination R² is 0.713. This shows that the canopy coverage sample data obtained by UAV is highly correlated with the field measurement data and can be used as a substitute for the field sample data.

[0060] The obtained samples are the initial samples, which need to be screened. Several screening methods are introduced below.

[0061] In some embodiments, the plurality of samples are screened using one or more of the following methods: Method 1: Stratified sampling. The canopy cover of the sparse forest plots is divided into zones, and an equal number of samples are drawn from each zone.

[0062] Specifically, to address the issue of uneven sample size distribution across different intervals, with an excessive number of low values ​​and a insufficient number of high values, a stratified sampling strategy was adopted to balance the sample size. Specifically, the canopy cover of the sparse forest plots was divided into zones, and an equal number of samples were drawn from each zone.

[0063] For example, Figure 4 This is a schematic diagram of the sampling method provided by the present invention, such as... Figure 4 As shown, the study area was divided into four intervals based on the canopy coverage value: (0, 0.05], (0.05, 0.1], (0.1, 0.15], and (0.15, 1]. An equal number of samples were drawn from each interval to construct a training set to improve the model's ability to estimate medium and high canopy coverage.

[0064] It should be noted that existing methods do not perform quality control in terms of sample distribution. This invention uses existing canopy coverage products as a stratified sampling mask to obtain sample points in stages, thus ensuring the uniform distribution of sample canopy coverage values.

[0065] Method 2: Manual removal method. The samples are screened using visible light images of sparse forest plots as a reference.

[0066] Specifically, Figure 5 This is a schematic diagram of the visible light image of the UAV provided by the present invention. The visible light image is also known as the background image. Figure 5 The distribution of vegetation in the sample can be determined. To ensure uniform vegetation distribution within the quadrat and reduce the influence of mixed information, high-resolution visible light images of sparse forest plots acquired by UAVs were used as a reference to remove samples located at the forest edge and near roads and buildings.

[0067] Method 3: Coefficient of variation method. Calculate the average of the coefficients of variation for the 8 bands of each grid, and remove samples with coefficients of variation greater than the average plus one standard deviation.

[0068] Specifically, the coefficient of variation is a statistical indicator used to measure the relative dispersion of a dataset; it is defined as the ratio of the standard deviation to the mean. The formula for calculating the coefficient of variation is: in, The coefficient of variation (CV) is represented by σ, where σ represents the standard deviation of the sample, and μ represents the mean of the sample. A higher CV indicates greater relative dispersion of the data; conversely, a lower CV indicates relatively concentrated data. The CV can be used to measure the evenness of woody vegetation distribution.

[0069] Based on remote sensing imagery (e.g., PlanetScope imagery, which includes both growing season and non-growing season remote sensing imagery), the mean and standard deviation of the coefficient of variation for multiple bands (e.g., coastal blue, blue, green, second green, yellow, red, red edge, near-infrared) for each grid are calculated, and samples with a coefficient of variation greater than the mean plus one standard deviation are removed.

[0070] After obtaining the canopy coverage data of the sparse forest plots, the following section details the construction process of the sparse forest canopy coverage estimation model.

[0071] In some embodiments, the sparse forest canopy coverage monitoring method provided by the present invention further includes: Based on remote sensing images of sparse forest plots during the growing season, non-growing season, and SRTM data, spectral information, vegetation index, salinity index, texture features, and topographic features were extracted to construct a feature set. Regression analysis was performed on the canopy coverage sample data and feature set of sparse forest plots; Based on the evaluation of feature contribution, target features are selected from the feature set.

[0072] Specifically, spectral information, vegetation indices, salinity indices, and texture features of the growing season are extracted from remote sensing images of the sparse forest plots during the growing season. Non-growing season spectral information, vegetation indices, and salinity indices are extracted from remote sensing images of the sparse forest plots during the non-growing season. Topographic features are extracted from the SRTM data of the sparse forest plots. A feature set is constructed based on the extracted spectral information, vegetation indices, salinity indices, and texture features of the growing season, as well as the spectral information, vegetation indices, salinity indices, and topographic features of the non-growing season.

[0073] Gradient Boosting Regression Tree (GBRT) is an ensemble learning method that uses a series of CART (Classification and Regression Tree) regression trees to form a base weak learner. This model adds the negative gradient of the loss function to the residual of the previous learner, gradually reducing the residual of the loss function in each training iteration, thus causing the output to gradually converge to a local or global optimum. GBRT regression models can flexibly handle various types of data and are highly robust to outliers, and have been widely applied to various regression tasks.

[0074] A gradient boosting regression tree model was used to perform regression analysis on the canopy cover sample data and feature set of sparse forest plots. Target features were selected from the feature set based on feature contribution evaluation.

[0075] In some embodiments, the spectral information includes information on eight bands—coastal blue, blue, green, second green, yellow, red, red edge, and near-infrared—during the growing season and non-growing season, as well as the difference between the same band information during the growing season and non-growing season.

[0076] Specifically, the remote sensing imagery used is PlanetScope's Level 3B imagery, which has already undergone geometric, radiometric, and orthorectified corrections, requiring no further preprocessing. PlanetScope imagery was acquired during both the growing season (e.g., August) and the non-growing season (e.g., April). PlanetScope sensor and band information are shown in Table 2.

[0077] Table 2 PlanetScope sensor and band information

[0078] Extract all band information from PlanetScope images during both growing and non-growing seasons, namely, coastal blue, blue, green, second green, yellow, red, red edge, and near-infrared bands, and calculate the difference between the same band information during the growing and non-growing seasons.

[0079] Vegetation indices include normalized vegetation index (NDI) for both growing and non-growing seasons, normalized green light NDI, normalized difference water vegetation index, soil-regulating vegetation index, improved soil-regulating vegetation index, atmospheric resistance vegetation index, differential vegetation index, difference vegetation index, enhanced vegetation index, total vegetation index, ratio vegetation index, improved triangular vegetation index, MERIS terrestrial chlorophyll index, plant senescence reflectance index, red-edged chlorophyll index, wide dynamic vegetation index, visible light atmospheric resistance index, improved photochemical reflectance index, and the difference between the same vegetation index for both growing and non-growing seasons.

[0080] Specifically, 18 vegetation indices were calculated using PlanetScope imagery from both the growing season and the non-growing season, as well as the differences between the same vegetation indices from the growing season and the non-growing season. The 18 vegetation indices are: Normalized Difference Vegetation Index (NDVI), Green Light Normalized Difference Vegetation Index (NDVE), Normalized Difference Aquatic Vegetation Index (NDA), Soil-Regulating Vegetation Index (SDE), Improved Soil-Regulating Vegetation Index (EMV) (Atmospheric Resistance Vegetation Index), Differential Vegetation Index (DVI), Difference Vegetation Index (DPI), Enhanced Vegetation Index (EVI), Total Vegetation Index (SVI), Ratio Vegetation Index (RVI), Improved Triangular Vegetation Index (EMI), MERIS Terrestrial Chlorophyll Index (MERIS), Plant Senescence Reflectance Index (SIR), Red-Edged Chlorophyll Index (REDI), Wide Dynamic Range Vegetation Index (WDL), Visible Light Atmospheric Resistance Index (VALI), Improved Photochemical Reflectance Index (EMI), and the differences between the same vegetation indices from the growing season and the non-growing season.

[0081] The salinity index includes salinity indices 1 to 13 for both the growing and non-growing seasons, as well as the difference between the same salinity index for both the growing and non-growing seasons.

[0082] Specifically, 13 salinity indices were calculated using visible and near-infrared bands of PlanetScope imagery from both growing and non-growing seasons, as well as the differences between the same salinity indices in the growing and non-growing seasons. The 13 salinity indices range from salinity index 1 to salinity index 13.

[0083] Texture features include the second angular moments, contrast, correlation, variance, inverse moments, mean sum, variance sum, entropy sum, entropy sum, and difference entropy of the first and second principal component components during the growing season; the first and second principal component components are obtained by principal component analysis of the remote sensing images during the growing season.

[0084] Specifically, principal component analysis was performed on PlanetScope images from the growing season to obtain the first and second principal component components. Ten gray-level co-occurrence texture features were calculated for the first and second principal component components respectively. The ten gray-level co-occurrence texture features are: second moment of angle, contrast, correlation, variance, inverse moment, sum of mean, sum of variance, sum of entropy, and entropy sum of difference entropy.

[0085] Topographic features include slope, aspect, and elevation.

[0086] Therefore, 140 features can be extracted from growing season remote sensing images, non-growing season remote sensing images, and SRTM data.

[0087] It should be noted that in practical applications, multi-source satellite data is not limited to the currently provided data categories. For example, satellite optical data is not limited to PlanetScope data. Any optical satellite data, such as WorldView, can be used as a substitute for PlanetScope data to achieve the same technical effect. The same applies to terrain data.

[0088] It should also be noted that existing methods have not proven the effectiveness of non-growing season data in estimating canopy cover in sparse forests. This invention incorporates non-growing season remote sensing images and uses reflectance, spectral index, salinity index, texture features, topographic features, and the differences between these features in the growing and non-growing seasons as feature variables, thereby increasing multi-temporal information about sparse forests and improving the estimation results.

[0089] A total of 8875 samples of sparse forest canopy cover were obtained in the study area. Based on the obtained sample data, gradient boosting tree regression analysis was performed using 140 features, including remote sensing spectra from both growing and non-growing seasons (after data preprocessing), vegetation indices, salinity indices, texture features, and topographic features. Through feature contribution evaluation, the top 21 contributing features were selected as target features and input into the sparse forest canopy cover estimation model.

[0090] The target features are: elevation, second green band in the non-growing season, average value of the first principal component, average value of the second principal component, blue band in the non-growing season, difference between yellow bands in the growing and non-growing seasons, salinity index 12 in the non-growing season, contrast of the first principal component, yellow band in the non-growing season, second green band in the growing season, green band in the growing season, normalized difference water index in the growing season, difference between second green bands in the growing and non-growing seasons, inverse difference moment of the second principal component, salinity index 6 in the non-growing season, difference entropy of the first principal component, blue band in the growing season, green band in the non-growing season, inverse difference moment of the first principal component, difference between coastal blue bands in the growing and non-growing seasons, and improved photochemical reflectance index in the growing season.

[0091] In some embodiments, the sparse forest canopy coverage monitoring method provided by the present invention further includes: The accuracy of the sparse forest canopy coverage estimation model was evaluated using the coefficient of determination, root mean square error, and mean absolute error.

[0092] Specifically, the coefficient of determination R², root mean square error (RMSE), and mean absolute error (MAE) are selected as accuracy indicators for the sparse forest canopy cover estimation model. The formulas for R², RMSE, and MAE are shown below: In the formula, The canopy cover value predicted by the sparse forest canopy cover estimation model. This represents the average value of the canopy cover predicted by the sparse forest canopy cover estimation model. denoted as the measured canopy coverage value of the sample, and n is the total number of validation samples.

[0093] Meanwhile, to avoid the random results caused by a single data segmentation, the sparse forest canopy coverage estimation model was subjected to 10-fold cross-validation and the average coefficient of determination R² was calculated to verify the stability of the model.

[0094] Of the 8875 samples randomly selected from UAV data, 70% were used as the training set. After optimization of the gradient boosting tree model parameters, the following parameters were chosen: ntree (number of decision trees) = 670, shrinkage (learning rate) = 0.01, and sampling rate (random sampling rate) = 0.55. The remaining 30% of the samples were used as the validation set, and the accuracy was evaluated against the UAV canopy coverage statistics. Figure 6 This is an accuracy verification diagram of the canopy cover estimation model provided by the present invention, such as... Figure 6 As shown, the correlation coefficient r = 0.81, the coefficient of determination R² = 0.66, the root mean square error RMSE = 0.054, and the mean absolute error MAE = 0.04.

[0095] Figure 7 This is the second flowchart of the sparse forest canopy coverage monitoring method provided by the present invention, for reference. Figure 7 As shown, the present invention provides a method for monitoring the canopy coverage of sparse forests, comprising two parts: obtaining samples of canopy coverage of sparse forests based on UAV data and estimating the canopy coverage of sparse forests.

[0096] The content of obtaining samples of sparse forest canopy coverage based on drone data includes: UAVs equipped with LiDAR acquire UAV data (i.e., point cloud data and visible light data). Data preprocessing is performed on the UAV data (point cloud denoising, point cloud classification, stitching; visible light image orthorectification and mosaicking, etc.) to obtain LiDAR point clouds and visible light images. The LiDAR point clouds are then processed to obtain digital surface models and digital elevation models. Subtracting the digital surface model from the digital elevation model yields the canopy height model. Field sampling was conducted on sparse forest plots to obtain the average grass height. Thresholding was performed using the average grass height to obtain binary images. The binary images were divided into multiple samples, with multiple pixels forming a grid. The canopy cover statistical value of each sample was obtained based on the proportion of pixels classified as woody vegetation within each grid.

[0097] The initial samples were screened using one or more of the following methods: stratified sampling, coefficient of variation method, and manual removal method, to obtain canopy coverage samples.

[0098] The estimation of canopy cover in sparse forests includes the following: Acquire PlanetScope images, including growing season images and non-growing season images. Perform data preprocessing (e.g., image mosaicking and cropping) on ​​the growing season and non-growing season images. Extract features from the preprocessed PlanetScope images, including spectral indices, index indices, salinity indices, and texture features.

[0099] Acquire SRTM data, resample the SRTM data, and obtain terrain features, including slope, aspect, and elevation.

[0100] A gradient boosting tree model was trained using canopy cover samples, along with spectral indices, indicator indices, salinity indices, texture features, and topographic features. Based on feature contribution evaluation, the spectral indices, indicator indices, salinity indices, texture features, and topographic features were optimized to obtain target features. A sparse forest canopy cover estimation model was then constructed using these target features to estimate canopy cover. Finally, the accuracy of the estimated canopy cover was verified using canopy cover samples.

[0101] The canopy cover of a sparse forest in a certain area was calculated using a trained sparse forest canopy cover estimation model. The resulting canopy cover distribution is shown below. Figure 8 As shown.

[0102] This invention provides a method for obtaining sparse forest canopy cover samples based on UAV data. This method can efficiently acquire large amounts of sample data and fills the gaps in current technology for extracting low, sparse trees and shrubs. By combining field measurement data with high-precision UAV point clouds, this invention can effectively distinguish sparse forest from the background.

[0103] This invention provides an algorithm for fine-grained monitoring of canopy cover in sparse forests. The algorithm calculates canopy cover using UAV data, resulting in a large sample of canopy cover in sparse forests. This sample is then combined with high-resolution satellite optical data from both growing and non-growing seasons, as well as topographic data. A machine learning algorithm is used for model extrapolation to ultimately obtain a canopy cover distribution map of the study area.

[0104] The following describes the sparse forest canopy coverage monitoring device provided by the present invention. The sparse forest canopy coverage monitoring device described below can be referred to in correspondence with the sparse forest canopy coverage monitoring method described above.

[0105] Figure 9 This is a schematic diagram of the sparse forest canopy coverage monitoring device provided by the present invention, as shown below. Figure 9 As shown, the present invention provides a sparse forest canopy coverage monitoring device, comprising: Extraction module 901 is used to extract the target features of the sparse forest to be monitored from remote sensing images of the growing season, remote sensing images of the non-growing season, and SRTM data of the sparse forest to be monitored. The estimation module 902 is used to input the target features of the sparse forest to be monitored into the sparse forest canopy coverage estimation model and output the spatial distribution map of the canopy coverage of the sparse forest to be monitored; the sparse forest canopy coverage estimation model is constructed based on the target features of the sparse forest plots and the canopy coverage sample data of the sparse forest plots.

[0106] In some embodiments, the apparatus further includes a first filtering module, which is specifically used for: Based on the growing season remote sensing images, non-growing season remote sensing images, and SRTM data of the sparse forest plots, spectral information, vegetation index, salinity index, texture features, and topographic features are extracted to construct a feature set. Regression analysis was performed on the canopy coverage sample data of the sparse forest plots and the feature set; The target features are selected from the feature set based on the feature contribution evaluation.

[0107] In some embodiments, the apparatus further includes an acquisition module, the acquisition module being configured to: The point cloud data of the sparse forest plots are processed to obtain DSM data and DEM data. The DSM data and DEM data are subtracted to obtain CHM data. Pixels corresponding to CHM data with grass height greater than the average grass height are classified as woody vegetation, and pixels corresponding to CHM data with grass height less than or equal to the average grass height are used as background points to obtain a binary image; the average grass height is obtained by field sampling of the sparse forest plots; The binary image is divided into multiple samples, with each pixel forming a grid; each grid is aligned with the pixel edges of the remote sensing image. The proportion of pixels classified as woody vegetation within each grid is used as the canopy coverage rate for each sample.

[0108] In some embodiments, the apparatus further includes a second filtering module, the second filtering module being specifically used for: The multiple samples are screened using one or more of the following methods: The canopy coverage of the sparse forest plots was divided into zones, and an equal number of samples were drawn from each zone. The samples were screened using the visible light images of the sparse forest plots as a reference. Calculate the mean and standard deviation of the coefficients of variation for multiple bands in each grid, and discard samples whose coefficients of variation are greater than the mean plus one standard deviation.

[0109] In some embodiments, the apparatus further includes an evaluation module, wherein the evaluation is specifically used for: The accuracy of the sparse forest canopy coverage estimation model was evaluated using the coefficient of determination, root mean square error, and mean absolute error.

[0110] In some embodiments, the spectral information includes information on eight bands—coastal blue, blue, green, second green, yellow, red, red edge, and near-infrared—during the growing season and non-growing season, as well as the difference between the same band information during the growing season and non-growing season. The vegetation indices include normalized vegetation index (NDI) for both growing season and non-growing season, normalized green light NDI, normalized difference water vegetation index, soil-regulating vegetation index, improved soil-regulating vegetation index, atmospheric resistance vegetation index, differential vegetation index, difference vegetation index, enhanced vegetation index, total vegetation index, ratio vegetation index, improved triangular vegetation index, MERIS terrestrial chlorophyll index, plant senescence reflectance index, red-edged chlorophyll index, wide dynamic range vegetation index, visible light atmospheric resistance index, improved photochemical reflectance index, and the difference between the same vegetation index for both growing season and non-growing season. The salinity index includes salinity indices 1 to 13 for the growing season and non-growing season, as well as the difference between the same salinity index for the growing season and non-growing season. The texture features include the second moment of the first principal component and the second principal component, contrast, correlation, variance, inverse moment, mean sum, variance sum, entropy sum, entropy sum, and difference entropy; the first principal component and the second principal component are obtained by principal component analysis of the growing season remote sensing image. The terrain features include slope, aspect, and elevation.

[0111] In some embodiments, each of the grids is 3×3 pixels in size.

[0112] It should be noted that the sparse forest canopy coverage monitoring device provided by the present invention can realize all the method steps implemented in the above method embodiments and can achieve the same technical effect. Here, the parts that are the same as those in the method embodiments and the beneficial effects will not be described in detail.

[0113] Figure 10 This is a schematic diagram of the structure of the electronic device provided by the present invention, such as... Figure 10 As shown, the electronic device may include a processor 1010, a communications interface 1020, a memory 1030, and a communication bus 1040, wherein the processor 1010, the communications interface 1020, and the memory 1030 communicate with each other via the communication bus 1040. The processor 1010 can call logical instructions in the memory 1030 to execute a sparse forest canopy cover monitoring method, which includes: extracting target features of the sparse forest to be monitored from remote sensing images of the growing season, remote sensing images of the non-growing season, and SRTM data; inputting the target features of the sparse forest to be monitored into a sparse forest canopy cover estimation model, and outputting a spatial distribution map of the canopy cover of the sparse forest to be monitored; the sparse forest canopy cover estimation model is constructed based on the target features of the sparse forest plots and the canopy cover sample data of the sparse forest plots.

[0114] Furthermore, the logical instructions in the aforementioned memory 1030 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0115] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the sparse forest canopy cover monitoring method provided by the above methods. The method includes: extracting target features of the sparse forest to be monitored from growing season remote sensing images, non-growing season remote sensing images, and SRTM data of the sparse forest to be monitored; inputting the target features of the sparse forest to be monitored into a sparse forest canopy cover estimation model, and outputting a spatial distribution map of the canopy cover of the sparse forest to be monitored; the sparse forest canopy cover estimation model is constructed based on the target features of the sparse forest plots and the canopy cover sample data of the sparse forest plots.

[0116] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the sparse forest canopy cover monitoring method provided by the above methods. The method includes: extracting target features of the sparse forest to be monitored from growing season remote sensing images, non-growing season remote sensing images, and SRTM data of the sparse forest to be monitored; inputting the target features of the sparse forest to be monitored into a sparse forest canopy cover estimation model, and outputting a spatial distribution map of the canopy cover of the sparse forest to be monitored; the sparse forest canopy cover estimation model is constructed based on the target features of the sparse forest plots and the canopy cover sample data of the sparse forest plots.

[0117] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0118] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0119] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for monitoring the canopy cover of sparse forests, characterized in that, include: The target features of the sparse forest to be monitored are extracted from remote sensing images of the growing season, remote sensing images of the non-growing season, and SRTM data. The target features of the sparse forest to be monitored are input into the sparse forest canopy coverage estimation model, and the spatial distribution map of the canopy coverage of the sparse forest to be monitored is output. The sparse forest canopy coverage estimation model is constructed based on the target features of the sparse forest plots and the canopy coverage sample data of the sparse forest plots.

2. The method for monitoring the canopy coverage of sparse forests according to claim 1, characterized in that, The method further includes: Based on the growing season remote sensing images, non-growing season remote sensing images, and SRTM data of the sparse forest plots, spectral information, vegetation index, salinity index, texture features, and topographic features are extracted to construct a feature set. Regression analysis was performed on the canopy coverage sample data of the sparse forest plots and the feature set; The target features are selected from the feature set based on the feature contribution evaluation.

3. The method for monitoring the canopy coverage of sparse forests according to claim 2, characterized in that, The method further includes: The point cloud data of the sparse forest plots are processed to obtain DSM data and DEM data. The DSM data and DEM data are subtracted to obtain CHM data. Pixels corresponding to CHM data with grass height greater than the average grass height are classified as woody vegetation, and pixels corresponding to CHM data with grass height less than or equal to the average grass height are used as background points to obtain a binary image; the average grass height is obtained by field sampling of the sparse forest plots; The binary image is divided into multiple samples, with each pixel forming a grid; each grid is aligned with the pixel edges of the remote sensing image. The proportion of pixels classified as woody vegetation within each grid is used as the canopy coverage rate for each sample.

4. The method for monitoring the canopy coverage of sparse forests according to claim 3, characterized in that, The method further includes: The multiple samples are screened using one or more of the following methods: The canopy coverage of the sparse forest plots was divided into zones, and an equal number of samples were drawn from each zone. The samples were screened using the visible light images of the sparse forest plots as a reference. Calculate the mean and standard deviation of the coefficients of variation for multiple bands in each grid, and discard samples whose coefficients of variation are greater than the mean plus one standard deviation.

5. The method for monitoring the canopy coverage of sparse forests according to claim 1, characterized in that, The method further includes: The accuracy of the sparse forest canopy coverage estimation model was evaluated using the coefficient of determination, root mean square error, and mean absolute error.

6. The method for monitoring the canopy coverage of sparse forests according to claim 2, characterized in that, The spectral information includes information on eight bands: coastal blue, blue, green, second green, yellow, red, red edge, and near-infrared, both during the growing season and outside the growing season, as well as the difference between the same band information during the growing season and outside the growing season. The vegetation indices include normalized vegetation index (NDI) for both growing season and non-growing season, normalized green light NDI, normalized difference water vegetation index, soil-regulating vegetation index, improved soil-regulating vegetation index, atmospheric resistance vegetation index, differential vegetation index, difference vegetation index, enhanced vegetation index, total vegetation index, ratio vegetation index, improved triangular vegetation index, MERIS terrestrial chlorophyll index, plant senescence reflectance index, red-edged chlorophyll index, wide dynamic range vegetation index, visible light atmospheric resistance index, improved photochemical reflectance index, and the difference between the same vegetation index for both growing season and non-growing season. The salinity index includes salinity indices 1 to 13 for the growing season and non-growing season, as well as the difference between the same salinity index for the growing season and non-growing season. The texture features include the second moment of the first principal component and the second principal component, contrast, correlation, variance, inverse moment, mean sum, variance sum, entropy sum, entropy sum, and difference entropy; the first principal component and the second principal component are obtained by principal component analysis of the growing season remote sensing image. The terrain features include slope, aspect, and elevation.

7. The method for monitoring the canopy coverage of sparse forests according to claim 3, characterized in that, Each of the grids is 3×3 pixels in size.

8. A device for monitoring the canopy cover of sparse forests, characterized in that, include: The extraction module is used to extract the target features of the sparse forest to be monitored from remote sensing images of the growing season, remote sensing images of the non-growing season, and SRTM data. The estimation module is used to input the target features of the sparse forest to be monitored into the sparse forest canopy coverage estimation model and output the spatial distribution map of the canopy coverage of the sparse forest to be monitored; the sparse forest canopy coverage estimation model is constructed based on the target features of the sparse forest plots and the canopy coverage sample data of the sparse forest plots.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the sparse forest canopy coverage monitoring method as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the sparse forest canopy coverage monitoring method as described in any one of claims 1 to 7.

Images