A non-cultivation habitat vegetation feature extraction method and system

Abstract

The application provides a non-cultivation habitat vegetation feature extraction method and system, in the extraction method, first, a machine learning classification method is used to classify the cultivated land system landscape in the region to be extracted; then, according to the result of the cultivated land system landscape classification, the vegetation growth feature extraction analysis in the non-cultivation habitat is selected, the vegetation coverage of the research area is estimated, and the vegetation growth situation is evaluated; finally, the best spatial scale of remote sensing data monitoring plant diversity is determined, and the texture features are analyzed, and the texture features are used as independent variables to establish a plant diversity prediction model to predict the plant diversity feature situation of the research area. The precision and efficiency of the non-cultivation habitat vegetation feature extraction are improved, the prediction result of the vegetation feature model of the non-cultivation habitat can provide a scientific basis for improving the existing non-cultivation habitat quality and promoting the farmland ecosystem restoration, and is an effective means for carrying out the effect evaluation of the comprehensive management of mountains, rivers, forests, farmlands, lakes and grasses.

Description

Technical Field

[0001] This invention relates to the field of high-resolution remote sensing image analysis and application technology, and in particular to a method and system for extracting vegetation features from non-cultivated habitats. Background Technology

[0002] Farmland and its surrounding ditches, wastelands, small woodlands (forest belts), shrub grasslands, field roads, and other non-cultivated habitats form a mosaic of arable land landscapes, serving as a primary source of resources for human survival. Maintaining a certain proportion of highly heterogeneous non-cultivated habitats within the arable land landscape system can provide the resources and environment necessary for the survival of most species within the landscape (such as food sources, species sources, refuges, and breeding grounds). The growth and community biodiversity of vegetation in non-cultivated habitats have a significant impact on the biodiversity of farmland landscapes. Accurately identifying and acquiring vegetation growth characteristics, community biodiversity, and their spatial location information is crucial. Traditional field survey methods can obtain highly accurate biodiversity data at the quadrat scale.

[0003] However, this method requires a lot of human and material resources. Therefore, remote sensing data has gradually become an important data source for biodiversity monitoring. Due to the limitation of remote sensing data resolution, it is still difficult to fully extract vegetation characteristics of non-cultivated habitats at a certain scale, making it difficult to meet the biodiversity mapping requirements of a certain area. Summary of the Invention

[0004] In view of this, the present invention provides a method and system for extracting vegetation features in non-cultivated habitats, so as to realize the prediction of vegetation features in non-cultivated habitats, and thus achieve the technological progress of extracting non-cultivated habitat features from the quadrat scale to a larger scale.

[0005] To achieve the above objectives, the present invention provides the following solution:

[0006] This invention provides a method for extracting vegetation features in non-cultivated habitats, the prediction method comprising the following steps:

[0007] High-resolution remote sensing images of the sample area are acquired, and feature extraction is performed based on the high-resolution images to obtain vegetation feature data and high-resolution image data; the vegetation feature data includes: spectral features, geometric features, vegetation texture features and vegetation remote sensing features;

[0008] A vegetation feature prediction model for non-cultivated habitats was established based on vegetation feature data and high-resolution image data for each non-cultivated habitat sample area. The prediction model includes a vegetation cover prediction model and a vegetation diversity prediction model. The vegetation cover prediction model is a model established using the pixel dichotomy model principle, with vegetation cover as the dependent variable and vegetation remote sensing features as the independent variable. The vegetation diversity prediction model is a model established using the linear regression principle, with vegetation diversity as the dependent variable and vegetation texture features as the independent variable.

[0009] The accuracy of the vegetation cover prediction model and the vegetation diversity prediction model is evaluated by combining low-altitude remote sensing image data from UAVs and field survey data, and the model with the highest accuracy is output.

[0010] The vegetation remote sensing feature data and vegetation texture feature data variables are respectively input into the vegetation cover prediction model and the vegetation diversity prediction model with the highest model accuracy to predict the spatial distribution characteristics of vegetation cover and vegetation diversity in the area to be predicted.

[0011] Furthermore, a prediction model for vegetation characteristics in non-cultivated habitats is established based on high-resolution image data and vegetation feature data, including:

[0012] A model with vegetation cover as the dependent variable and vegetation remote sensing features as the independent variable was established using the pixel dichotomy model principle; a model with vegetation diversity as the dependent variable and vegetation texture features as the independent variable was established using the linear regression principle.

[0013] By inputting the relevant parameters characterizing the remote sensing features and texture features of vegetation into the corresponding models, multiple models with different parameters are established.

[0014] Accordingly, the model accuracy of the vegetation cover prediction model and the vegetation diversity prediction model is evaluated by combining low-altitude remote sensing image data from UAVs and field survey data, including:

[0015] Based on UAV low-altitude remote sensing image data, field survey data, and models established with different parameters, the prediction error of each model is calculated.

[0016] Based on the magnitude of the prediction error, select explanatory variable parameters that can accurately explain vegetation cover and vegetation diversity, and output the model.

[0017] Furthermore, the prediction error of each model is calculated, including:

[0018] Using formula Calculate the mean square error of measured data and predicted results of vegetation characteristics in non-cultivated habitats. ;

[0019] Using formula Calculate the correlation coefficient between measured data and predicted results of vegetation characteristics in non-cultivated habitats. ;

[0020] In the formula, and The first The prediction results of image feature variables for each sample plot and the first Measured data of vegetation characteristics of non-cultivated habitat in each sample plot; The number of items represented; and They are The predicted results of each sample plot and the average value of vegetation characteristics in non-cultivated habitats.

[0021] Furthermore, determining the precision of the explanatory variable parameters includes:

[0022] If the judgment result is If the explanatory variable of the model is 'a', then the accuracy is higher.

[0023] If the judgment result is If b is used as the explanatory variable in the model, then the model has higher accuracy.

[0024] If the judgment result is If the explanatory variable of the model is 'a', then the accuracy is higher.

[0025] If the judgment result is If b is used as the explanatory variable in the model, then the model has higher accuracy.

[0026] In the formula, a and b represent image feature variables, R 2 and RMSE are the correlation coefficient and root mean square error between the vegetation characteristics estimated by the model and the measured values, respectively. R 2 The closer the value is to 1, the higher the precision; the smaller the RMSE value, the higher the accuracy.

[0027] Furthermore, feature extraction is performed based on high-resolution images, including: image segmentation scale optimization and feature extraction and optimization;

[0028] The preferred image segmentation scale includes:

[0029] A multi-scale segmentation algorithm is used to set scale parameters and heterogeneity parameters;

[0030] Using a multi-level scale parameter optimization tool, the data levels added to the project are automatically identified. The dataset is iteratively segmented by continuously increasing the scale parameters. The average homogeneity local variance of the segmented objects in all levels is calculated to obtain the segmentation results.

[0031] Compare the segmentation results to determine the optimal segmentation scale;

[0032] The feature extraction and optimization include:

[0033] Spectral, geometric, textural, and remote sensing features of the high-resolution image of the sample area are extracted.

[0034] Training samples are selected through visual interpretation and then filtered using a random forest model to reduce feature redundancy, resulting in filtered feature variables.

[0035] Furthermore, the filtered feature variables are obtained, including:

[0036] The classification feature value of each sample area is input as a variable into the random forest model. The importance of each variable is calculated according to the importance function in the random forest model. The importance of the variables is ranked according to the magnitude of the average decrease in node impurity.

[0037] Iteratively eliminate feature variables with small importance values, and select the best features based on the change in modeling accuracy of the remaining variables after elimination.

[0038] Furthermore, the scope of non-cultivated habitats is determined, including: using the optimization results of the importance function and constructing an optimal feature set from this dimension, and using a random forest classifier to complete the classification of landscape categories under this dimension;

[0039] By comparing the validation samples and classification results, a confusion matrix is ​​obtained. Global accuracy, mapping accuracy, user accuracy, and Kappa coefficient are calculated from the confusion matrix to evaluate the classification results of the images and compare the degree of agreement between the classification results and the actual results.

[0040] The landscape classification results under the random forest classifier were obtained, and the non-cultivated habitats were extracted.

[0041] Furthermore, high-resolution image data was obtained, including centimeter-level high-resolution visible light images acquired using Gaofen-2 imagery and low-altitude UAVs, from which vegetation cover and plant diversity were calculated.

[0042] The present invention also provides a system for extracting vegetation features in non-cultivated habitats, the system comprising:

[0043] The sample data acquisition module is used to acquire high-resolution remote sensing image data of the sample area, and perform feature extraction based on the high-resolution image to obtain vegetation feature data and high-resolution image data; the vegetation feature data includes: spectral features, geometric features, vegetation texture features and vegetation remote sensing features.

[0044] The model building module is used to build a vegetation feature prediction model for non-cultivated habitats based on the vegetation feature data and high-resolution image data of each non-cultivated habitat sample area obtained by the sample data acquisition module. The prediction model includes a vegetation cover prediction model and a vegetation diversity prediction model. The vegetation cover prediction model is a model built using the pixel dichotomy model principle, with vegetation cover as the dependent variable and vegetation remote sensing features as the independent variable. The vegetation diversity prediction model is a model built using the linear regression principle, with vegetation diversity as the dependent variable and vegetation texture features as the independent variable.

[0045] The model evaluation module is used to evaluate the model accuracy of the vegetation cover prediction model and the vegetation diversity prediction model established by the model building module by combining UAV low-altitude measured remote sensing image data and field survey data, and outputs the model with the highest model accuracy.

[0046] The prediction module is used to input vegetation remote sensing feature data and vegetation texture features into the vegetation cover prediction model and vegetation diversity prediction model with the highest model accuracy obtained by the model evaluation module, respectively, to predict the spatial distribution characteristics of vegetation cover and vegetation diversity in the area to be predicted.

[0047] This invention provides a method and system for extracting vegetation characteristics in non-cultivated habitats. The method involves first acquiring the vegetation characteristic image classification features and landscape classification results for each sample area; then, determining the extent of non-cultivated habitat based on the landscape classification results; and establishing a spatial distribution prediction model for non-cultivated habitat vegetation characteristics based on the vegetation characteristic image classification feature variables of each non-cultivated habitat sample area. Finally, vegetation characteristic data of the area to be predicted is collected, and the measured data and predicted data are compared to adjust the model and its accuracy. The prediction model is then determined to predict the spatial distribution of non-cultivated habitat vegetation characteristics in the area to be predicted. This method enables accurate identification and acquisition of vegetation growth characteristics, community biodiversity, and their spatial location information using image data. Simulating the spatial distribution of vegetation characteristics in non-cultivated habitats using readily available image data distinguishes it from traditional field survey methods, allowing for the acquisition of higher-precision biodiversity data at a certain scale, and providing a basis for optimizing biodiversity monitoring methods. Attached Figure Description

[0048] To more clearly illustrate the technical solutions in the embodiments of the present 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 the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0049] Figure 1A flowchart illustrating a method for extracting vegetation features in non-cultivated habitats provided in an embodiment of the present invention;

[0050] Figure 2 A schematic diagram illustrating the principle of a method for extracting vegetation features in non-cultivated habitats provided in an embodiment of the present invention;

[0051] Figure 3 This is a schematic diagram of the structural composition of a non-cultivated habitat vegetation feature extraction system provided in an embodiment of the present invention. Detailed Implementation

[0052] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0053] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0054] like Figure 1 and 2 As shown in the embodiment of the present invention, the method for extracting vegetation features of non-cultivated habitats includes the following steps:

[0055] Step 101: Obtain high-resolution remote sensing image data of the sample area, and extract features based on the high-resolution images to obtain vegetation feature data and high-resolution image data.

[0056] Feature extraction based on high-resolution images includes image segmentation scale optimization and feature extraction and optimization.

[0057] (1) Image segmentation scale optimization includes: using the multi-scale segmentation algorithm in eCognition software, setting scale parameters and heterogeneity parameters (including shape parameter weights and spectral parameter weights); using the multi-level scale parameter optimization tool (ESP2), automatically identifying the data layers added to the project, iteratively segmenting the dataset by continuously increasing the scale parameters, calculating the average homogeneity local variance of the segmented objects in all layers, and obtaining the segmentation results; comparing the segmentation results to determine the optimal segmentation scale.

[0058] (2) Vegetation feature data include: spectral features (gray mean, standard deviation, brightness and maximum difference, etc.), geometric features (area, perimeter, aspect ratio, density, rectangle fit, etc.), vegetation texture features (GLCM homogeneity, GLCM variance, GLCM heterogeneity, GLDV second moment, GLDV entropy, etc., based on gray co-occurrence matrix and gray difference vector, etc.), vegetation remote sensing features (normalized vegetation index, enhanced vegetation index, visible light vegetation index resistant to atmospheric influence, water index and building area index, etc.).

[0059] Spectral, geometric, textural, and remote sensing features of the high-resolution imagery of the region to be predicted are extracted. Training samples are selected through visual interpretation, and a C5.0 algorithm is constructed and filtered using a random forest model to reduce feature redundancy, resulting in filtered feature variables. Specifically, the classification feature value of each sample area is input as a variable into the random forest model. The importance of each variable is calculated according to the importance function in the random forest model, and the importance of the variables is ranked according to the magnitude of the average decrease in node impurity. Then, feature variables with smaller importance values ​​are iteratively eliminated, and the best features are selected based on the change in modeling accuracy of the remaining variables after elimination.

[0060] Furthermore, the scope of non-cultivated habitat vegetation is determined, including: classifying landscape categories based on feature optimization results, including: using the optimization results of the importance function and constructing an optimized feature set from this dimension, and using a random forest classifier to complete the classification of landscape categories under this dimension; obtaining a confusion matrix by comparing validation samples and classification results, calculating global accuracy, mapping accuracy, user accuracy, and Kappa coefficient from the confusion matrix to evaluate the image classification results, and comparing the degree of agreement between the classification results and the actual results; extracting non-cultivated habitats from the classification results to determine the model prediction range.

[0061] The object-oriented random forest algorithm can describe most of the features of land cover in the image. When combined with a suitable classifier, it can achieve good classification results.

[0062] The process of acquiring high-resolution image data includes: Gaofen-2 image data acquisition process, radiometric calibration, atmospheric correction, orthorectification, image mosaicking and cropping, mask extraction, acquisition threshold, vegetation cover estimation, etc.; and UAV image data acquisition process, including: UAV image data acquisition process, including: acquiring photos through flight path planning and aerial photography, aerial triangulation and image stitching, output of results, vegetation diversity estimation, etc.

[0063] Step 102: Establish a vegetation characteristic prediction model for each non-cultivated habitat sample area based on vegetation characteristic data and high-resolution image data. The prediction model includes: a vegetation cover prediction model and a vegetation diversity prediction model. The specific implementation includes the following steps:

[0064] S201. Establish a model with vegetation cover as the explained variable and vegetation remote sensing features as the explained variable using the pixel dichotomy model principle; establish a model with vegetation diversity as the explained variable and vegetation texture features as the explained variable using the linear regression principle.

[0065] S202. Input the relevant parameters representing the remote sensing features and texture features of vegetation into the corresponding models respectively;

[0066] The remote sensing index of each pixel is substituted into the pixel binary model; the texture features extracted at the optimal segmentation scale are substituted into the linear regression model.

[0067] In the pixel-based binary model, the threshold for the remote sensing index is determined, and the band calculation tool of ENVI software is used to build the model in conjunction with corresponding mask files for grassland, woodland, etc. The estimated vegetation cover is divided into four categories: high cover, medium-high cover, medium cover, and low cover. The parameters of the pixel-based binary model are adjusted according to the proportion of each vegetation cover category. Taking the remote sensing index NDVI as an example:

[0068] ;

[0069] In the formula: This refers to the local vegetation coverage. NDVI value for bare land or areas without vegetation cover; The NDVI value represents the area that is entirely covered by vegetation.

[0070] In the linear regression model, the pixel size for obtaining texture features is determined based on the semivariogram, and the model is built by combining the texture features of grassland and woodland types. The parameters of the linear regression model are adjusted according to whether the texture feature index is significantly correlated with vegetation diversity.

[0071] y = ax + b;

[0072] In the formula: y is the correlation index representing vegetation diversity; x is the vegetation texture feature; a is the independent variable coefficient; b is the constant term.

[0073] Predicting vegetation diversity in non-cultivated habitats using high-resolution texture features includes: establishing a spherical model to simulate high-resolution texture features and vegetation diversity indices using a semivariogram, and determining the optimal scale for monitoring vegetation diversity with high-resolution data; collecting actual data on vegetation features in non-cultivated habitats to obtain measured vegetation diversity; and establishing a linear regression model using high-resolution texture features as explanatory variables and measured vegetation diversity as the explained variable, and selecting the model with the best fit.

[0074] In vegetation diversity prediction models, spectral heterogeneity and textural heterogeneity need to be used to represent plant species diversity at an appropriate spatial scale.

[0075] The optimal scale for monitoring species diversity using GF-2 data was determined using a spherical model. The calculation formula for the spherical model is as follows:

[0076] ;

[0077] ;

[0078] ;

[0079] In the formula: For the variance of the gold nugget, For structural variance, The autocorrelation threshold, + For sill values, The lag distance interval. It represents the spatial heterogeneity of the random part, that is, the magnitude of the randomness of the variable.

[0080] Step 103: Combine UAV low-altitude remote sensing imagery and field surveys to evaluate the accuracy of the model.

[0081] S301. Evaluate the accuracy of the estimation;

[0082] In vegetation cover prediction models, the measured vegetation cover needs to be calculated based on centimeter-level high-resolution visible light images acquired by low-altitude UAVs, and the correlation coefficient R is used to determine the actual vegetation cover. 2 The estimation effectiveness was tested against the root mean square error (RMSE).

[0083] Based on the measured vegetation cover of non-cultivated habitat in each sample area and the prediction results for each sample area, the prediction error of the random forest model is calculated, specifically including:

[0084] Using formula Calculate the mean square error of measured data and predicted results of vegetation characteristics in non-cultivated habitats. ;

[0085] Using formula Calculate the correlation coefficient between measured data and predicted results of vegetation characteristics in non-cultivated habitats. ;

[0086] In the formula, and The first The prediction results of image feature variables for each sample plot and the first Measured data of vegetation characteristics of non-cultivated habitat in each sample plot; The number of items represented; and They are The predicted results of each sample plot and the average value of vegetation characteristics in non-cultivated habitats; It is the Pearson correlation coefficient between the prediction results and vegetation characteristics.

[0087] S302. Based on the magnitude of the prediction error, select the explanatory variable parameters that can more accurately explain vegetation cover and vegetation diversity, and output the model.

[0088] Determine whether the selection of explanatory variables for the model meets the accuracy requirements, specifically including:

[0089] If the judgment result is If the explanatory variable of the model is 'a', then the accuracy is higher.

[0090] If the judgment result is If b is used as the explanatory variable in the model, then the model has higher accuracy.

[0091] If the judgment result is If the explanatory variable of the model is 'a', then the accuracy is higher.

[0092] If the judgment result is If b is used as the explanatory variable in the model, then the model has higher accuracy.

[0093] In the formula, a and b represent image feature variables, and These represent the correlation coefficient and root mean square error between the vegetation characteristics estimated by the model and the measured values, respectively. 2 The closer the value is to 1, the higher the precision; the smaller the RMSE value, the higher the accuracy.

[0094] Step 104: Input the high-resolution image data and the extracted vegetation feature data into the vegetation feature prediction model of the area to be predicted, and predict the spatial distribution characteristics of vegetation coverage and vegetation diversity in the area to be predicted.

[0095] By utilizing the principles of pixel binary model and linear regression, a spatial distribution map of vegetation features in non-cultivated habitats is output, thus completing the extraction of non-cultivated habitat features.

[0096] like Figure 3 As shown, the present invention also provides a system for predicting vegetation characteristics in non-cultivated habitats, the system comprising:

[0097] The sample data acquisition module 301 is used to acquire high-resolution remote sensing image data of the sample area, and perform feature extraction based on the high-resolution image to obtain vegetation feature data and high-resolution image data.

[0098] The sample data acquisition module 301 specifically includes: a segmentation scale optimization submodule, which uses the multi-scale segmentation algorithm in eCognition software to set scale parameters and determines the optimal segmentation scale through the multi-level scale parameter optimization tool (ESP2); and a landscape classification submodule, which obtains a confusion matrix by comparing the validation samples and classification results, and calculates global accuracy, mapping accuracy, user accuracy, and Kappa coefficient to evaluate the image classification results and obtain landscape classification results.

[0099] The model building module 302 is used to build a vegetation cover prediction model and a vegetation diversity prediction model based on the vegetation characteristic data and high-resolution image data of each non-cultivated habitat sample area obtained by the sample data acquisition module 301. The vegetation cover prediction model is a model built using the pixel dichotomy model principle, with vegetation cover as the dependent variable and vegetation remote sensing features as the independent variable. The vegetation diversity prediction model is a model built using the linear regression principle, with vegetation diversity as the dependent variable and vegetation texture features as the independent variable.

[0100] The model building module 302 specifically includes: a pixel-based binary model building submodule, which uses the pixel-based binary model principle to build an initial prediction model with vegetation cover as the explained variable and remote sensing data as the explanatory variable; a pixel-based binary model parameter adjustment submodule, which is used to adjust the model parameters during the model building process; and an accuracy estimation and evaluation submodule, which is used to determine the explanatory variables included in the optimal prediction model. If the accuracy of the accuracy estimation and evaluation is optimal, the prediction model is output to obtain a spatial distribution prediction model of vegetation characteristics in non-cultivated habitats.

[0101] The model evaluation module 303 is used to evaluate the model accuracy of the vegetation cover prediction model and the vegetation diversity prediction model established by the model building module 302 by combining the low-altitude remote sensing image data of UAV and the field survey data, and output the model with the highest model accuracy.

[0102] The prediction module 304 is used to input vegetation remote sensing feature data and vegetation texture features into the vegetation cover prediction model and vegetation diversity prediction model with the highest model accuracy obtained by the model evaluation module 303, respectively, to predict the spatial distribution characteristics of vegetation cover and vegetation diversity in the area to be predicted.

[0103] As for the non-cultivated habitat vegetation feature prediction system of this invention, since it corresponds to the non-cultivated habitat vegetation feature prediction method in the above embodiment, the description is relatively simple. For related similarities, please refer to the description of the non-cultivated habitat vegetation feature prediction method in the above embodiment, which will not be described in detail here.

[0104] According to specific embodiments provided by the present invention, the technical solution disclosed in the present invention has the following technical effects:

[0105] The modeling process of this invention does not require a pre-reserved validation set but instead employs cross-validation. By simulating the spatial distribution of vegetation characteristics in non-cultivated habitats using readily available image data, it can differentiate itself from traditional field survey methods and obtain high-precision biodiversity data at a certain scale, providing a basis for optimizing biodiversity monitoring methods.

[0106] Vegetation growth and community biodiversity in non-cultivated habitats significantly impact biodiversity in farmland landscapes. Accurate identification and acquisition of vegetation growth characteristics, community biodiversity, and their spatial location information are crucial. Traditional field survey methods can obtain high-precision biodiversity data at the quadrat scale.

[0107] However, this method requires a lot of human and material resources. Therefore, remote sensing data has gradually become an important data source for biodiversity monitoring. Due to the limitation of remote sensing data resolution, it is still difficult to fully extract vegetation characteristics of non-cultivated habitats at a certain scale, making it difficult to meet the biodiversity mapping requirements of a certain area.

[0108] 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 or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for extracting vegetation features from non-cultivated habitats, characterized in that, Includes the following steps: High-resolution remote sensing images of the sample area are acquired, and feature extraction is performed based on the high-resolution images to obtain vegetation feature data and high-resolution image data. The vegetation feature data includes: spectral features, geometric features, vegetation texture features, and vegetation remote sensing features. Feature extraction based on the high-resolution images includes: image segmentation scale selection and feature extraction and selection. The image segmentation scale selection includes: using a multi-scale segmentation algorithm to set scale parameters and heterogeneity parameters; using a multi-level scale parameter selection tool to automatically identify the data layers added to the project, iteratively segmenting the dataset by continuously increasing the scale parameters, calculating the average homogeneity local variance of the segmented objects in all layers, and obtaining the segmentation result; comparing the segmentation results... The optimal segmentation scale is determined. Feature extraction and selection include: extracting spectral, geometric, textural, and remote sensing features from the high-resolution image of the sample area; selecting training samples through visual interpretation and filtering them in a random forest model to reduce feature redundancy, resulting in filtered feature variables; obtaining the filtered feature variables includes: inputting the classification feature value of each sample area as a variable into the random forest model; calculating the importance of each variable according to the importance function in the random forest model; ranking the importance of variables according to the magnitude of the average decrease in node impurity; iteratively eliminating feature variables with importance values ​​less than a preset value; and selecting features based on the change in modeling accuracy of the remaining variables after elimination. A vegetation feature prediction model for non-cultivated habitats was established based on vegetation feature data and high-resolution image data for each non-cultivated habitat sample area. The prediction model includes a vegetation cover prediction model and a vegetation diversity prediction model. The vegetation cover prediction model is a model established using the pixel dichotomy model principle, with vegetation cover as the dependent variable and vegetation remote sensing features as the independent variable. The vegetation diversity prediction model is a model established using the linear regression principle, with vegetation diversity as the dependent variable and vegetation texture features as the independent variable. The accuracy of the vegetation cover prediction model and the vegetation diversity prediction model is evaluated by combining low-altitude remote sensing image data from UAVs and field survey data, and the model with the highest accuracy is output. The vegetation remote sensing feature data and vegetation texture feature data variables are respectively input into the vegetation cover prediction model and the vegetation diversity prediction model with the highest model accuracy to predict the spatial distribution characteristics of vegetation cover and vegetation diversity in the area to be predicted.

2. The method for extracting vegetation features in non-cultivated habitats according to claim 1, characterized in that, A vegetation feature prediction model for non-cultivated habitats was established based on high-resolution image data and vegetation feature data, including: A model with vegetation cover as the dependent variable and vegetation remote sensing features as the independent variable was established using the pixel dichotomy model principle; a model with vegetation diversity as the dependent variable and vegetation texture features as the independent variable was established using the linear regression principle. By inputting the relevant parameters characterizing the remote sensing features and texture features of vegetation into the corresponding models, multiple models with different parameters are established. Accordingly, the model accuracy of the vegetation cover prediction model and the vegetation diversity prediction model is evaluated by combining low-altitude remote sensing image data from UAVs and field survey data, including: Based on UAV low-altitude remote sensing image data, field survey data, and models established with different parameters, the prediction error of each model is calculated. Based on the magnitude of the prediction error, select explanatory variable parameters that can accurately explain vegetation cover and vegetation diversity, and output the model.

3. The method for extracting vegetation features in non-cultivated habitats according to claim 2, characterized in that, Calculate the prediction error of each model, including: Using formula Calculate the mean square error of measured data and predicted results of vegetation characteristics in non-cultivated habitats. ; Using formula Calculate the correlation coefficient between measured data and predicted results of vegetation characteristics in non-cultivated habitats. ; In the formula, and The first The prediction results of image feature variables for each sample plot and the first Measured data of vegetation characteristics of non-cultivated habitat in each sample plot; The number of items represented; and They are The predicted results of each sample plot and the average value of vegetation characteristics in non-cultivated habitats.

4. The method for extracting vegetation features in non-cultivated habitats according to claim 3, characterized in that, Determining the precision of explanatory variable parameters includes: If the judgment result is If the explanatory variable of the model is 'a', then the accuracy is higher. If the judgment result is If b is the explanatory variable of the model, then the accuracy is higher. If the judgment result is If the explanatory variable of the model is 'a', then the accuracy is higher. If the judgment result is If b is the explanatory variable of the model, then the accuracy is higher. In the formula, a and b represent image feature variables, R 2 and RMSE are the correlation coefficient and root mean square error between the vegetation characteristics estimated by the model and the measured values, respectively. R 2 The closer the value is to 1, the higher the precision; the smaller the RMSE value, the higher the accuracy.

5. The method for extracting vegetation features in non-cultivated habitats according to claim 1, characterized in that, Determining the extent of non-cultivated habitats includes: using the selection results of the importance function and constructing a feature set from the importance dimension, and using a random forest classifier to classify the landscape categories under this dimension; By comparing the validation samples and classification results, a confusion matrix is ​​obtained. Global accuracy, mapping accuracy, user accuracy, and Kappa coefficient are calculated from the confusion matrix to evaluate the classification results of the images and compare the degree of agreement between the classification results and the actual results. The landscape classification results under the random forest classifier were obtained, and the non-cultivated habitats were extracted.

6. The method for extracting vegetation features in non-cultivated habitats according to claim 1, characterized in that, High-resolution image data was obtained, including centimeter-level high-resolution visible light images acquired using Gaofen-2 imagery and low-altitude UAVs, and the resulting vegetation cover and plant diversity were calculated.

7. A system for extracting vegetation features from non-cultivated habitats, characterized in that, The system includes: The sample data acquisition module is used to acquire high-resolution remote sensing image data of the sample area, and perform feature extraction based on the high-resolution images to obtain vegetation feature data and high-resolution image data. The vegetation feature data includes: spectral features, geometric features, vegetation texture features, and vegetation remote sensing features. Feature extraction based on high-resolution images includes: image segmentation scale selection and feature extraction and selection. The image segmentation scale selection includes: using a multi-scale segmentation algorithm to set scale parameters and heterogeneity parameters; using a multi-level scale parameter selection tool to automatically identify the data layers added to the project, iteratively segmenting the dataset by continuously increasing the scale parameters, and calculating the average homogeneity local square of the segmented objects in all layers. The segmentation results are obtained by comparing the segmentation results and determining the optimal segmentation scale. The feature extraction and selection includes: extracting spectral features, geometric features, texture features, and remote sensing features of the high-resolution image of the sample area; selecting training samples through visual interpretation and filtering them in a random forest model to reduce feature redundancy, and obtaining the filtered feature variables, including: inputting the classification feature value of each sample area as a variable into the random forest model, calculating the importance of each variable according to the importance function in the random forest model, and ranking the importance of the variables according to the magnitude of the average decrease in node impurity; iteratively eliminating feature variables with importance values ​​less than a preset value, and selecting features based on the change in modeling accuracy of the remaining variables after elimination. The model building module is used to build a vegetation feature prediction model for non-cultivated habitats based on the vegetation feature data and high-resolution image data of each non-cultivated habitat sample area obtained by the sample data acquisition module. The prediction model includes a vegetation cover prediction model and a vegetation diversity prediction model. The vegetation cover prediction model is a model built using the pixel dichotomy model principle, with vegetation cover as the dependent variable and vegetation remote sensing features as the independent variable. The vegetation diversity prediction model is a model built using the linear regression principle, with vegetation diversity as the dependent variable and vegetation texture features as the independent variable. The model evaluation module is used to evaluate the model accuracy of the vegetation cover prediction model and the vegetation diversity prediction model established by the model building module by combining UAV low-altitude measured remote sensing image data and field survey data, and outputs the model with the highest model accuracy. The prediction module is used to input vegetation remote sensing feature data and vegetation texture features into the vegetation cover prediction model and vegetation diversity prediction model with the highest model accuracy obtained by the model evaluation module, respectively, to predict the spatial distribution characteristics of vegetation cover and vegetation diversity in the area to be predicted.

Images