CN-121459186-B - Unmanned aerial vehicle-based arbor-shrub vegetation detection and analysis method and system

CN121459186BCN 121459186 BCN121459186 BCN 121459186BCN-121459186-B

Abstract

The application relates to the technical field of ecological remote sensing and intelligent restoration, in particular to a method and a system for detecting and analyzing arbor and shrub vegetation based on an unmanned plane, wherein the method comprises the steps of fusing canopy height, texture and morphological characteristics to realize arbor and shrub refined classification; the method comprises the steps of constructing a multidimensional feature vector by combining a multi-temporal unmanned aerial vehicle image and long-time sequence satellite data, generating a vegetation degradation graph with 1 meter resolution through an optimized random forest model, and formulating a differential rejuvenation scheme according to degradation grades, vegetation types and topography conditions. The method can effectively solve the problems of misclassification of vegetation, low degradation recognition precision and lack of suitability of rejuvenation measures, and improves the precision and efficiency of ecological restoration.

Inventors

Dong Zhepi
Zhang Aochong
HU SHI
WANG DONGLIANG
TIAN JING
CHEN PENGFEI

Assignees

中国科学院地理科学与资源研究所

Dates

Publication Date: 20260508
Application Date: 20260105

Claims (8)

1. The arbor and shrub vegetation detection and analysis method based on the unmanned aerial vehicle is characterized by comprising the following steps of: acquiring an orthophoto and a digital surface model DSM of vegetation in a preset area acquired by an unmanned aerial vehicle, and generating a canopy height model CHM based on the digital terrain model DTM; Extracting a height information CHM value of the vegetation based on the CHM, and primarily classifying the vegetation into arbor, shrub and herb; Extracting texture features and morphological features of the vegetation from the orthographic images, correcting the preliminary classification result by combining the height information of the vegetation, and obtaining a refined vegetation classification result; the multi-temporal unmanned aerial vehicle image and the long-time sequence satellite remote sensing data are fused, a multi-dimensional feature vector is constructed, vegetation degradation level identification is carried out through a machine learning model, and a vegetation degradation graph with 1 meter spatial resolution is generated; formulating a differential updating rejuvenation scheme according to the vegetation degradation map, vegetation types and terrain conditions, wherein, The vegetation types include arbor, shrub, and herb; the multi-temporal unmanned aerial vehicle image and the long-time sequence satellite remote sensing data are fused to construct a multi-dimensional feature vector, which comprises the following steps: Acquiring Landsat images in a preset period of the preset area based on a Google EARTH ENGINE platform, calculating an NDVI time sequence, and generating a vegetation degradation grade diagram with a resolution of 25 meters through Theil-Sen trend analysis and Mann-Kendall inspection; Acquiring RGB (red green blue) orthographic images of the unmanned aerial vehicle at two preset time nodes in a vegetation growing season, respectively calculating VDVI, EXG, IRGBVI three visible light vegetation indexes, and obtaining 6 original index layers; Resampling an index layer with 3 cm resolution to 1m resolution through bilinear interpolation, and performing geographic registration with the vegetation degradation level map; subtracting the 6 original index layers from each other in a band-by-band manner to generate VDVI diff 、EXG diff 、IRGBVI diff difference layers; Stacking 6 original index layers and 3 difference layers into a 9-band TIF file; the vegetation degradation level identification by the machine learning model comprises the following steps: taking each pixel of the vegetation degradation level diagram as a sample unit, extracting 5 statistical features in total from the mean value, standard deviation, median, 25% fractional number and 75% fractional number of each band in a 25×25 window of a 9-band TIF file with a resolution of 1 meter, and forming a 45-dimensional feature vector; setting n estimators =100、random state =42 with a random forest classifier, and enabling class weight = 'bandwidth' parameters; Training an initial model by using the 45-dimensional feature vector, and screening the first 15 effective features through feature importance ranking; the optimization model was retrained based on 15 of the valid features and used for subsequent predictions.
2. The unmanned aerial vehicle-based arbor and shrub vegetation detection and analysis method according to claim 1, wherein the extracting the height information CHM value of the vegetation based on the CHM and performing preliminary classification of arbor, shrub, herb on the vegetation comprises: Marking the vegetation with the CHM value greater than 4 meters as a candidate arbor; Marking the vegetation with the CHM value greater than 1 meter and less than 2 meters as candidate shrubs or arbor seedlings; Marking the vegetation with the CHM value less than 0.3 meters as a candidate herb; Further comprises: Acquiring a gradient value of each pixel in the orthographic image based on the elevation gradient of the digital terrain model DTM, and outputting a gradient grid atlas; extracting the section curvature q of the orthographic image according to the digital terrain model DTM, and marking a region with the gradient value larger than 15 degrees and the section curvature q smaller than 0 as a gully region; And when the vegetation is in a gully region and the CHM value is more than 1 meter and less than 2 meters, performing self-adaptive adjustment, and re-marking the vegetation with the CHM value of more than 1 meter and less than 2 meters as a candidate shrub or arbor seedling.
3. The unmanned aerial vehicle-based arbor and shrub vegetation detection and analysis method according to claim 2, wherein the extracting the texture features and morphological features of the vegetation from the orthographic image, and correcting the preliminary classification result by combining the height information of the vegetation, comprises: three texture indexes of energy, contrast and entropy of each candidate object are calculated by adopting a gray level co-occurrence matrix GLCM algorithm, wherein, The candidate objects comprise the candidate arbor, the candidate shrub, the arbor seedling and the candidate herb; calculating aspect ratio and shape index of each candidate object as morphological characteristics; classifying the candidate object with the CHM value greater than 1 meter and less than 2 meters as a arbor seedling if the aspect ratio is greater than 0.8 and the shape index is between 1.0 and 1.3; for the candidate object having the CHM value greater than 1 meter and less than 2 meters, if the aspect ratio is less than 0.6 and the shape index is greater than 1.8 and is clustered in clusters, then remains as shrubs; Objects with CHM values less than 0.3 meters, energy values greater than 0.7, entropy values less than 2.5, and continuous and uniform texture are classified as herbs.
4. The unmanned aerial vehicle-based arbor and shrub vegetation detection and analysis method of claim 1, wherein the generating a vegetation degradation map with a spatial resolution of 1 meter comprises: a 25 multiplied by 25 sliding window is adopted for the preset area, and a corresponding 45-dimensional feature vector is extracted for each 1-meter pixel; screening 15 feature columns corresponding to the optimization model from a feature matrix; Inputting a feature matrix formed by the screened 15 feature columns into the optimization model, and outputting a two-class label of each 1-meter pixel; And remolding a one-dimensional prediction result into a two-dimensional grid according to the number of rows and columns of an original image, endowing the two-dimensional grid with spatial reference information consistent with the orthographic image, and outputting the vegetation degradation graph with 1 meter spatial resolution.
5. The unmanned aerial vehicle-based arbor and shrub vegetation detection and analysis method according to claim 4, wherein the making of a differential update rejuvenation scheme according to the vegetation degradation map, vegetation type and terrain conditions comprises: Dividing the preset area into four terrains of a low hilly area, a fixed sand area, a semi-fixed sand area and a valley low area; setting stubble leveling modes for each type of terrain: The low hilly area adopts strip stubble leveling along a contour line, the bandwidth is 20-30 meters, the reserved belt is 50-75 meters, the fixed sand area adopts checkerboard block stubble leveling, the single block area is smaller than 5 hectares, the semi-fixed sand area adopts vertical main wind direction strip Ping Cha, the stubble leveling bandwidth is smaller than 25 meters, the reserved belt is greater than or equal to 75 meters, the valley low area adopts interlaced belt wheel stubble leveling, and the operation area of each time is 25-30 percent; Setting stubble leveling strength, stubble remaining height and auxiliary measures corresponding to the terrain conditions according to the vegetation degradation map and the vegetation types.
6. The unmanned aerial vehicle-based arbor and shrub vegetation detection and analysis method of claim 5, further comprising: Calculating a differential grade based on the digital surface model DSM and correcting the CHM value according to the differential grade, Subtracting 0.2 meters from the CHM value when the differential slope is greater than 15 ° and equal to or less than 20 °; subtracting 0.3 meters from the CHM value when the differential grade is greater than 20 ° and equal to or less than 25 °; subtracting 0.4 meter from the CHM value when the differential gradient is greater than 25 ° and equal to or less than 30 °; Classifying the candidate object with the corrected CHM value being greater than 1 meter and less than 2 meters as a arbor seedling if the aspect ratio is greater than 0.8 and the shape index is between 1.0 and 1.3; for the candidate object with the corrected CHM value being greater than 1 meter and less than 2 meters, if the aspect ratio is less than 0.6 and the shape index is greater than 1.8 and is clustered in a cluster, then remaining as shrubs; for objects with corrected CHM values of less than 0.3 meters, energy values of greater than 0.7, entropy values of less than 2.5, and continuous and uniform texture, the object is classified as herb.
7. The unmanned aerial vehicle-based arbor and shrub vegetation detection and analysis method of claim 6, further comprising: performing secondary correction on the CHM value according to the section curvature q of the orthographic image extracted by the digital terrain model DTM; increasing the CHM value by 0.3 meters when the profile curvature q is less than 0; subtracting 0.3 meters from the CHM value when the profile curvature q is greater than 0; classifying the candidate object with the secondary corrected CHM value being greater than 1 meter and less than 2 meters as a arbor seedling if the aspect ratio is greater than 0.8 and the shape index is between 1.0 and 1.3; For the candidate objects with the CHM value being more than 1 meter and less than 2 meters after the secondary correction, if the aspect ratio is less than 0.6 and the shape index is more than 1.8 and is clustered in a cluster shape, the candidate objects remain as shrubs; For objects with a CHM value less than 0.3m, an energy value greater than 0.7, an entropy value less than 2.5 and a continuous and uniform texture after the secondary correction, the objects are classified as herbs.
8. An unmanned aerial vehicle-based arbor and shrub vegetation detection and analysis system for use in the unmanned aerial vehicle-based arbor and shrub vegetation detection and analysis method according to any one of claims 1 to 7, comprising: the vegetation height extraction module is used for acquiring an orthographic image of vegetation in a preset area acquired by the unmanned aerial vehicle and generating a canopy height model CHM based on the digital surface model DSM and the digital terrain model DTM; The vegetation primary classification module is used for extracting the height information of the vegetation based on the CHM and primarily classifying the vegetation into arbor, shrub and herb; The feature fusion correction module is used for extracting texture features and morphological features of the vegetation from the orthographic images, correcting the preliminary classification result by combining the height information of the vegetation, and obtaining a refined vegetation classification result; The multi-source data fusion module is used for fusing the multi-time-phase unmanned aerial vehicle image and the long-time sequence satellite remote sensing data, constructing a multi-dimensional feature vector, identifying vegetation degradation level through a machine learning model, and generating a vegetation degradation graph with 1 meter of spatial resolution; A rejuvenation scheme generating module for making a differential updating rejuvenation scheme according to the vegetation degradation map, vegetation type and terrain condition, The vegetation types include arbor, shrub, and herb.

Description

Unmanned aerial vehicle-based arbor-shrub vegetation detection and analysis method and system Technical Field The invention belongs to the technical field of ecological environment monitoring and remote sensing, and particularly relates to a method and a system for detecting and analyzing arbor and shrub vegetation based on an unmanned aerial vehicle. Background The existing vegetation updating and rejuvenation practice is mostly dependent on ground investigation or low-resolution remote sensing data for degradation evaluation, and accurate identification of fine distinction and degradation degree of arbor and shrub vegetation types is difficult to realize in a large-scale area. Although unmanned aerial vehicle remote sensing technology has been used to acquire high resolution images and generate Digital Surface Models (DSMs), vegetation classification by only relying on canopy height is susceptible to interference from morphologically similar objects such as arbor seedlings, dense herbs, etc., resulting in bias in classification results. Meanwhile, although long-time sequence satellite data can reflect macroscopic trend of vegetation degradation, spatial resolution is usually in the level of tens of meters, and the problem of scale mismatch exists between the long-time sequence satellite data and centimeter-level details of unmanned aerial vehicle images, so that fine drawing of degradation areas is limited. Therefore, the demands of merging multisource remote sensing data and combining a machine learning method to perform high-precision vegetation classification and degradation recognition are increasingly highlighted. The existing method is easy to be influenced by redundant information when the feature dimension is higher, and the rejuvenation strategy is updated in a lack of differentiation aiming at different terrains and vegetation types, so that the establishment of rejuvenation measures and effect evaluation lack systematic support. Disclosure of Invention The invention aims to solve the technical problems that in the prior art, the arbor and shrub vegetation detection and analysis method and system based on the unmanned aerial vehicle are provided, and aims to solve the problems that arbor seedlings and shrubs, dense herbs and shrubs are mixed up due to vegetation classification only depending on canopy height, and degradation area identification precision is insufficient due to scale mismatch between long-time sequence remote sensing data and high-resolution unmanned aerial vehicle images, and meanwhile the defects that the rejuvenation measures lack terrain suitability and vegetation type pertinence are overcome. The technical scheme adopted for solving the technical problems is as follows: the invention provides an unmanned aerial vehicle-based arbor and shrub vegetation detection and analysis method, which comprises the following steps: acquiring an orthophoto and a digital surface model DSM of vegetation in a preset area acquired by an unmanned aerial vehicle, and generating a canopy height model CHM based on the digital terrain model DTM; Extracting a height information CHM value of the vegetation based on the CHM, and primarily classifying the vegetation into arbor, shrub and herb; Extracting texture features and morphological features of the vegetation from the orthographic images, correcting the preliminary classification result by combining the height information of the vegetation, and obtaining a refined vegetation classification result; the multi-temporal unmanned aerial vehicle image and the long-time sequence satellite remote sensing data are fused, a multi-dimensional feature vector is constructed, vegetation degradation level identification is carried out through a machine learning model, and a vegetation degradation graph with 1 meter spatial resolution is generated; formulating a differential updating rejuvenation scheme according to the vegetation degradation map, vegetation types and terrain conditions, wherein, The vegetation types include arbor, shrub, and herb. In some embodiments of the application, the extracting the height information CHM value of the vegetation based on the CHM and performing preliminary classification of the vegetation on trees, shrubs, herbs includes: Marking the vegetation with the CHM value greater than 4 meters as a candidate arbor; Marking the vegetation with the CHM value greater than 1 meter and less than 2 meters as candidate shrubs or arbor seedlings; Marking the vegetation with the CHM value less than 0.3 meters as a candidate herb; Further comprises: Acquiring a gradient value of each pixel in the orthographic image based on the elevation gradient of the digital terrain model DTM, and outputting a gradient grid atlas; extracting the section curvature q of the orthographic image according to the digital terrain model DTM, and marking a region with the gradient value larger than 15 degrees and the section curvature q smaller than 0 as a gully r