An unmanned aerial vehicle-based detection and analysis method and system for arbor, shrub and grass vegetation

Abstract

The application relates to the technical field of ecological remote sensing and intelligent restoration, in particular to a method and system for detecting and analyzing arbor, shrub and grass vegetation based on a UAV, which comprises the following steps: fusing canopy height, texture and morphological features to realize fine classification of arbor, shrub and grass; combining multi-temporal UAV images and long-time sequence satellite data to construct a multi-dimensional feature vector, generating a 1-meter resolution vegetation degradation map through an optimized random forest model; and formulating a differentiated rejuvenation scheme according to degradation grades, vegetation types and terrain conditions. The application can effectively solve the problems of vegetation misclassification, low degradation identification accuracy and lack of adaptability of rejuvenation measures, and improve the accuracy and efficiency of ecological restoration.

Description

Technical Field

[0001] This invention belongs to the field of ecological environment monitoring and remote sensing technology, specifically a method and system for detecting and analyzing tree, shrub and grass vegetation based on unmanned aerial vehicles (UAVs). Background Technology

[0002] Existing vegetation regeneration practices largely rely on ground surveys or low-resolution remote sensing data for degradation assessment, making it difficult to achieve fine differentiation of tree, shrub, and grass vegetation types and accurate identification of degradation levels over large areas. Although UAV remote sensing technology has been used to acquire high-resolution imagery and generate digital surface models (DSMs), relying solely on canopy height for vegetation classification is susceptible to interference from morphologically similar objects such as tree seedlings and dense herbaceous plants, leading to biased classification results. Furthermore, while long-term satellite data can reflect macroscopic trends in vegetation degradation, its spatial resolution is typically on the order of tens of meters, creating a scale mismatch with the centimeter-level details of UAV imagery and limiting the detailed mapping of degraded areas.

[0003] Therefore, the need for high-precision vegetation classification and degradation identification by integrating multi-source remote sensing data and combining machine learning methods is becoming increasingly prominent. Existing methods are susceptible to redundant information when the feature dimension is high, and lack differentiated update and rejuvenation strategies for different terrains and vegetation types, resulting in a lack of systematic support for the formulation of rejuvenation measures and the evaluation of their effects. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a method and system for detecting and analyzing tree, shrub and grass vegetation based on UAVs, in order to address the above-mentioned deficiencies of the prior art. The invention aims to solve the problems of confusion between tree seedlings and shrubs, and between dense herbs and shrubs caused by relying solely on canopy height for vegetation classification in the prior art, as well as the problem of insufficient accuracy in identifying degraded areas caused by scale mismatch between long-term remote sensing data and high-resolution UAV images. At the same time, it overcomes the deficiencies of rejuvenation measures lacking terrain adaptability and vegetation type specificity.

[0005] The technical solution adopted by this invention to solve the technical problem is as follows:

[0006] This invention provides a method for detecting and analyzing tree, shrub, and grass vegetation based on unmanned aerial vehicles (UAVs), comprising:

[0007] Acquire orthophotos and digital surface models (DSMs) of vegetation within a preset area collected by drones, and generate canopy height models (CHMs) based on digital terrain models (DTMs).

[0008] Based on the CHM, the height information CHM value of the vegetation is extracted, and the vegetation is preliminarily classified as trees, shrubs, and herbs.

[0009] The texture and morphological features of the vegetation are extracted from the orthophoto, and the height information of the vegetation is combined to correct the preliminary classification results and obtain refined vegetation classification results.

[0010] By integrating multi-temporal UAV imagery and long-term satellite remote sensing data, a multi-dimensional feature vector is constructed. A machine learning model is used to identify the vegetation degradation level and generate a vegetation degradation map with a spatial resolution of 1 meter.

[0011] Based on the aforementioned vegetation degradation map, vegetation type, and topographic conditions, differentiated regeneration and rejuvenation plans were formulated; among which...

[0012] The vegetation types include trees, shrubs, and herbs.

[0013] In some embodiments of this application, the step of extracting the height information CHM value of the vegetation based on the CHM and performing a preliminary classification of the vegetation as trees, shrubs, and herbs includes:

[0014] The vegetation with a CHM value greater than 4 meters was marked as a candidate tree;

[0015] The vegetation with a CHM value greater than 1 meter and less than 2 meters is marked as a candidate shrub or tree seedling;

[0016] The vegetation with a CHM value less than 0.3 meters was marked as a candidate herb;

[0017] Also includes:

[0018] Based on the elevation gradient of the digital terrain model (DTM), the slope value of each pixel in the orthophoto is obtained, and a slope raster map is output.

[0019] The profile curvature q of the orthophoto is extracted based on the digital terrain model (DTM). Areas with a slope value greater than 15° and a profile curvature q less than 0 are marked as gully areas.

[0020] When vegetation located in gully areas with a CHM value greater than 1 meter and less than 2 meters is marked as a candidate shrub or tree seedling, an adaptive adjustment is performed, and vegetation with a CHM value greater than 1 meter and less than 2 meters is remarked as a candidate tree.

[0021] In some embodiments of this application, the step of extracting the texture and morphological features of the vegetation from the orthophoto and refining the preliminary classification results by combining the vegetation height information includes:

[0022] The Gray-Level Co-occurrence Matrix (GLCM) algorithm is used to calculate three texture metrics for each candidate object: energy, contrast, and entropy.

[0023] The candidate objects include the candidate trees, the candidate shrubs, the tree seedlings, and the candidate herbs;

[0024] Calculate the aspect ratio and shape index of each candidate object as morphological features;

[0025] For the candidate objects with a CHM value greater than 1 meter and less than 2 meters, if the aspect ratio is greater than 0.8 and the shape index is between 1.0 and 1.3, they are classified as tree seedlings.

[0026] For candidate objects with a CHM value greater than 1 meter and less than 2 meters, if the aspect ratio is less than 0.6, the shape index is greater than 1.8, and they are distributed in a clump-like cluster, they are retained as shrubs.

[0027] Objects with a CHM value less than 0.3 meters, an energy value greater than 0.7, an entropy value less than 2.5, and a continuous and uniform texture are classified as herbaceous.

[0028] In some embodiments of this application, the fusion of multi-temporal UAV imagery and long-term satellite remote sensing data to construct a multi-dimensional feature vector includes:

[0029] Landsat images of the preset area within a preset number of years were obtained using the Google Earth Engine platform, and NDVI time series were calculated. A vegetation degradation level map with a resolution of 25 meters was generated by Theil-Sen trend analysis and Mann-Kendall test.

[0030] RGB orthophotos of drones were acquired at two preset time points during the vegetation growing season. Three visible light vegetation indices, namely VDVI, EXG, and IRGBVI, were calculated respectively, and six original index layers were obtained.

[0031] The 3 cm resolution index layer was resampled to 1 m resolution using bilinear interpolation and georeferenced with the vegetation degradation level map.

[0032] The six original index layers are subtracted band by band to generate VDVI. diff EXG diff IRGBVI diff Three difference layers;

[0033] The six original index layers and the three difference layers are stacked into a 9-band TIF file.

[0034] In some embodiments of this application, the identification of vegetation degradation levels using a machine learning model includes:

[0035] Using each pixel of the vegetation degradation level map as a sample unit, within a 25×25 window of the corresponding 1-meter resolution 9-band TIF file, five statistical features are extracted for each band: mean, standard deviation, median, 25th percentile, and 75th percentile, forming a 45-dimensional feature vector.

[0036] A random forest classifier is used, with n set to n. estimators =100, random state =42, and enable class weight ='balanced' parameter;

[0037] The initial model was trained using the 45-dimensional feature vectors, and the top 15 effective features were selected by ranking them by feature importance.

[0038] The optimized model was retrained based on the 15 effective feature sets and used for subsequent predictions.

[0039] In some embodiments of this application, generating a vegetation degradation map with a spatial resolution of 1 meter includes:

[0040] A 25×25 sliding window is used for the preset area, and a corresponding 45-dimensional feature vector is extracted for each 1-meter pixel;

[0041] Select 15 feature columns from the feature matrix that correspond to the optimization model;

[0042] The feature matrix formed by the 15 selected feature columns is input into the optimization model, and the binary classification label of each 1-meter pixel is output.

[0043] The one-dimensional prediction results are reshaped into a two-dimensional raster according to the number of rows and columns of the original image, and given spatial reference information consistent with the orthophoto, and the vegetation degradation map with a spatial resolution of 1 meter is output.

[0044] In some embodiments of this application, the step of formulating a differentiated regeneration and rejuvenation plan based on the vegetation degradation map, vegetation type, and terrain conditions includes:

[0045] The preset area is divided into four types of terrain: low mountain and hilly area, fixed sand area, semi-fixed sand area, and low valley area.

[0046] For each type of terrain, a coping method is set:

[0047] In the low hilly areas, strip coppicing along contour lines is adopted, with a width of 20–30 meters and a retention strip of 50–75 meters. In fixed sand areas, checkerboard-like block coppicing is adopted, with a single block area of ​​less than 5 hectares. In semi-fixed sand areas, strip coppicing perpendicular to the prevailing wind direction is adopted, with a coppicing width of less than 25 meters and a retention strip of 75 meters or more. In the low-lying river valleys, alternating row and strip coppicing is adopted, with each operation covering 25%–30% of the total area.

[0048] Based on the vegetation degradation map and the vegetation type, the coppicing intensity, stubble height and auxiliary measures corresponding to the terrain conditions are set.

[0049] In some embodiments of this application, it also includes:

[0050] The differential slope is calculated based on the Digital Surface Model (DSM), and the CHM value is corrected according to the differential slope; wherein,

[0051] When the differential slope is greater than 15° and less than or equal to 20°, the CHM value is reduced by 0.2 meters;

[0052] When the differential slope is greater than 20° and less than or equal to 25°, the CHM value is reduced by 0.3 meters;

[0053] When the differential slope is greater than 25° and less than or equal to 30°, the CHM value is reduced by 0.4 meters;

[0054] For candidate objects with a corrected CHM value greater than 1 meter and less than 2 meters, if the aspect ratio is greater than 0.8 and the shape index is between 1.0 and 1.3, they are classified as tree seedlings.

[0055] For candidate objects with a corrected CHM value greater than 1 meter and less than 2 meters, if the aspect ratio is less than 0.6, the shape index is greater than 1.8, and they are distributed in a clump-like cluster, they are retained as shrubs.

[0056] Objects with a corrected CHM value less than 0.3 meters, an energy value greater than 0.7, an entropy value less than 2.5, and a continuous and uniform texture are classified as herbaceous.

[0057] In some embodiments of this application, it also includes:

[0058] The CHM value is then corrected a second time based on the profile curvature q of the orthophoto extracted from the digital terrain model (DTM).

[0059] When the profile curvature q is less than 0, the CHM value is increased by 0.3 meters;

[0060] When the profile curvature q is greater than 0, subtract 0.3 meters from the CHM value;

[0061] For candidate objects whose CHM value after secondary correction is greater than 1 meter and less than 2 meters, if the aspect ratio is greater than 0.8 and the shape index is between 1.0 and 1.3, they are classified as tree seedlings.

[0062] For candidate objects with a CHM value greater than 1 meter and less than 2 meters after secondary correction, if the aspect ratio is less than 0.6, the shape index is greater than 1.8, and the distribution is in a clump-like cluster, then they are retained as shrubs.

[0063] Objects with a CHM value less than 0.3 m after secondary correction, an energy value greater than 0.7, an entropy value less than 2.5, and a continuous and uniform texture are classified as herbaceous.

[0064] Based on the same technical concept, the present invention also provides a UAV-based tree, shrub, and grass vegetation detection and analysis system, applied to the aforementioned UAV-based tree, shrub, and grass vegetation detection and analysis method, comprising:

[0065] The vegetation height extraction module is used to acquire orthophotos of vegetation within a preset area collected by a drone, and generate a canopy height model CHM based on the digital surface model (DSM) and the digital terrain model (DTM).

[0066] The vegetation preliminary classification module is used to extract the height information of the vegetation based on the CHM and to perform preliminary classification of the vegetation into trees, shrubs and herbs.

[0067] The feature fusion correction module is used to extract the texture and morphological features of the vegetation from the orthophoto, combine the height information of the vegetation, correct the preliminary classification results, and obtain a refined vegetation classification result.

[0068] The multi-source data fusion module is used to fuse multi-temporal UAV imagery and long-term satellite remote sensing data, construct multi-dimensional feature vectors, identify vegetation degradation levels through machine learning models, and generate vegetation degradation maps with a spatial resolution of 1 meter.

[0069] The rejuvenation plan generation module is used to formulate differentiated renewal and rejuvenation plans based on the vegetation degradation map, vegetation type, and terrain conditions; wherein,

[0070] The vegetation types include trees, shrubs, and herbs.

[0071] Beneficial effects:

[0072] Compared with existing technologies, this invention solves the problem of vegetation misclassification caused by relying solely on canopy height by integrating canopy height, texture features, and morphological features. By constructing multi-temporal dynamic difference features and employing an iterative optimization strategy based on feature importance, it achieves accurate downscaling of vegetation degradation levels from 25 meters to 1 meter resolution under limited sample conditions. By coupling topographic conditions, vegetation type, and degradation level, it generates spatially adaptable renewal and rejuvenation schemes, avoiding "one-size-fits-all" management. This method can complete large-scale, high-precision vegetation status diagnosis and rejuvenation decision support without relying on manual ground surveys, significantly improving the scientific rigor and implementation efficiency of ecological restoration projects. Attached Figure Description

[0073] Figure 1 This is a flowchart of the method for detecting and analyzing tree, shrub, and grass vegetation based on unmanned aerial vehicles in an embodiment of the present invention;

[0074] Figure 2 This is a functional block diagram of the tree, shrub and grass vegetation detection and analysis system based on UAV in an embodiment of the present invention. Detailed Implementation

[0075] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0076] In the description of this application, it should be understood that the terms "center", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application.

[0077] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0078] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the communication between the inner sides of two components. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0079] See Figure 1 As shown, this invention provides a method for detecting and analyzing tree, shrub, and grass vegetation based on unmanned aerial vehicles (UAVs), including:

[0080] S101: Acquire orthophotos and digital surface models (DSM) of vegetation within a preset area collected by drones, and generate a canopy height model (CHM) based on the digital terrain model (DTM).

[0081] S201: Extract vegetation height information CHM values ​​based on CHM, and perform preliminary classification of vegetation into trees, shrubs and herbs;

[0082] S301: Extract the texture and morphological features of vegetation from orthophotos, combine them with vegetation height information, correct the preliminary classification results, and obtain refined vegetation classification results.

[0083] S401: By integrating multi-temporal UAV imagery with long-term satellite remote sensing data, a multi-dimensional feature vector is constructed. A machine learning model is used to identify the vegetation degradation level and generate a vegetation degradation map with a spatial resolution of 1 meter.

[0084] S501: Based on vegetation degradation maps, vegetation types, and topographic conditions, develop differentiated regeneration and rejuvenation plans; among which,

[0085] The vegetation types include trees, shrubs, and herbs.

[0086] In one specific embodiment of this application, the height information (CHM value) of vegetation is extracted based on CHM, and the vegetation is preliminarily classified into trees, shrubs, and herbs, including:

[0087] Vegetation with a CHM value greater than 4 meters was marked as a candidate tree;

[0088] Vegetation with a CHM value greater than 1 meter and less than 2 meters was marked as a candidate shrub or tree seedling;

[0089] Vegetation with a CHM value less than 0.3 meters was marked as a candidate herb;

[0090] Also includes:

[0091] Based on the elevation gradient of the digital terrain model (DTM), the slope value of each pixel in the orthophoto is obtained and the slope raster map is output.

[0092] The profile curvature q of the orthophoto is extracted from the digital terrain model (DTM). Areas with a slope greater than 15° and a profile curvature q less than 0 are marked as gully areas.

[0093] When vegetation in gully areas with a CHM value greater than 1 meter and less than 2 meters is marked as a candidate shrub or tree seedling, adaptive adjustment is performed, and vegetation with a CHM value greater than 1 meter and less than 2 meters is remarked as a candidate tree.

[0094] The beneficial effect is that CHM data can directly reflect the vertical structural characteristics of vegetation, providing an objective and quantitative basis for classification. By setting a reasonable height threshold, trees, shrubs, and herbaceous plants can be quickly distinguished, avoiding the high cost and low efficiency of traditional methods that rely on manual observation or complex image recognition algorithms. Compared with classification methods based on texture, color, or spectral features, height-based classification methods are simpler, more stable, and less affected by lighting conditions or seasonal changes. Especially in areas with dense vegetation cover or complex terrain, CHM can more accurately reflect the actual height of vegetation. Furthermore, fixed height thresholds can fail in complex terrain (such as steep slopes and gullies). By introducing a terrain curvature compensation mechanism to construct a closed-loop adaptive threshold adjustment method, the discrimination accuracy is further improved.

[0095] In one specific embodiment of this application, texture and morphological features of vegetation are extracted from orthophotos, and the preliminary classification results are corrected by combining vegetation height information, including:

[0096] The Gray-Level Co-occurrence Matrix (GLCM) algorithm is used to calculate three texture metrics for each candidate object: energy, contrast, and entropy.

[0097] Candidates include candidate trees, candidate shrubs, tree seedlings, and candidate herbs;

[0098] Calculate the aspect ratio and shape index of each candidate object as morphological features;

[0099] For candidate objects with a CHM value greater than 1 meter and less than 2 meters, if the length-to-width ratio is greater than 0.8 and the shape index is between 1.0 and 1.3, they are classified as tree seedlings.

[0100] For candidate objects with a CHM value greater than 1 meter and less than 2 meters, if the aspect ratio is less than 0.6, the shape index is greater than 1.8, and the distribution is in a clump-like cluster, then they are retained as shrubs.

[0101] Objects with a CHM value less than 0.3 meters, an energy value greater than 0.7, an entropy value less than 2.5, and a continuous and uniform texture are classified as herbaceous.

[0102] In one specific embodiment of this application, a multi-dimensional feature vector is constructed by fusing multi-temporal UAV imagery and long-term satellite remote sensing data, including:

[0103] Landsat images of a preset area within a preset number of years were acquired using the Google Earth Engine platform, and NDVI time series were calculated. A vegetation degradation level map with a resolution of 25 meters was generated through Theil-Sen trend analysis and Mann-Kendall test.

[0104] RGB orthophotos of drones were acquired at two preset time points during the vegetation growing season. Three visible light vegetation indices, namely VDVI, EXG, and IRGBVI, were calculated respectively, and six original index layers were obtained.

[0105] The 3 cm resolution index layer was resampled to 1 meter resolution using bilinear interpolation and georeferenced with the vegetation degradation level map.

[0106] The six original index layers are subtracted band by band to generate VDVI. diff EXG diff IRGBVI diff Three difference layers;

[0107] Stack the 6 original index layers and 3 difference layers into a 9-band TIF file.

[0108] The beneficial effect is that this method is not only applicable to typical ecosystems such as forests and grasslands, but also to vegetation degradation monitoring in complex environments such as wetlands and farmland. The response characteristics of different vegetation types can be optimized by adjusting the index parameters. It is applicable from large-scale regional monitoring to small-scale local degradation area identification (such as a certain woodland or grassland). Furthermore, it can flexibly adjust the data resolution and time span to meet the needs of different users.

[0109] In one specific embodiment of this application, vegetation degradation level identification is performed using a machine learning model, including:

[0110] Using each pixel of the vegetation degradation level map as a sample unit, within a 25×25 window of the corresponding 1-meter resolution 9-band TIF file, five statistical features are extracted for each band: mean, standard deviation, median, 25th percentile, and 75th percentile, forming a 45-dimensional feature vector.

[0111] A random forest classifier is used, with n set to n. estimators =100, randomstate =42, and enable class weight ='balanced' parameter;

[0112] The initial model was trained using 45-dimensional feature vectors, and the top 15 effective features were selected by ranking them by importance.

[0113] The model was retrained and optimized based on 15 effective feature sets and used for subsequent predictions.

[0114] In one specific embodiment of this application, generating a vegetation degradation map with a spatial resolution of 1 meter includes:

[0115] A 25×25 sliding window is used for the preset area, and a corresponding 45-dimensional feature vector is extracted for each 1-meter pixel;

[0116] Select 15 feature columns from the feature matrix that correspond to the optimized model;

[0117] The feature matrix formed by the 15 selected feature columns is input into the optimization model, and the binary classification label of each 1-meter pixel is output.

[0118] The one-dimensional prediction results are reshaped into a two-dimensional raster according to the number of rows and columns of the original image, and given spatial reference information consistent with the orthophoto, and a vegetation degradation map with a spatial resolution of 1 meter is output.

[0119] In one specific embodiment of this application, a differentiated regeneration and rejuvenation plan is formulated based on vegetation degradation maps, vegetation types, and topographic conditions, including:

[0120] The pre-defined area is divided into four types of terrain: low mountain and hilly area, fixed sand area, semi-fixed sand area, and low valley area.

[0121] For each type of terrain, a coppicing method is set:

[0122] In low mountain and hilly areas, strip coppicing along contour lines is adopted, with a width of 20–30 meters and a retention strip of 50–75 meters. In fixed sand areas, checkerboard-like block coppicing is adopted, with a single block area of ​​less than 5 hectares. In semi-fixed sand areas, strip coppicing perpendicular to the prevailing wind direction is adopted, with a coppicing width of less than 25 meters and a retention strip of 75 meters or more. In low-lying river valleys, alternating row and strip coppicing is adopted, with each operation covering 25%–30% of the total area.

[0123] Based on the vegetation degradation map and vegetation type, set the coppicing intensity, stubble height and auxiliary measures corresponding to the terrain conditions.

[0124] In one specific embodiment of this application, it further includes:

[0125] Differential slope is calculated based on the Digital Surface Model (DSM), and the CHM value is corrected according to the differential slope; among which,

[0126] When the differential slope is greater than 15° and less than or equal to 20°, subtract 0.2 meters from the CHM value;

[0127] When the differential slope is greater than 20° and less than or equal to 25°, subtract 0.3 meters from the CHM value;

[0128] When the differential slope is greater than 25° and less than or equal to 30°, subtract 0.4 meters from the CHM value;

[0129] For candidate objects with a corrected CHM value greater than 1 meter and less than 2 meters, if the aspect ratio is greater than 0.8 and the shape index is between 1.0 and 1.3, they are classified as tree seedlings.

[0130] For candidate objects with a corrected CHM value greater than 1 meter and less than 2 meters, if the aspect ratio is less than 0.6, the shape index is greater than 1.8, and the distribution is in a clump-like cluster, then they are retained as shrubs.

[0131] Objects with a corrected CHM value less than 0.3 meters, an energy value greater than 0.7, an entropy value less than 2.5, and a continuous and uniform texture are classified as herbaceous.

[0132] The beneficial effect is that in the initial CHM classification, vegetation with a height of 1-2 meters (candidate shrubs / tree seedlings) has a high misclassification rate due to terrain slope interference (e.g., when the slope of hilly areas is >15°, the CHM value is compressed and seedlings are misclassified as shrubs). By constructing a dynamic threshold adjustment mechanism, the failure of fixed thresholds under complex terrain is avoided, and the accuracy of analysis and detection is further improved.

[0133] In one specific embodiment of this application, it further includes:

[0134] The CHM value is then corrected a second time based on the profile curvature q of the orthophoto extracted from the digital terrain model (DTM).

[0135] When the profile curvature q is less than 0, increase the CHM value by 0.3 meters;

[0136] When the profile curvature q is greater than 0, subtract 0.3 meters from the CHM value;

[0137] For candidate objects with a CHM value greater than 1 meter and less than 2 meters after secondary correction, if the aspect ratio is greater than 0.8 and the shape index is between 1.0 and 1.3, they are classified as tree seedlings.

[0138] For candidate objects with a CHM value greater than 1 meter and less than 2 meters after secondary correction, if the aspect ratio is less than 0.6, the shape index is greater than 1.8, and the distribution is in a clump-like cluster, then they are retained as shrubs.

[0139] Objects with a CHM value less than 0.3 meters after secondary correction, an energy value greater than 0.7, an entropy value less than 2.5, and a continuous and uniform texture are classified as herbaceous.

[0140] The beneficial effect is that, since the correction rules only cover the 15°–30° range, when the slope is <15° or >30° (such as steep cliffs or plains), the CHM value is not corrected, resulting in a vegetation height estimation bias of >15%, which causes significant misjudgments in mountainous remote sensing classification. By introducing the aforementioned profile curvature q for dynamic terrain curvature compensation mechanism, combined with secondary correction, rule breaks can be effectively eliminated, further ensuring the accuracy of classification results.

[0141] See Figure 2 As shown, the present invention also provides a UAV-based tree, shrub, and grass vegetation detection and analysis system, applied in UAV-based tree, shrub, and grass vegetation detection and analysis methods, including:

[0142] The vegetation height extraction module is used to acquire orthophotos of vegetation within a preset area collected by a drone, and generate a canopy height model CHM based on the digital surface model (DSM) and the digital terrain model (DTM).

[0143] The vegetation preliminary classification module is used to extract vegetation height information based on CHM and to perform preliminary classification of vegetation into trees, shrubs and herbs.

[0144] The feature fusion correction module is used to extract the texture and morphological features of vegetation from orthophotos, combine them with the height information of vegetation, correct the preliminary classification results, and obtain refined vegetation classification results.

[0145] The multi-source data fusion module is used to fuse multi-temporal UAV imagery and long-term satellite remote sensing data, construct multi-dimensional feature vectors, identify vegetation degradation levels through machine learning models, and generate vegetation degradation maps with a spatial resolution of 1 meter.

[0146] The rejuvenation plan generation module is used to develop differentiated renewal and rejuvenation plans based on vegetation degradation maps, vegetation types, and terrain conditions; among them,

[0147] The vegetation types include trees, shrubs, and herbs.

[0148] Next, using a specific region as an example, we will describe in detail the complete implementation process of a method and system for detecting and analyzing tree, shrub, and grass vegetation based on unmanned aerial vehicles (UAVs).

[0149] The region covers a total area of ​​approximately 120 square kilometers and includes four typical geomorphic units: low mountains and hills, fixed sand dunes, semi-fixed sand dunes, and river valleys. The vegetation types include artificial forests of Pinus sylvestris (trees), Caragana korshinskii (shrubs), and Stipa herbaceous communities. These areas exhibit varying degrees of degradation and urgently require high-precision classification and differentiated rejuvenation decision support.

[0150] This embodiment first performs orthophoto and DSM acquisition. A drone equipped with a Zenmuse P1 full-frame camera was used to conduct aerial photography during two key time periods (June and August) of the vegetation growing season. The flight altitude was set at 90 meters, the ground sampling distance at 3 centimeters, and the forward and lateral overlap rates were both set at 80%. High-precision POS data was acquired using RTK / PPK differential positioning technology, and Pix4Dmapper software was used for aerial triangulation and image stitching to generate a global RGB orthophoto and the corresponding digital surface model (DSM). Simultaneously, a 1:10000 scale digital terrain model (DTM) (5 m raster resolution) for the area was acquired and resampled to 3 cm resolution using bicubic convolution interpolation to ensure strict alignment with the DSM on the spatial grid.

[0151] Subsequently, a Crown Height Model (CHM) is generated. The obtained Direct Scaling Model (DSM) is subtracted pixel-by-pixel from the resampled Direct Scaling Model (DTM), i.e., CHM = DSM - DTM, to generate the Crown Height Model (CHM). This CHM accurately reflects the vertical height of the top of the vegetation canopy relative to the bare ground, providing basic data for subsequent initial vegetation height classification.

[0152] Next, initial vegetation height classification is performed. Based on CHM (Content Height Map), height thresholds are set to initially classify vegetation: all connected objects with CHM values ​​greater than 4 meters are marked as candidate trees; objects with CHM values ​​between 1 and 2 meters are marked as candidate shrubs or tree seedlings; and objects with CHM values ​​less than 0.3 meters are marked as candidate herbs. To further prevent misclassification caused by special terrain, the slope value of each pixel in the orthophoto is obtained through the elevation gradient of the Digital Terrain Model (DTM), and a slope raster atlas is output. The profile curvature q of the orthophoto is extracted based on the DTM. Areas with slope values ​​greater than 15° and profile curvature q less than 0 are marked as gullies. When vegetation in gullies with CHM values ​​greater than 1 meter and less than 2 meters is marked as candidate shrubs or tree seedlings, adaptive adjustment is performed, and vegetation with CHM values ​​greater than 1 meter and less than 2 meters is remarked as candidate trees. This process is implemented using the object-oriented segmentation module in ENVI software, employing a multi-scale segmentation algorithm with a scale parameter set to 30, a shape factor of 0.3, and a compactness of 0.7, ensuring that the object boundary matches the actual vegetation canopy outline.

[0153] Subsequently, texture and morphological feature extraction was performed. For each generated candidate object, texture and morphological features of its internal pixels were extracted from a 3 cm resolution orthophoto. Texture features were calculated using the Gray-Level Co-occurrence Matrix (GLCM) algorithm, with a window size of 7×7 pixels and the orientation averaged at 0°, 45°, 90°, and 135°. Energy, Contrast, and Entropy were calculated. Morphological features were derived by calculating the minimum bounding rectangle of the object to obtain the aspect ratio and shape index. The shape index was calculated using the following formula:

[0154] ;

[0155] In the formula: the numerator (Perimeter) represents the boundary length of the figure, and the denominator represents the circumference of the circle;

[0156] All feature values ​​are associated with attributes based on object IDs, forming a structured feature table.

[0157] Next, refined vegetation classification is performed. Combining the initial height classification results with extracted texture and morphological features, ambiguous categories are corrected. Specific rules are as follows: For objects with CHM values ​​in the 1–2 meter range, if their aspect ratio is greater than 0.8 and their shape index is between 1.0 and 1.3 (close to circular or elliptical), they are identified as tree seedlings and reclassified as trees; if their aspect ratio is less than 0.6, their shape index is greater than 1.8, and they exhibit a clump-like clustered distribution (verified through neighborhood density analysis), they are retained as shrubs; for objects with CHM values ​​less than 0.3 meters, if their texture energy value is higher than 0.7, their entropy value is lower than 2.5, and they are continuously and uniformly distributed throughout the entire object area, they are classified as herbs. The final output is a 1-meter resolution vegetation type layer containing trees, shrubs, and herbs. This layer maintains spatial consistency through resampling and subsequent degradation analysis.

[0158] Subsequently, multi-temporal UAV and satellite data fusion was performed. First, Landsat 5 / 7 / 8 / 9 series satellite imagery was used on the Google Earth Engine (GEE) platform to acquire NDVI time-series data for a preset area over 20 years. The Theil-Sen slope estimation method was used to calculate the NDVI change trend, and combined with the Mann-Kendall (MK) significance test, areas with insignificant changes were removed, generating a vegetation degradation level map with a spatial resolution of 25 meters. The degradation level was divided into four categories: stable (no significant change), slight degradation (NDVI decreases but p>0.05), moderate degradation (NDVI decreases significantly, p≤0.05 and slope<-0.001), and severe degradation (NDVI decreases significantly and slope<-0.003). The latter three categories were merged into a binary label of "degradation" for model training.

[0159] Meanwhile, based on 3 cm UAV orthophotos acquired in June and August, three visible light vegetation indices were calculated: VDVI = (2G-RB) / (2G+R+B), EXG = 2G-RB, and IRGBVI = (GB) / (R+G+B), forming six index layers (three indices for each time phase). In the formulas, "G", "R", and "B" represent the values ​​of the green, red, and blue bands of pixels in the UAV orthophoto image, respectively. G (Green): Represents the pixel value of the green band in the image. In visible light images, the green band is often used to reflect the health status and chlorophyll content of plants because plants have high reflectivity in the green band. R (Red): Represents the pixel value of the red band in the image. The red band is also sensitive to plant health status, especially in vegetation index calculations, where it is used to detect the photosynthetic capacity of plants. B (Blue): Represents the pixel value of the blue band in the image. The blue band is mainly used in vegetation analysis to correct for atmospheric scattering effects or to be combined with other bands to improve the accuracy of vegetation indices. Subsequently, these six sets of 3 cm resolution index layers were resampled to 1 meter resolution using bilinear interpolation and georegistered with the 25-meter degraded map to ensure coordinate system consistency. Next, pixel-by-pixel subtraction was performed on the 1-meter layers of the same index at two different time phases to generate VDVI. diff EXG diff IRGBVI diff Three difference layers reflect the dynamic changes of vegetation index during the growing season.

[0160] Next, multidimensional feature vector construction is performed. The above 6 original index layers are stacked with 3 difference layers to form a 9-band 1-meter resolution TIF file. Centered on each pixel of the 25-meter degradation map, a 25×25 pixel sliding window (covering a 25-meter × 25-meter area) is drawn on the corresponding 9-band 1-meter data. For each band within the window, five statistics are calculated: mean, standard deviation, median, 25th percentile, and 75th percentile, resulting in a total of 9×5=45-dimensional feature vectors (see attached). Figure 2 The label is 15. Each 25-meter pixel corresponds to a 45-dimensional sample, and the label is its degradation binary classification result (degraded / non-degraded).

[0161] The next step is machine learning degradation identification. A random forest classifier from the scikit-learn library is used, with an initial n value. estimators =100, random state =42, and enable class weight ='balanced' to address the sample imbalance problem. The model is trained using all 45 features. After training, the importance scores of each feature are extracted (based on the reduction in Gini impurity), sorted in descending order, and the top 15 most discriminative features are selected (e.g., the mean of August EXG, the standard deviation of June VDVI, the 75th percentile of IRGBVI_diff, etc.) to form an optimized feature set. The random forest model is retrained based on this 15-dimensional feature subset to obtain the optimized degradation recognition model.

[0162] Next, a 1-meter degradation map is generated. For the 1-meter 9-band data within the preset area, a 25×25 sliding window is used to traverse each 1-meter pixel, extracting its corresponding 45-dimensional feature vector. Subsequently, 15 feature columns corresponding to the optimized model are selected from this vector to form a prediction input matrix. This matrix is ​​input into a trained random forest model, outputting a binary classification prediction label (degraded / non-degraded) for each 1-meter pixel. Finally, the one-dimensional prediction result array is reshaped into a two-dimensional raster matrix according to the number of rows and columns of the original image, and given spatial reference information consistent with the UAV orthophoto (including projection, coordinate origin, and pixel size), outputting a vegetation degradation map with a 1-meter spatial resolution.

[0163] Finally, a differentiated rejuvenation plan was developed. This step involved deep integration of multiple input layers: a 1-meter degradation map, a 1-meter vegetation type layer, and a topographic zoning map. The topographic zoning map, through analysis of the slope, curvature, and dune movement index of the 5-meter resolution DEM, divided the study area into four categories: low mountain and hilly areas (slope 5°–15°, bedrock exposed), fixed sand areas (dust height <2m, vegetation cover >40%), semi-fixed sand areas (dust height 2–5m, vegetation cover 20%–40%), and low valley areas (elevation <1200m, near the river channel). Different coppicing operation modes are preset for each type of terrain: in low mountain and hilly areas, strip coppicing along contour lines is adopted, with a coppicing width of 20-30 meters and a retention width of 50-75 meters to reduce soil erosion; in fixed sand areas, checkerboard-shaped block coppicing is adopted, with a single block area not exceeding 5 hectares to prevent wind erosion from expanding; in semi-fixed sand areas, strip coppicing perpendicular to the prevailing wind direction (NW-SE direction) is adopted, with a coppicing width not exceeding 25 meters and a retention width of not less than 75 meters to impede sand dune movement; in low-lying river valley areas, alternating row and strip coppicing is adopted, with the area of ​​each operation controlled at 25%-30% of the total area to ensure water conservation function.

[0164] Based on this, specific rejuvenation parameters are set by combining vegetation degradation levels (refined from a 1-meter degradation map into three levels: Level 1 for mild degradation, Level 2 for moderate degradation, and Level 3 for severe degradation) and vegetation types (trees, shrubs, and herbs). For example, for severely degraded shrubs (such as Caragana korshinskii), in semi-fixed sandy areas, strip coppicing with a stubble height of 15 cm is implemented, supplemented by hole fertilization (0.5 kg of organic fertilizer per hole); for moderately degraded tree seedlings, in low mountain and hilly areas, only thinning is carried out without coppicing, maintaining a density of 800 plants / hectare; for mildly degraded herbs, in low-lying river valleys, intermittent cutting is implemented, with a stubble height of 5 cm, to promote tiller regeneration. All strategy parameters are stored in the system database in the form of a rule base and are automatically matched and output by the rejuvenation scheme generation module.

[0165] This invention proposes a complete process for vegetation degradation and regeneration based on unmanned aerial vehicles (UAVs), which has the following significant effects:

[0166] (1) Achieving a scientific and procedural approach to vegetation degradation diagnosis: Existing technologies lack a unified basis for determining the need for vegetation regeneration and rejuvenation. This invention introduces high-resolution UAV imagery data, combines it with a digital surface model (DSM) and a digital terrain model (DTM) to generate a digital canopy height model (CHM), and calculates various visible vegetation indices, achieving refined classification and growth status analysis of trees, shrubs, and herbaceous vegetation. Based on this, machine learning methods are used to train the UAV imagery features with weak degradation level labels, outputting a refined degradation level map, providing an objective, quantitative, and procedural diagnostic basis for vegetation regeneration and rejuvenation, significantly improving the scientific nature and accuracy of decision-making.

[0167] (2) Providing precise and adaptable renewal and rejuvenation measures: Existing technologies often target single vegetation types, lacking comprehensive consideration of different vegetation types, different degrees of degradation, and different terrain conditions. This invention divides vegetation into three levels based on fine degradation levels and, combined with four terrain conditions—low mountain and hilly areas, fixed sand areas, semi-fixed sand areas, and low river valley areas—proposes differentiated coppicing methods (such as manual coppicing, mechanical coppicing, strip coppicing, and alternating row and strip coppicing) and specific tree, shrub, and grass vegetation renewal and rejuvenation implementation plans (such as complete renewal, selective renewal, and tending pruning of trees; intensive coppicing, selective coppicing, and light tending pruning of shrubs; and tillage improvement and reconstruction, coppicing combined with reseeding, and protective utilization of grasslands). This multi-dimensional and refined strategy ensures the precision of rejuvenation measures and their adaptability to the local ecological environment, avoids the negative impact of a "one-size-fits-all" approach, and improves the success rate of rejuvenation.

[0168] (3) Establishing an objective evaluation system for rejuvenation effects: The effects of existing rejuvenation technologies are often difficult to verify. This invention establishes a complete and quantifiable effect evaluation system by setting up a standardized control experimental design (enhanced treatment group, conventional treatment group, and natural control group), and dynamically monitoring key indicators such as vegetation sprouting, survival rate, growth status, and biomass accumulation before and after coppicing and at different stages. This is combined with subsequent UAV image data for continuous evaluation. This not only scientifically verifies the effectiveness of rejuvenation measures and provides data support for subsequent management, but also provides a reliable basis for cross-regional applicability assessment and technology promotion.

[0169] (4) Improved operational efficiency and reduced labor costs: Traditional vegetation surveys and effect assessments rely on extensive manual field investigations, which are time-consuming, labor-intensive, and difficult to cover large areas. This invention utilizes UAV technology for high-frequency, large-scale data collection, significantly improving data acquisition efficiency and reducing the intensity of fieldwork and labor costs. At the same time, the automated processing and analysis based on UAV imagery also greatly shortens the time cycle from data collection to diagnostic decision-making, making the renewal and rejuvenation work more efficient.

[0170] (5) Promoting the continuous improvement of the ecological environment in desertified areas: The complete set of vegetation detection, diagnosis, rejuvenation and evaluation processes proposed in this invention can systematically solve the problem of vegetation degradation in desert and potentially desertified areas. Precise diagnosis and targeted rejuvenation measures can effectively promote the rapid recovery and healthy growth of vegetation, enhance the ecological functions of vegetation such as windbreak and sand fixation, and soil and water conservation, thereby continuously improving the regional ecological environment and providing strong technical support for desertification control.

[0171] The above description is merely one embodiment of the present invention, but it cannot be used to limit the scope of the present invention. Any structural changes made based on the present invention, as long as they do not lose the essence of the present invention, should be considered to fall within the protection scope of the present invention and be subject to its restrictions.

[0172] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process and related descriptions of the system described above can be found in the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0173] It should be noted that the system provided in the above embodiments is only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the modules or steps in the embodiments of the present invention can be further decomposed or combined. For example, the modules in the above embodiments can be merged into one module, or further divided into multiple sub-modules to complete all or part of the functions described above. The names of the modules and steps involved in the embodiments of the present invention are only for distinguishing the various modules or steps and are not considered as an improper limitation of the present invention.

[0174] Those skilled in the art will recognize that the modules and method steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. The programs corresponding to the software modules and method steps can be placed in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or any other form of storage medium known in the art. To clearly illustrate the interchangeability of electronic hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in electronic hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the invention.

[0175] The term "comprising" or any other similar term is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus / device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent in such process, method, article, or apparatus / device.

[0176] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after such changes or substitutions will all fall within the scope of protection of the present invention.

[0177] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention.

Claims

1. A method for detecting and analyzing tree, shrub, and grass vegetation based on unmanned aerial vehicles (UAVs), characterized in that, include: Acquire orthophotos and digital surface models (DSMs) of vegetation within a preset area collected by drones, and generate canopy height models (CHMs) based on digital terrain models (DTMs). Based on the CHM, the height information CHM value of the vegetation is extracted, and the vegetation is preliminarily classified as trees, shrubs, and herbs. The texture and morphological features of the vegetation are extracted from the orthophoto, and the height information of the vegetation is combined to correct the preliminary classification results and obtain refined vegetation classification results. By integrating multi-temporal UAV imagery and long-term satellite remote sensing data, a multi-dimensional feature vector is constructed. A machine learning model is used to identify the vegetation degradation level and generate a vegetation degradation map with a spatial resolution of 1 meter. Based on the aforementioned vegetation degradation map, vegetation type, and topographic conditions, differentiated regeneration and rejuvenation plans were formulated; among which... The vegetation types include trees, shrubs, and herbs; The fusion of multi-temporal UAV imagery and long-term satellite remote sensing data constructs a multi-dimensional feature vector, including: Landsat images of the preset area within a preset number of years were obtained using the Google Earth Engine platform, and NDVI time series were calculated. A vegetation degradation level map with a resolution of 25 meters was generated by Theil-Sen trend analysis and Mann-Kendall test. RGB orthophotos of drones were acquired at two preset time points during the vegetation growing season. Three visible light vegetation indices, namely VDVI, EXG, and IRGBVI, were calculated respectively, and six original index layers were obtained. The 3 cm resolution index layer was resampled to 1 m resolution using bilinear interpolation and georeferenced with the vegetation degradation level map. The six original index layers are subtracted band by band to generate VDVI. diff EXG diff IRGBVI diff Three difference layers; The six original index layers and the three difference layers are stacked into a 9-band TIF file; The identification of vegetation degradation levels using machine learning models includes: Using each pixel of the vegetation degradation level map as a sample unit, within a 25×25 window of the corresponding 1-meter resolution 9-band TIF file, five statistical features are extracted for each band: mean, standard deviation, median, 25th percentile, and 75th percentile, forming a 45-dimensional feature vector. A random forest classifier is used, with n set to n. estimators =100, random state =42, and enable class weight ='balanced' parameter; The initial model was trained using the 45-dimensional feature vectors, and the top 15 effective features were selected by ranking them by feature importance. The optimized model was retrained based on the 15 effective features and used for subsequent predictions.

2. The method for detecting and analyzing tree, shrub, and grass vegetation based on unmanned aerial vehicles according to claim 1, characterized in that, The process of extracting the height information (CHM value) of the vegetation based on the CHM, and performing a preliminary classification of the vegetation into trees, shrubs, and herbs, includes: The vegetation with a CHM value greater than 4 meters was marked as a candidate tree; The vegetation with a CHM value greater than 1 meter and less than 2 meters is marked as a candidate shrub or tree seedling; The vegetation with a CHM value less than 0.3 meters was marked as a candidate herb; Also includes: Based on the elevation gradient of the digital terrain model (DTM), the slope value of each pixel in the orthophoto is obtained, and a slope raster map is output. The profile curvature q of the orthophoto is extracted based on the digital terrain model (DTM). Areas with a slope value greater than 15° and a profile curvature q less than 0 are marked as gully areas. When vegetation located in gully areas with a CHM value greater than 1 meter and less than 2 meters is marked as a candidate shrub or tree seedling, an adaptive adjustment is performed, and vegetation with a CHM value greater than 1 meter and less than 2 meters is remarked as a candidate tree.

3. The method for detecting and analyzing tree, shrub, and grass vegetation based on unmanned aerial vehicles according to claim 2, characterized in that, The step of extracting the texture and morphological features of the vegetation from the orthophoto image, and combining this with the vegetation height information to correct the preliminary classification results includes: The Gray-Level Co-occurrence Matrix (GLCM) algorithm is used to calculate three texture metrics for each candidate object: energy, contrast, and entropy. The candidate objects include the candidate trees, the candidate shrubs, the tree seedlings, and the candidate herbs; Calculate the aspect ratio and shape index of each candidate object as morphological features; For the candidate objects with a CHM value greater than 1 meter and less than 2 meters, if the aspect ratio is greater than 0.8 and the shape index is between 1.0 and 1.3, they are classified as tree seedlings. For candidate objects with a CHM value greater than 1 meter and less than 2 meters, if the aspect ratio is less than 0.6, the shape index is greater than 1.8, and they are distributed in a clump-like cluster, they are retained as shrubs. Objects with a CHM value less than 0.3 meters, an energy value greater than 0.7, an entropy value less than 2.5, and a continuous and uniform texture are classified as herbaceous.

4. The method for detecting and analyzing tree, shrub, and grass vegetation based on unmanned aerial vehicles according to claim 1, characterized in that, The generation of a vegetation degradation map with a spatial resolution of 1 meter includes: A 25×25 sliding window is used for the preset area, and a corresponding 45-dimensional feature vector is extracted for each 1-meter pixel; Select 15 feature columns from the feature matrix that correspond to the optimization model; The feature matrix formed by the 15 selected feature columns is input into the optimization model, and the binary classification label of each 1-meter pixel is output. The one-dimensional prediction results are reshaped into a two-dimensional raster according to the number of rows and columns of the original image, and given spatial reference information consistent with the orthophoto, and the vegetation degradation map with a spatial resolution of 1 meter is output.

5. The method for detecting and analyzing tree, shrub, and grass vegetation based on unmanned aerial vehicles according to claim 4, characterized in that, The step of formulating differentiated regeneration and rejuvenation plans based on the vegetation degradation map, vegetation type, and topographic conditions includes: The preset area is divided into four types of terrain: low mountain and hilly area, fixed sand area, semi-fixed sand area, and low valley area. For each type of terrain, a coping method is set: In the low hilly areas, strip coppicing along contour lines is adopted, with a width of 20–30 meters and a retention strip of 50–75 meters. In fixed sand areas, checkerboard-like block coppicing is adopted, with a single block area of ​​less than 5 hectares. In semi-fixed sand areas, strip coppicing perpendicular to the prevailing wind direction is adopted, with a coppicing width of less than 25 meters and a retention strip of 75 meters or more. In the low-lying river valleys, alternating row and strip coppicing is adopted, with each operation covering 25%–30% of the total area. Based on the vegetation degradation map and the vegetation type, the coppicing intensity, stubble height and auxiliary measures corresponding to the terrain conditions are set.

6. The method for detecting and analyzing tree, shrub, and grass vegetation based on unmanned aerial vehicles according to claim 5, characterized in that, Also includes: The differential slope is calculated based on the Digital Surface Model (DSM), and the CHM value is corrected according to the differential slope; wherein, When the differential slope is greater than 15° and less than or equal to 20°, the CHM value is reduced by 0.2 meters; When the differential slope is greater than 20° and less than or equal to 25°, the CHM value is reduced by 0.3 meters; When the differential slope is greater than 25° and less than or equal to 30°, the CHM value is reduced by 0.4 meters; For candidate objects with a corrected CHM value greater than 1 meter and less than 2 meters, if the aspect ratio is greater than 0.8 and the shape index is between 1.0 and 1.3, they are classified as tree seedlings. For candidate objects with a corrected CHM value greater than 1 meter and less than 2 meters, if the aspect ratio is less than 0.6, the shape index is greater than 1.8, and they are distributed in a clump-like cluster, they are retained as shrubs. Objects with a corrected CHM value less than 0.3 meters, an energy value greater than 0.7, an entropy value less than 2.5, and a continuous and uniform texture are classified as herbaceous.

7. The method for detecting and analyzing tree, shrub, and grass vegetation based on unmanned aerial vehicles according to claim 6, characterized in that, Also includes: The CHM value is then corrected a second time based on the profile curvature q of the orthophoto extracted from the digital terrain model (DTM). When the profile curvature q is less than 0, the CHM value is increased by 0.3 meters; When the profile curvature q is greater than 0, subtract 0.3 meters from the CHM value; For candidate objects whose CHM value after secondary correction is greater than 1 meter and less than 2 meters, if the aspect ratio is greater than 0.8 and the shape index is between 1.0 and 1.3, they are classified as tree seedlings. For candidate objects with a CHM value greater than 1 meter and less than 2 meters after secondary correction, if the aspect ratio is less than 0.6, the shape index is greater than 1.8, and the distribution is in a clump-like cluster, then they are retained as shrubs. Objects with a CHM value less than 0.3 m after secondary correction, an energy value greater than 0.7, an entropy value less than 2.5, and a continuous and uniform texture are classified as herbaceous.

8. A drone-based system for detecting and analyzing tree, shrub, and grass vegetation, applied in the drone-based method for detecting and analyzing tree, shrub, and grass vegetation as described in any one of claims 1-7, characterized in that, include: The vegetation height extraction module is used to acquire orthophotos of vegetation within a preset area collected by a drone, and generate a canopy height model CHM based on the digital surface model (DSM) and the digital terrain model (DTM). The vegetation preliminary classification module is used to extract the height information of the vegetation based on the CHM and to perform preliminary classification of the vegetation into trees, shrubs and herbs. The feature fusion correction module is used to extract the texture and morphological features of the vegetation from the orthophoto, combine the height information of the vegetation, correct the preliminary classification results, and obtain a refined vegetation classification result. The multi-source data fusion module is used to fuse multi-temporal UAV imagery and long-term satellite remote sensing data, construct multi-dimensional feature vectors, identify vegetation degradation levels through machine learning models, and generate vegetation degradation maps with a spatial resolution of 1 meter. The rejuvenation plan generation module is used to formulate differentiated renewal and rejuvenation plans based on the vegetation degradation map, vegetation type, and terrain conditions; wherein, The vegetation types include trees, shrubs, and herbs.

Images