Shrub biomass evaluation and prediction method based on remote sensing data

Abstract

The invention discloses a shrub biomass evaluation and prediction method based on remote sensing data, and relates to the technical field of shrub biomass prediction. Environment information of the target area is obtained, and a shrub environment similarity coefficient of the target area is analyzed based on the environment information of the target area; selecting a model for evaluating and predicting the shrub biomass in the target area according to the shrub environment similarity coefficient as a target model; evaluating and predicting the biomass of the target area according to the preprocessed remote sensing data of the target area and the target model; therefore, when the shrub biomass evaluation prediction is carried out on the new area, whether the existing shrub biomass evaluation prediction model is selected to carry out shrub biomass evaluation prediction on the new area or not can be determined according to the actual condition, so that the difference between the shrub biomass evaluation prediction result of the new area and the actual shrub biomass evaluation prediction result is small; and the influence on ecological monitoring and environmental protection effects of the shrubs is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of shrub biomass prediction, and particularly to a method for evaluating and predicting shrub biomass based on remote sensing data. Background Art

[0002] The method for evaluating and predicting shrub biomass based on remote sensing data is to use remote sensing technology to obtain image data of ground shrubs, such as images taken by satellites or drones, and combine image processing and data analysis models to estimate the biomass of shrub areas. These methods analyze the spectral characteristics, structural characteristics, and growth conditions of vegetation, and then establish mathematical models to predict biomass, which are widely used in ecological monitoring, environmental protection, agricultural management, and other fields; In the actual process of evaluating and predicting shrub biomass based on remote sensing data, when it is necessary to evaluate and predict the shrub biomass of a new area, in order to save computing resources, or select a suitable model from existing shrub biomass evaluation and prediction models to evaluate and predict the shrub biomass of the new area, it can significantly save computing resources and time costs. By using the already trained model, the complex calculations and data processing from scratch can be avoided, and the prediction efficiency can be improved. At the same time, selecting a suitable model can also improve the prediction accuracy, especially in similar environments or conditions, and reduce the prediction error caused by regional differences. This method not only improves the resource utilization rate but also can provide reliable biomass evaluation results in a short time, facilitating quick decision-making and implementation; However, when evaluating and predicting the shrub biomass of a new area, the existing shrub biomass evaluation and prediction models may not necessarily be applicable to the new area. If an existing shrub biomass evaluation and prediction model is blindly selected to evaluate and predict the shrub biomass of the new area, it may lead to a large difference between the predicted result of the shrub biomass in the new area and the actual predicted result of the shrub biomass, and may even affect the effects of shrub ecological monitoring and environmental protection. Summary of the Invention

[0003] The object of the present invention is to solve the above-mentioned problems and provide a method for evaluating and predicting shrub biomass based on remote sensing data.

[0004] The present invention provides a method for evaluating and predicting shrub biomass based on remote sensing data, and the method includes: Obtain the remote sensing data of the target area, and preprocess the remote sensing data to obtain the preprocessed remote sensing data of the target area; Obtain the environmental information of the target area, and analyze the shrub environment similarity coefficient of the target area based on the environmental information of the target area; Select a model for evaluating and predicting the shrub biomass of the target area according to the shrub environment similarity coefficient as the target model; Evaluate and predict the biomass of the target area based on the preprocessed remote sensing data of the target area and the target model.

[0005] Optionally, the steps of obtaining the remote sensing data of the target area and preprocessing the remote sensing data are as follows: The remote sensing data includes optical remote sensing data, radar remote sensing data, lidar data, and hyperspectral remote sensing data; Perform spatial registration on the optical remote sensing data, radar remote sensing data, lidar data, and hyperspectral remote sensing data, and use an automatic matching and false matching elimination algorithm based on feature points to ensure the spatial consistency of multi-source data; and the automatic matching uses the SIFT feature matching combined with the RANSAC algorithm to eliminate false matches; For optical remote sensing data, combine the atmospheric correction model with terrain correction to eliminate the influence of atmospheric scattering and terrain shadows; For radar remote sensing data, through polarization calibration and adaptive filtering processing, reduce the speckle noise of radar data; the adaptive filtering uses Lee filtering combined with polarization decomposition method; For lidar data, generate a high-precision canopy height model based on the irregular triangular network (TIN) interpolation method, and retain the three-dimensional structural characteristics of shrubs; For hyperspectral remote sensing data, based on band information entropy and separability analysis, select the characteristic bands sensitive to shrub biomass; and the band selection includes red edge bands and shortwave infrared bands; Adopt a method combining principal component analysis and wavelet transform to fuse the spectral and texture information of optical and radar data; Use NDVI and NDWI to distinguish vegetation and non-vegetation areas; separate shrubs and low-growing trees based on the radar cross-polarization ratio; eliminate low-confidence areas through lidar point cloud density analysis.

[0006] Optionally, the steps of obtaining the environmental information of the target area and analyzing the shrub environmental similarity coefficient of the target area based on the environmental information of the target area are as follows: Obtain the radar remote sensing image of the target area, divide the radar remote sensing image into several grids, and obtain the altitude corresponding to each grid; Calculate the average altitude of the target area based on the altitude corresponding to each grid, and calculate the altitude standard deviation of the target area based on the average altitude; Obtain the areas where the shrub biomass evaluated and predicted by each existing shrub biomass evaluation and prediction model is located, denoted as the first areas, obtain the average altitude and altitude standard deviation of each first area, and calculate the absolute value of the difference between the average altitude and the average altitude of the target area, and the absolute value of the difference between the altitude standard deviation and the altitude standard deviation of the target area respectively; The absolute value of the average elevation and the absolute value of the elevation standard deviation are weighted and summed to obtain the elevation similarity coefficient between the target area and each first area; Calculate the shrub environment similarity coefficient between the target area and each first area according to the elevation similarity coefficient.

[0007] Optionally, the steps for calculating the shrub environment similarity coefficient of the target area according to the elevation similarity coefficient are as follows: Obtain the rainfall in the target area in the past year to obtain a rainfall sequence based on time order as the first rainfall sequence; Denote the area where the shrub biomass evaluated and predicted by each existing shrub biomass evaluation and prediction model is located as the first area, obtain the rainfall in each first area in the past year to obtain a rainfall sequence based on time order as the second rainfall sequence; Calculate the similarity value between the first rainfall sequence and the second rainfall sequence through the dynamic time warping algorithm to obtain the rainfall similarity coefficient; Calculate the shrub environment similarity coefficient between the target area and each first area according to the elevation similarity coefficient and the rainfall similarity coefficient.

[0008] Optionally, the steps for calculating the shrub environment similarity coefficient between the target area and each first area according to the elevation similarity coefficient and the rainfall similarity coefficient are as follows: Convert the elevation similarity coefficient and the rainfall similarity coefficient into a comprehensive feature vector, use the comprehensive feature vector as the input of a machine learning model, use the prediction of each shrub environment similarity coefficient label by the machine learning model as the prediction target, and use minimizing the sum of the prediction errors of all shrub environment similarity coefficient labels as the training target. Train the machine learning model until the sum of the prediction errors reaches convergence and then stop the model training. Determine the shrub environment similarity coefficient according to the model output result, where the machine learning model is a polynomial regression model.

[0009] Optionally, the steps for selecting a model for evaluating and predicting the shrub biomass of the target area according to the shrub environment similarity coefficient as the target model are as follows: Denote the area where the shrub biomass evaluated and predicted by each existing shrub biomass evaluation and prediction model is located as the first area. Compare the shrub environment similarity coefficient between the target area and each first area with a preset shrub environment similarity coefficient threshold. If the shrub environment similarity coefficient is not less than the preset shrub environment similarity coefficient threshold, denote the corresponding first area as the available area, denote the shrub biomass evaluation and prediction model of the available area as the available model, select the model with the largest shrub environment similarity coefficient in the available models as the target model, and input the remote sensing data of the target area into the target model to evaluate and predict the shrub biomass of the target area; If all the shrub environment similarity coefficients are less than the preset shrub environment similarity coefficient threshold, the target model is retrained based on the remote sensing data of the target area to evaluate and predict the shrub biomass of the target area.

[0010] Optionally, if all the shrub environment similarity coefficients are less than the preset shrub environment similarity coefficient threshold, the steps to retrain the target model based on the remote sensing data of the target area to evaluate and predict the shrub biomass of the target area are as follows: Use the remotely sensed data of the target area that has been preprocessed and feature-extracted as input to prepare for model training; Select a machine learning algorithm and input the remotely sensed data of the target area into the model for training. The training process includes: Use the correspondence between the remotely sensed data of the target area and the known biomass for training; Adjust the hyperparameters of the model to optimize the model performance; Evaluate the accuracy of the model through cross-validation and adjust the model parameters according to the evaluation results to ensure the optimal performance of the model in the target area; Use the trained target model to predict the shrub biomass of the remotely sensed data of the target area and output the estimated value of the shrub biomass of the target area.

[0011] Advantages of the present invention: The present invention proposes a method for evaluating and predicting shrub biomass based on remote sensing data. By obtaining the remote sensing data of the target area and preprocessing the remote sensing data, the preprocessed remote sensing data of the target area is obtained; obtaining the environmental information of the target area and analyzing the shrub environment similarity coefficient of the target area based on the environmental information of the target area; selecting a model for evaluating and predicting the shrub biomass of the target area according to the shrub environment similarity coefficient as the target model; evaluating and predicting the biomass of the target area according to the preprocessed remote sensing data of the target area and the target model; in this way, when evaluating and predicting the shrub biomass of a new area, it is possible to determine whether to select an existing model for evaluating and predicting the shrub biomass of the new area according to the actual situation, ensuring that the difference between the evaluated and predicted results of the shrub biomass of the new area and the actual evaluated and predicted results of the shrub biomass is small, and reducing the impact on the ecological monitoring and environmental protection effects of the shrubbery. Description of the Drawings

[0012] The present invention will be further described below with reference to the drawings.

[0013] Figure 1 It is a flowchart of a method for evaluating and predicting shrub biomass based on remote sensing data. Detailed Embodiments

[0014] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0015] All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0016] The embodiment of the present invention provides a method for evaluating and predicting shrub biomass based on remote sensing data. Refer to Figure 1 , Figure 1 which is a flowchart of a method for evaluating and predicting shrub biomass based on remote sensing data provided by the embodiment of the present invention. The method includes the following steps: Obtain the remote sensing data of the target area, and preprocess the remote sensing data to obtain the preprocessed remote sensing data of the target area; Obtain the environmental information of the target area, and analyze the shrub environmental similarity coefficient of the target area based on the environmental information of the target area; Select a model for evaluating and predicting the shrub biomass of the target area according to the shrub environmental similarity coefficient as the target model; Evaluate and predict the biomass of the target area according to the preprocessed remote sensing data of the target area and the target model.

[0017] Based on the method for evaluating and predicting shrub biomass based on remote sensing data provided by the embodiment of the present invention, when evaluating and predicting the shrub biomass of a new area, it can be determined according to the actual situation whether to select an existing model for evaluating and predicting the shrub biomass of the new area, ensuring that the difference between the evaluation and prediction results of the shrub biomass of the new area and the actual evaluation and prediction results of the shrub biomass is small, and reducing the impact on the ecological monitoring and environmental protection effects of the shrubbery.

[0018] In one embodiment, the steps of obtaining the remote sensing data of the target area and preprocessing the remote sensing data are as follows: The remote sensing data includes optical remote sensing data, radar remote sensing data, lidar data, and hyperspectral remote sensing data; Perform spatial registration on the optical remote sensing data, radar remote sensing data, lidar data, and hyperspectral remote sensing data, and use an automatic matching and false matching elimination algorithm based on feature points to ensure the spatial consistency of multi-source data; and the automatic matching uses the SIFT feature matching combined with the RANSAC algorithm to eliminate false matches; For the optical remote sensing data, combine the atmospheric correction model with the terrain correction to eliminate the influence of atmospheric scattering and terrain shadows; For radar remote sensing data, through polarization calibration and adaptive filtering processing, the speckle noise of radar data is reduced; the adaptive filtering adopts the Lee filtering combined with the polarization decomposition method.

[0019] For lidar data, a high-precision canopy height model is generated based on the triangulated irregular network (TIN) interpolation method, and the three-dimensional structural characteristics of shrubs are retained; For hyperspectral remote sensing data, based on the band information entropy and separability analysis, the characteristic bands sensitive to shrub biomass are selected; and the band selection includes the red-edge band and the short-wave infrared band; A method combining principal component analysis and wavelet transform is adopted to fuse the spectral and texture information of optical and radar data; The NDVI and NDWI are used to distinguish vegetation and non-vegetation areas; shrubs and low-growing trees are separated based on the radar cross-polarization ratio; the low-confidence areas are removed through lidar point cloud density analysis.

[0020] It should be noted that when obtaining and preprocessing the remote sensing data of the target area, multiple data sources and processing technologies are adopted, aiming to improve the data quality, ensure spatial consistency, reduce noise and extract the characteristic information closely related to shrub biomass. First of all, the remote sensing data includes different types of data such as optical remote sensing data, radar remote sensing data, lidar data and hyperspectral remote sensing data, and each data source provides information in different dimensions. It is crucial to perform spatial registration on these multi-source data, especially when the data sources and resolutions are inconsistent. An automatic matching and false matching elimination algorithm based on feature points is adopted, and the SIFT (scale-invariant feature transform) feature matching combined with the RANSAC (random sample consensus) algorithm can effectively identify and eliminate false matches, ensuring the spatial consistency between data, so as to provide a reliable data basis for subsequent analysis.

[0021] For optical remote sensing data, due to factors such as atmospheric scattering and terrain shadows that will affect the image quality, it is necessary to combine the atmospheric correction model and terrain correction technology to eliminate these adverse effects and ensure that the optical image truly reflects the ground features. The influence of atmospheric scattering is eliminated through atmospheric correction, and terrain correction can effectively reduce the shadows or reflection changes caused by the terrain, thereby improving the accuracy of the data.

[0022] Radar remote sensing data can provide effective information about the structure of ground objects due to its strong penetration ability. However, the speckle noise of radar data may affect the analysis results, so it is necessary to reduce the noise through polarization calibration and adaptive filtering. The combination of Lee filtering and polarization decomposition method can effectively smooth the noise and retain the ground features, enhancing the usability of the data.

[0023] The advantage of lidar data lies in its ability to provide highly accurate three-dimensional information, which is particularly suitable for extracting the canopy height characteristics of shrubs or trees. Through the triangulated irregular network (TIN) interpolation method, a high-precision canopy height model can be generated, thereby retaining the three-dimensional structural characteristics of shrubs and providing strong support for biomass assessment.

[0024] Hyperspectral remote sensing data, on the other hand, provides rich spectral information and can obtain the reflection characteristics of vegetation from multiple bands. By analyzing the band information entropy and separability, characteristic bands sensitive to shrub biomass are selected, especially the red edge band and the short-wave infrared band. These bands are highly sensitive to changes in vegetation growth, health status, and water content, and are helpful for accurately reflecting the changes in shrub biomass.

[0025] In addition, by combining principal component analysis (PCA) and wavelet transform, optical and radar data can be fused to extract spectral and texture information, thus more comprehensively reflecting the ground characteristics of the target area. To distinguish vegetation and non-vegetation areas, using NDVI (Normalized Difference Vegetation Index) and NDWI (Normalized Difference Water Index) helps to identify different types of vegetation-covered areas. Through the radar cross-polarization ratio, the shrub and low-tree areas can be further separated, while lidar point cloud density analysis can eliminate low-confidence areas and improve the reliability and accuracy of the data.

[0026] In one implementation, as described above, preprocessing remote sensing data not only helps to eliminate noise and errors in the data, but also optimizes the fusion of different data sources, making the final biomass prediction results more accurate and reliable. Through these refined processes, the spatial accuracy, spectral characteristics, and three-dimensional structural characteristics of remote sensing data can be effectively improved, making the prediction of shrub biomass assessment more in line with the actual situation, thereby enhancing the decision-making support capabilities for ecological monitoring, environmental protection, and resource management.

[0027] In one embodiment, the steps of obtaining the environmental information of the target area and analyzing the shrub environmental similarity coefficient of the target area based on the environmental information of the target area are as follows: Obtain the radar remote sensing image of the target area, divide the radar remote sensing image into several grids, and obtain the altitude corresponding to each grid; Calculate the average altitude of the target area according to the altitude corresponding to each grid, and calculate the altitude standard deviation of the target area based on the average altitude; Obtain the areas where the shrub biomass evaluated and predicted by each existing shrub biomass assessment prediction model are located, denoted as the first areas. Obtain the average altitude and altitude standard deviation of each first area, and calculate the absolute value of the difference between the average altitude and the average altitude of the target area, and the absolute value of the altitude standard deviation between the altitude standard deviation and the altitude standard deviation of the target area respectively; The absolute value of the average altitude and the absolute value of the altitude standard deviation are weighted and summed to obtain the altitude similarity coefficient between the target area and each first area; Calculate the shrub environment similarity coefficient between the target area and each first area according to the altitude similarity coefficient.

[0028] It should be noted that in the above process of calculating the altitude similarity coefficient, the data acquisition method mainly relies on high-resolution remote sensing data and Geographic Information System (GIS) technology. First, by obtaining the radar remote sensing image of the target area, the area is divided into grids using the spatial information of the remote sensing image, and the altitude data of each grid is extracted. These altitude data are usually extracted through Digital Elevation Model (DEM) or ground measurement data. Then, by calculating the average altitude and standard deviation of the target area and other known areas, the altitude characteristic differences of each area can be objectively reflected. Combining these high-precision geographical data can ensure that the calculated altitude similarity coefficient can more accurately reflect the terrain differences of each area, thus providing a reliable basis for the subsequent shrub environment similarity analysis.

[0029] It should be noted that the steps of weighting and summing the absolute value of the average altitude and the absolute value of the altitude standard deviation to obtain the altitude similarity coefficient between the target area and each first area are as follows: In the formula, is the altitude similarity coefficient, and respectively represent the absolute value of the average altitude and the absolute value of the altitude standard deviation, is and 's proportionality coefficient, and are both greater than 0.

[0030] It should be noted that the altitude similarity coefficient quantifies the similarity in altitude between the target area and the first area by calculating the altitude difference therebetween. Specifically, it takes into account two main factors: 1) the difference in the average altitude between the target area and the first area; and 2) the standard difference in altitude between the target area and the first area. By weighted summation of these two differences, a comprehensive altitude similarity coefficient is obtained. The larger the value of this coefficient, the more similar the altitude characteristics of the target area and the first area. Conversely, the smaller the coefficient, the greater the altitude difference between the two. A larger altitude similarity coefficient indicates a high similarity in altitude characteristics between the target area and the first area, which usually means that their environmental factors such as climate, soil type, vegetation type, etc. may also be similar. Since the growth of shrubs is closely related to altitude, especially at different altitudes, factors such as temperature, precipitation, and soil composition will affect the growth state and biomass of shrubs. For example, in mountainous areas at higher altitudes, the temperature is lower, and the vegetation may mainly consist of cold-tolerant shrubs, while in low-altitude areas, the vegetation may mainly consist of tropical or subtropical shrubs. If the altitude similarity between the target area and the first area is high, then the environmental conditions and vegetation types in these two areas are also likely to be similar, so that the prediction result of the shrub biomass assessment model in the first area for the target area is more accurate. For example, assume that the target area is located in a mid-altitude mountainous area, and the climate and soil conditions in this area are similar to those in a known area (the first area). If the altitude similarity coefficient between these two areas is very high, indicating a great similarity in climate and ecological conditions between them, then when predicting the target area using the biomass assessment model in the first area, the prediction result will be closer to the actual result. This similarity enables the model to accurately reflect the shrub biomass in the target area, thus providing effective information in ecological monitoring and environmental protection. If the altitude similarity is low, it may lead to a large prediction error because the environmental assumptions on which the model is based are quite different from the actual environment in the target area, thereby affecting the accurate assessment of biomass and even potentially leading to incorrect management decisions and affecting the effect of ecological protection.

[0031] In one embodiment, the steps for calculating the shrub environmental similarity coefficient of the target area according to the altitude similarity coefficient are as follows: Obtain the rainfall in the target area in the past year to obtain a rainfall sequence based on chronological order as the first rainfall sequence; Denote the area where the shrub biomass evaluated and predicted by each existing shrub biomass assessment prediction model is located as the first area, and obtain the rainfall in each first area in the past year to obtain a rainfall sequence based on chronological order as the second rainfall sequence; Calculate the similarity value between the first rainfall sequence and the second rainfall sequence through the dynamic time warping algorithm to obtain the rainfall similarity coefficient; Calculate the shrub environment similarity coefficient between the target area and each first area according to the altitude similarity coefficient and the rainfall similarity coefficient.

[0032] It should be noted that in the process of calculating the rainfall similarity coefficient, the data acquisition method mainly relies on long-term meteorological monitoring data and remote sensing technology. First, obtain the rainfall data of the target area in the past year, usually through meteorological stations, satellite remote sensing data or meteorological models for monitoring and prediction. The real-time rainfall data provided by meteorological stations can directly reflect the precipitation changes in the target area, while remote sensing data can estimate the precipitation by analyzing the cloud reflection characteristics. Secondly, the same method is applied to obtain the rainfall data of the first area in the past year to ensure that the precipitation sequences of the two are based on the same time range and time order. Finally, using the dynamic time warping (DTW) algorithm, the two precipitation sequences can be matched in time to calculate the similarity between them, thus obtaining the rainfall similarity coefficient. Through this method, accurate precipitation data can be obtained and its temporal consistency can be ensured, providing strong data support for subsequent shrub environment similarity analysis.

[0033] It should be noted that the rainfall similarity coefficient refers to the similarity between the target area and the first area calculated by using the dynamic time warping (DTW) algorithm by comparing the rainfall sequences in the past year. The dynamic time warping algorithm can evaluate the similarity by comparing two time series, especially when there is a desynchronization on the time axis. The larger the rainfall similarity coefficient, the more similar the precipitation patterns, seasonal variations and rainfall fluctuations between the target area and the first area, which also means that the hydrological environment and ecological conditions of these two areas are more similar in terms of precipitation.

[0034] Precipitation is an important environmental factor affecting the growth of shrubs, which is directly related to the water supply and growth cycle of vegetation. The greater the similarity of precipitation, the more similar the water resource availability, humidity, and vegetation growth conditions between the target area and the first area, thus making the shrub biomass assessment and prediction model of the first area more applicable to the target area. In other words, if the precipitation patterns in two areas are similar, then their vegetation growth characteristics are more likely to be similar, which makes the prediction results closer to the actual situation and with less error when using the model of the first area for prediction. For example, assume that the precipitation in the target area and the first area shows similar seasonal fluctuations and precipitation levels in the past year, then the shrub biomass model of the first area can better reflect the ecological characteristics of the target area and provide a more accurate biomass prediction. On the contrary, if the precipitation in the two areas varies greatly, then the model of the first area may not be applicable to the target area, because different precipitation conditions will lead to differences in vegetation growth, thus affecting the accuracy of biomass assessment. If the prediction error is large, it may affect the effectiveness of ecological monitoring and environmental protection decisions. For example, in some areas, the shrub biomass may be underestimated or overestimated incorrectly, resulting in inappropriate management measures or resource allocation, thus affecting the effectiveness of ecological protection. Therefore, ensuring a large similarity of precipitation helps to improve the prediction accuracy of the model, and thus better support ecological monitoring and protection work.

[0035] In one embodiment, the steps of calculating the shrub environmental similarity coefficient between the target area and each first area according to the altitude similarity coefficient and the precipitation similarity coefficient are as follows: Convert the altitude similarity coefficient and the precipitation similarity coefficient into a comprehensive feature vector, use the comprehensive feature vector as the input of the machine learning model. The machine learning model takes predicting each shrub environmental similarity coefficient label as the prediction target with the comprehensive feature vector of each group, and takes minimizing the sum of prediction errors for all shrub environmental similarity coefficient labels as the training target. Train the machine learning model until the sum of prediction errors reaches convergence and then stop the model training. Determine the shrub environmental similarity coefficient according to the model output result, where the machine learning model is a polynomial regression model.

[0036] It should be noted that when calculating the shrub environment similarity coefficient between the target area and each first area according to the altitude similarity coefficient and the rainfall similarity coefficient, it is first necessary to convert these two similarity coefficients into a comprehensive feature vector. The key to this step is to integrate the altitude similarity coefficient and the rainfall similarity coefficient into a single vector as the input features of the model. Each group of comprehensive feature vectors represents the environmental similarity between a target area and a first area, containing the similarity information of these two areas in terms of altitude and precipitation. Next, these comprehensive feature vectors are used as inputs to train a polynomial regression model, aiming to predict the shrub environment similarity coefficient label. The polynomial regression model can capture non-linear relationships and thus can better fit complex environmental similarity data. By using the training data, the model will learn the relationship between the altitude and rainfall similarity coefficients and the shrub environment similarity coefficient, so as to be able to generate accurate predictions. During the training process, the goal of the model is to minimize the sum of prediction errors, that is, by adjusting the model parameters, making the errors between all predicted values and the actual shrub environment similarity coefficient labels as small as possible until the prediction error reaches a convergence state. This process can be achieved through optimization algorithms such as gradient descent. When the error is minimized, the model training stops and can be used to predict the shrub environment similarity coefficient between the target area and other areas. Using the trained polynomial regression model, the shrub environment similarity coefficient between the target area and each first area can be accurately calculated, and this result will help to select the most suitable shrub biomass assessment and prediction model. On this basis, more accurate biomass prediction can be provided, reducing the impact of prediction errors between different regions, so as to ensure more scientific and effective decision-making in the fields of ecological monitoring, environmental protection, etc.

[0037] In one embodiment, the steps of selecting a model for evaluating and predicting the shrub biomass of the target area according to the shrub environment similarity coefficient as the target model are as follows: Denote the area where the shrub biomass evaluated and predicted by each existing shrub biomass evaluation and prediction model as the first area. Compare the shrub environment similarity coefficient between the target area and each first area with the preset shrub environment similarity coefficient threshold. If the shrub environment similarity coefficient is not less than the preset shrub environment similarity coefficient threshold, denote the corresponding first area as the available area, denote the shrub biomass evaluation and prediction model of the available area as the available model, select the model with the largest shrub environment similarity coefficient in the available models as the target model, and input the remote sensing data of the target area into the target model to evaluate and predict the shrub biomass of the target area; If all the shrub environment similarity coefficients are less than the preset shrub environment similarity coefficient threshold, retrain the target model according to the remote sensing data of the target area to evaluate and predict the shrub biomass of the target area.

[0038] It should be noted that the shrub environment similarity coefficients between the target area and the corresponding areas of each existing shrub biomass assessment and prediction model are compared. By setting a preset threshold for the shrub environment similarity coefficient, when the similarity coefficient of a certain area is not less than this threshold, it is considered that the shrub biomass assessment model of this area is available. Then, the model with the largest shrub environment similarity coefficient (i.e., the most matching the environment of the target area) is selected from these available models as the target model. By selecting the most similar model, the accuracy of biomass assessment and prediction can be maximized, and the interference of environmental differences on the prediction results can be reduced. At this time, the remote sensing data of the target area will be input into the target model for predicting the shrub biomass. If the shrub environment similarity coefficients of all existing models are less than the preset threshold, it indicates that none of the existing models can fit the target area. At this time, a new model needs to be trained based on the remote sensing data of the target area. This new model will be adjusted entirely based on the environmental data of the target area to better predict the shrub biomass of the target area. For example: Suppose the environmental characteristics of the target area are similar to those of the tropical climate, and the three existing shrub biomass assessment and prediction models are applicable to temperate, tropical, and cold regions respectively. By calculating the similarity of altitude and rainfall, we find that the tropical model is the most similar to the environment of the target area and has the largest shrub environment similarity coefficient. Therefore, it will be selected as the target model. If the environment of the target area is quite different from all existing models, resulting in similarity coefficients lower than the threshold, the system will choose to retrain a model based on the remote sensing data of the target area to ensure that the final prediction results can fully reflect the actual ecological conditions of the target area.

[0039] In one embodiment, if all the shrub environment similarity coefficients are less than the preset shrub environment similarity coefficient threshold, the steps for retraining the target model to evaluate and predict the shrub biomass of the target area based on the remote sensing data of the target area are as follows: Use the remotely sensed data of the target area that has been preprocessed and feature-extracted as input to prepare for model training; Select a machine learning algorithm and input the remotely sensed data of the target area into the model for training. The training process includes: Use the correspondence between the remotely sensed data of the target area and the known biomass for training.

[0040] Adjust the hyperparameters of the model to optimize the model performance; Evaluate the accuracy of the model through the cross-validation method and adjust the model parameters according to the evaluation results to ensure the optimal performance of the model in the target area; Use the trained target model to predict the shrub biomass of the remotely sensed data of the target area and output the estimated value of the shrub biomass of the target area.

[0041] It should be noted that when all the shrub environment similarity coefficients are less than the preset threshold, it indicates that the existing shrub biomass assessment model cannot accurately adapt to the environmental characteristics of the target area. Therefore, it is necessary to train a new target model based on the remote sensing data of the target area. First, the remote sensing data of the target area has been preprocessed and key features have been extracted, and these data will be used as the input for model training. Next, it is necessary to select appropriate machine learning algorithms, such as random forest, support vector machine (SVM), or deep neural network, etc., to adapt to the complex environmental characteristics and biomass distribution of the target area. The core of the training process is to compare the remote sensing data of the target area with the known actual biomass data, and optimize the model through the relationship between the known input and output. By adjusting the hyperparameters of the model (such as learning rate, number of trees, depth, etc.), the generalization ability and prediction accuracy of the model can be improved. The cross-validation method is an important step in the training process. By dividing the dataset into multiple subsets and repeatedly training and validating the model, overfitting can be avoided and the robustness of the model can be evaluated. Through these processes, it can be ensured that the model performs best for the target area, that is, the prediction results have high accuracy and reliability. Finally, the trained target model will be used to predict the shrub biomass of the remote sensing data of the target area, and the biomass estimation value of the target area will be output. The accuracy of this process directly affects the effect of subsequent ecological monitoring and environmental protection decisions. For example: Suppose the target area is located in a mountainous area of a certain tropical rainforest and the existing model is not applicable. In this case, researchers will collect the remote sensing data of this area (such as satellite images, lidar data, etc.) and extract relevant features, such as vegetation index, terrain information, etc. By selecting the support vector machine (SVM) as the training algorithm and using the known shrub biomass samples, a shrub biomass prediction model suitable for the tropical mountainous area is trained. During the training process, the performance of the model may be optimized by adjusting the kernel function and C parameter of the SVM, and its performance on different subsets is evaluated through cross-validation, and finally it is ensured that the trained model can effectively predict the shrub biomass of this area.

[0042] The above has described an embodiment of the present invention in detail, but the content described is only a preferred embodiment of the present invention and cannot be used to artificially limit the scope of implementation of the present invention. All equivalent changes and improvements made according to the scope of the application of the present invention should still fall within the scope covered by the patent of the present invention.

Claims

1. A method for evaluating and predicting shrub biomass based on remote sensing data, characterized in that, It includes the following steps: Obtain the remote sensing data of the target area, and preprocess the remote sensing data to obtain the preprocessed remote sensing data of the target area; Obtain the environmental information of the target area, and analyze the shrub environmental similarity coefficient of the target area based on the environmental information of the target area; Select a model for evaluating and predicting the shrub biomass of the target area according to the shrub environmental similarity coefficient as the target model; Evaluate and predict the biomass of the target area according to the preprocessed remote sensing data of the target area and the target model.

2. The shrub biomass assessment and prediction method based on remote sensing data according to claim 1, characterized in that The steps of obtaining the remote sensing data of the target area and preprocessing the remote sensing data are as follows: The remote sensing data includes optical remote sensing data, radar remote sensing data, lidar data, and hyperspectral remote sensing data; Perform spatial registration on the optical remote sensing data, radar remote sensing data, lidar data, and hyperspectral remote sensing data, and use an automatic matching and false matching rejection algorithm based on feature points to ensure the spatial consistency of multi-source data; and for automatic matching, use the SIFT feature matching combined with the RANSAC algorithm to reject false matches; For the optical remote sensing data, combine the atmospheric correction model with terrain correction to eliminate the influence of atmospheric scattering and terrain shadows; For the radar remote sensing data, reduce the speckle noise of the radar data through polarization calibration and adaptive filtering; the adaptive filtering uses the Lee filter combined with the polarization decomposition method; For the lidar data, generate a high-precision canopy height model based on the irregular triangular network (TIN) interpolation method to retain the three-dimensional structural characteristics of the shrubs; For the hyperspectral remote sensing data, select the characteristic bands sensitive to shrub biomass based on band information entropy and separability analysis; and the band selection includes the red edge band and the shortwave infrared band; Adopt a method combining principal component analysis and wavelet transform to fuse the spectral and texture information of optical and radar data; Use NDVI and NDWI to distinguish vegetation and non-vegetation areas; separate shrubs and low trees based on the radar cross-polarization ratio; eliminate low-confidence areas through lidar point cloud density analysis.

3. A method for evaluating and predicting shrub biomass based on remote sensing data according to claim 1, characterized in that, The steps of obtaining the environmental information of the target area and analyzing the shrub environmental similarity coefficient of the target area based on the environmental information of the target area are as follows: Obtain the radar remote sensing image of the target area, divide the radar remote sensing image into several grids, and obtain the altitude corresponding to each grid; Calculate the average altitude of the target area according to the altitude corresponding to each grid, and calculate the altitude standard deviation of the target area based on the average altitude; Obtain the areas where the shrub biomass evaluated and predicted by each existing shrub biomass evaluation and prediction model is located, denoted as the first areas, obtain the average altitude and altitude standard deviation of each first area, and calculate the absolute value of the difference between the average altitude and the average altitude of the target area, and the absolute value of the difference between the altitude standard deviation and the altitude standard deviation of the target area respectively; Perform a weighted sum of the absolute value of the average altitude difference and the absolute value of the altitude standard deviation difference to obtain the altitude similarity coefficient between the target area and each first area; Calculate the shrub environmental similarity coefficient between the target area and each first area according to the altitude similarity coefficient.

4. A method for evaluating and predicting shrub biomass based on remote sensing data according to claim 3, characterized in that, The steps of calculating the shrub environmental similarity coefficient of the target area according to the altitude similarity coefficient are as follows: Obtain the rainfall in the target area in the past year to get a rainfall sequence based on time order, which is used as the first rainfall sequence; Record the area where the shrub biomass evaluated and predicted by each existing shrub biomass evaluation and prediction model is located as the first area, obtain the rainfall in each first area in the past year, and get a rainfall sequence based on time order, which is used as the second rainfall sequence; Calculate the similarity value between the first rainfall sequence and the second rainfall sequence through the dynamic time warping algorithm to obtain the rainfall similarity coefficient; Calculate the shrub environment similarity coefficient between the target area and each first area according to the altitude similarity coefficient and the rainfall similarity coefficient.

5. The shrub biomass assessment and prediction method based on remote sensing data according to claim 4, characterized in that, The steps of calculating the shrub environment similarity coefficient between the target area and each first area according to the altitude similarity coefficient and the rainfall similarity coefficient are as follows: Convert the altitude similarity coefficient and the rainfall similarity coefficient into a comprehensive feature vector, use the comprehensive feature vector as the input of the machine learning model, take predicting each shrub environment similarity coefficient label by each group of comprehensive feature vectors as the prediction target, and take minimizing the sum of prediction errors for all shrub environment similarity coefficient labels as the training target. Train the machine learning model until the sum of prediction errors reaches convergence and then stop the model training. Determine the shrub environment similarity coefficient according to the model output result, where the machine learning model is a polynomial regression model.

6. The method for evaluating and predicting shrub biomass based on remote sensing data according to claim 1, wherein, The steps of selecting a model for evaluating and predicting the shrub biomass in the target area according to the shrub environment similarity coefficient as the target model are as follows: Record the area where the shrub biomass evaluated and predicted by each existing shrub biomass evaluation and prediction model is located as the first area. Compare the shrub environment similarity coefficients of the target area and each first area with the preset shrub environment similarity coefficient threshold. If the shrub environment similarity coefficient is not less than the preset shrub environment similarity coefficient threshold, record the corresponding first area as the available area, record the shrub biomass evaluation and prediction model in the available area as the available model, select the model with the largest shrub environment similarity coefficient in the available models as the target model, and input the remote sensing data of the target area into the target model to evaluate and predict the shrub biomass in the target area; If all shrub environment similarity coefficients are less than the preset shrub environment similarity coefficient threshold, retrain the target model to evaluate and predict the shrub biomass in the target area according to the remote sensing data of the target area.

7. A method for evaluating and predicting shrub biomass based on remote sensing data according to claim 6, characterized in that The steps of retraining the target model to evaluate and predict the shrub biomass in the target area according to the remote sensing data of the target area if all shrub environment similarity coefficients are less than the preset shrub environment similarity coefficient threshold are as follows: Use the remotely sensed data of the target area that has been preprocessed and feature-extracted as the input to prepare for model training; Select a machine learning algorithm and input the remotely sensed data of the target area into the model for training. The training process includes: Use the correspondence between the remotely sensed data of the target area and the known biomass for training; Adjust the hyperparameters of the model to optimize the model performance; Evaluate the accuracy of the model through the cross-validation method and adjust the model parameters according to the evaluation results to ensure the best performance of the model in the target area; Use the trained target model to predict the shrub biomass of the remote sensing data in the target area, and output the estimated value of the shrub biomass in the target area.

Images