A remote sensing monitoring method and system for grassland biomass

Abstract

The present invention discloses a remote sensing monitoring method and system for grassland biomass, comprising the following steps: constructing a high-dimensional feature matrix representing the state of the grassland ecosystem by using multi-modal remote sensing data fusion technology; generating spatio-temporal correlation parameters of grassland biomass based on the high-dimensional feature matrix and a dynamic spatio-temporal feature analysis framework; training a non-linear biomass prediction model by combining a deep neural network and a multi-task optimization algorithm; jointly analyzing the results of the prediction model with high-resolution images, and applying region segmentation technology to perform hierarchical processing on heterogeneous grassland areas; performing uncertainty propagation analysis on the partition estimation results and multi-modal data, and outputting optimized grassland biomass estimation results; combining the partition estimation results with cross-spatio-temporal scale data through a collaborative monitoring mechanism of unmanned aerial vehicles and satellite remote sensing to dynamically update the spatio-temporal correlation parameters; generating a spatial distribution map of grassland biomass and an analysis result of dynamic change trends according to the optimized monitoring data.

Description

Technical Field

[0001] The present invention relates to the technical field of remote sensing monitoring, and particularly to a method and system for remotely monitoring grassland biomass. Background Art

[0002] With the rapid development of remote sensing technology and ecological monitoring requirements, the monitoring of grassland biomass has become an important part of precision agriculture and ecological management. As an important characterization index of the grassland ecosystem, grassland biomass can not only reflect the productivity of the grassland, but also provide information on the health status and dynamic changes of the grassland. However, due to the spatial heterogeneity and spatio-temporal dynamics of the grassland, the existing remote sensing monitoring technologies for grassland biomass face many challenges in terms of accuracy and applicability.

[0003] In the prior art, traditional remote sensing monitoring methods for grassland biomass mostly rely on single data sources or estimation methods based on simple empirical models. These methods often show significant limitations when faced with complex terrains, diverse grassland ecological environments, and dynamically changing biomass characteristics. Specifically, the existing methods mainly have the following deficiencies:

[0004] 1. Insufficient data fusion ability: The existing technology usually only relies on a single remote sensing data source or simple multi-source data superposition, lacking an effective data fusion mechanism, resulting in a single dimension of monitoring data and being unable to fully reflect the complex characteristics of the grassland ecosystem;

[0005] 2. Low model accuracy: Traditional grassland biomass estimation models are mostly based on linear regression or simple exponential formulas, failing to fully consider non-linear factors and grassland heterogeneity characteristics, resulting in low adaptability and prediction accuracy of the models;

[0006] 3. Inadequate response to dynamic changes: The existing methods lack in-depth analysis of the spatio-temporal dynamic characteristics of grassland biomass and are unable to accurately capture seasonal changes and heterogeneity distribution characteristics, resulting in insufficient reliability of the monitoring results in dynamic assessment;

[0007] 4. Weak uncertainty processing ability: Traditional remote sensing monitoring methods lack effective analysis of the quantification and propagation of uncertainties in multi-source data, making it difficult to evaluate the reliability of monitoring results and optimize the data processing process;

[0008] 5. Contradiction between spatial resolution and coverage: The existing technology is difficult to balance the requirements of high spatial resolution and large-scale coverage, and it is impossible to achieve refined monitoring and comprehensive coverage of grassland biomass simply relying on unmanned aerial vehicle or satellite data.

[0009] Therefore, how to provide a method and system for remotely monitoring grassland biomass is an urgent problem to be solved by those skilled in the art. Summary of the Invention

[0010] An object of the present invention is to provide a method and system for remotely sensing and monitoring grassland biomass. The present invention uses technologies such as multi-modal remote sensing data fusion, deep learning neural network modeling, and dynamic spatio-temporal feature analysis, and details the specific steps for achieving accurate estimation and dynamic monitoring of grassland biomass, with the advantages of high data fusion efficiency, high model prediction accuracy, and strong dynamic change response ability.

[0011] A method and system for remotely sensing and monitoring grassland biomass according to an embodiment of the present invention includes the following steps:

[0012] S1. Using multi-modal remote sensing data fusion technology, integrating unmanned aerial vehicle images, high-resolution satellite images, and temporal remote sensing data, generating a cross-resolution and multi-temporal joint feature space, and extracting grassland spectral features, texture features, and geometric structure features to construct a high-dimensional feature matrix representing the state of the grassland ecosystem;

[0013] S2. Based on the high-dimensional feature matrix and the dynamic spatio-temporal feature analysis framework, using the methods of temporal change decomposition and spatial heterogeneity clustering to generate spatio-temporal correlation parameters of grassland biomass and construct a description set of grassland dynamic change features;

[0014] S3. Using the description set of dynamic change features, combining a deep neural network and a multi-task optimization algorithm to train a non-linear biomass prediction model to achieve high-precision estimation of grassland biomass for the characteristics of multi-source data and the heterogeneous distribution of grasslands;

[0015] S4. Jointly analyzing the results of the prediction model with high-resolution images, applying regional segmentation technology to perform hierarchical processing on heterogeneous grassland areas to generate refined grassland biomass zonal estimation results to capture complex local changes in grasslands;

[0016] S5. Conducting uncertainty propagation analysis on the zonal estimation results and multi-modal data, using an uncertainty quantification optimization algorithm to evaluate the error sources of the monitoring data, and outputting optimized grassland biomass estimation results;

[0017] S6. Through the collaborative monitoring mechanism of unmanned aerial vehicles and satellite remote sensing, combining the zonal estimation results with cross-temporal and cross-scale data to dynamically update the spatio-temporal correlation parameters and optimize the real-time performance in the time dimension and the integrity in the space dimension of the grassland biomass monitoring results;

[0018] S7. Generating a high-resolution grassland biomass spatial distribution map and dynamic change trend analysis results based on the optimized monitoring data to form a comprehensive assessment of the state of the grassland ecosystem and the changes in grassland biomass.

[0019] Optionally, the specific content of S1 includes:

[0020] S11. Obtain UAV remote sensing image data, collect detailed information on the grassland surface through a multispectral sensor to generate high-resolution images, and at the same time obtain multi-temporal high-resolution satellite image data covering the study area and other remote sensing data;

[0021] S12. Perform data preprocessing on the UAV images and satellite images, including radiometric correction, geometric correction, and noise removal, to generate a standardized image set, making the multi-source data comparable in spectral, geometric, and temporal dimensions;

[0022] S13. Apply a data registration algorithm to spatially align remote sensing data with different resolutions and multi-temporal data, and use the SIFT algorithm to perform geometric registration on the images to generate a cross-resolution, multi-temporal remote sensing data set;

[0023] S14. Extract multi-modal features of the remote sensing data, including spectral features, texture features, and geometric structure features. The spectral features are calculated using the normalized difference vegetation index NDVI and the enhanced vegetation index EVI:

[0024]

[0025] where NIR represents the reflectance of the near-infrared band, R represents the reflectance of the red band, and B represents the reflectance of the blue band;

[0026] S15. Extract the texture features of the images, calculate the contrast, homogeneity, and energy indicators of the images using the gray-level co-occurrence matrix, and generate a multi-dimensional texture feature matrix;

[0027] S16. Extract the geometric structure features of the remote sensing images, combine the watershed segmentation algorithm and the edge detection algorithm to identify the grassland terrain boundaries and morphological structures, and generate a geometric feature description set related to the spatial distribution;

[0028] S17. Construct a high-dimensional feature matrix representing the state of the grassland ecosystem, and fuse the extracted spectral features, texture features, and geometric structure features into a joint feature vector.

[0029] Optionally, the S2 specifically includes:

[0030] S21. Use the multi-temporal remote sensing data in the high-dimensional feature matrix to establish a time series decomposition model, and extract seasonal and trend change features through the weighted moving average method:

[0031]

[0032] where T(t) is the trend component of the time series, X(t - i) is the observation value at the previous i moments, W(i) is the weighting factor, and n is the moving window length;

[0033] S22. Combine the spatial heterogeneity clustering method, and perform clustering processing on the spatial features in the high-dimensional feature matrix through the density-based spatial clustering algorithm to generate spatial partitions within the grassland area;

[0034] S23. Analyze the dynamic features of each spatial partition, and construct a grassland biomass dynamic change model with time and space dimensions as the core:

[0035] M(x, t) = F(S(x), T(t), P(x, t));

[0036] Among them, M(x, t) represents the dynamic features at position x and time t, S(x) is the spatial feature, T(t) is the time series feature, and P(x, t) is the external influence factor;

[0037] S24. Through the adaptive time window method, dynamically update the spatio-temporal correlation parameters within the partition, and generate a grassland dynamic feature description set according to the seasonal fluctuations and heterogeneous distribution changes of the grassland biomass.

[0038] Optionally, the specific steps of S3 are as follows:

[0039] S31. Divide the dynamic change feature description set into a training data set and a validation data set. Through normalization processing, map the spatio-temporal correlation parameters and external environmental variables to a unified numerical range to adapt to the input requirements of the deep neural network;

[0040] S32. Construct a multi-layer deep neural network structure, including an input layer, a hidden layer, and an output layer. The input layer receives the key features in the dynamic change feature description set. The hidden layer realizes non-linear feature transformation through an activation function, and the output layer generates the predicted value of the grassland biomass:

[0041] y = f(W3·g(W2·g(W1·x + b1) + b2) + b3);

[0042] Among them, x is the input feature vector, W1, W2, W3 are weight matrices, b1, b2, b3 are bias vectors, g(·) is the activation function, and f(·) is the activation function of the output layer;

[0043] S33. Combine the multi-task optimization algorithm, and define the loss function for the grassland biomass prediction task. The loss function consists of the mean square error and the uncertainty quantification error:

[0044] L = α·MSE + β·UQE;

[0045] Among them, α and β are task weight parameters, UQE is the uncertainty measure of the predicted value;

[0046] S34. Optimize the network weights and bias parameters using the backpropagation algorithm, adjust the network structure according to the gradient update principle of the loss function, and gradually minimize the prediction error;

[0047] S35. Verify the optimized model, calculate the accuracy metrics of the prediction results using the validation dataset, including the correlation coefficient and relative error, to enable the applicability of the model under the grassland heterogeneity distribution and multi-source data characteristics;

[0048] S36. Take the trained and validated deep neural network model as the final non-linear biomass prediction model, and generate the estimated results of the spatial distribution of grassland biomass.

[0049] Optionally, the specific steps of S4 are as follows:

[0050] S41. Conduct a spatial overlap analysis of the grassland biomass distribution results generated by the prediction model and the high-resolution remote sensing image to calibrate the spatial distribution consistency between the prediction results and the image;

[0051] S42. Based on the calibrated data, use the region segmentation technology to segment the heterogeneous grassland areas. Adopt the watershed algorithm to generate the initial segmentation regions according to the spectral differences and topographic change characteristics of the image;

[0052] S43. Optimize the initial segmentation regions, combine the edge detection method and morphological operations to eliminate the over-segmented and under-segmented regions. The optimized segmentation regions are represented in vector form as:

[0053] Ri = {x|x∈Ω, g(x)>τi};

[0054] where, R i is the i-th segmentation region, x is the pixel point, Ω is the global research area, g(x) is the pixel feature function within the region, and τ i is the segmentation threshold;

[0055] S44. Combine the segmentation regions, analyze the biomass distribution characteristics within each region, calculate the mean and standard deviation of the biomass within the region, and generate the zonal estimation parameters:

[0056]

[0057] where, μ i is the biomass mean of the i-th segmentation region, σ i is the standard deviation, y(x) is the predicted biomass value of the pixel point x, and |R i | is the number of pixels within the region;

[0058] S45. Combine the zonal estimation parameters with the prediction model results to generate a refined zonal estimation map of grassland biomass, record the biomass distribution and statistical information of each region, and use it to capture the complex local change characteristics of the grassland.

[0059] Optionally, the S5 specifically includes:

[0060] S51. Integrate the zonal estimation results with multi-modal remote sensing data, analyze the error distribution of each data source through the uncertainty propagation model, and use the error distribution function to describe the random characteristics of the error sources:

[0061]

[0062] where e(x) is the error component, μ e is the error mean, is the error variance;

[0063] S52. Construct an uncertainty propagation formula, calculate the uncertainty measure of the zonal biomass estimation results, and describe the uncertainty range through the propagation relationship of the errors:

[0064]

[0065] where U(y) is the uncertainty of the output, y is the biomass estimation value, x i is the input feature variable, σ i is the standard deviation of the variable x i ;

[0066] S53. Use the uncertainty quantification optimization algorithm to adjust the estimation model parameters based on the optimization objective function:

[0067] min[L(y)+γ·U(y)];

[0068] where L(y) is the error loss function between the estimation value and the actual observed value, and γ is the uncertainty weight parameter;

[0069] S54. Verify the optimized biomass estimation results, analyze the mean square error and standardized residual distribution within the region, and verify whether the uncertainty level of the optimization results meets the predetermined threshold conditions;

[0070] S55. Output the optimized grassland biomass estimation results, including the estimation value and its corresponding uncertainty range.

[0071] Optionally, the S6 specifically includes:

[0072] S6

[0073] S62. Perform spatial registration processing on cross-scale data, and use a multi-scale registration algorithm to align the detailed information of UAV images with the macroscopic coverage data of satellite images. The registered data is represented in joint resolution as:

[0074] D(x,t) = Fmerge(Dsatellite(x,t), D UAV (x));

[0075] where D(x,t) is the spatio-temporal joint remote sensing data, D satellite (x,t) is the satellite data, D UAV (x) is the UAV image data, and F merge is the data fusion function;

[0076] S63. Use the spatio-temporal scale compensation algorithm to dynamically balance the temporal and spatial resolutions D(x,t) in the joint remote sensing data, so as to achieve the consistency of the UAV and satellite data in the time dimension and the integrity in the spatial dimension;

[0077] S64. Combine the partition estimation results, and optimize the spatio-temporal correlation parameters of the grassland biomass based on the dynamic update model. The updated correlation parameter matrix can be expressed as:

[0078] P(t,x) = α·Pprev(t,x) + β·Pnew(t,x);

[0079] where P(t,x) is the optimized correlation parameter matrix, P prev (t,x) is the previous parameter, P new (t,x) is the correlation parameter of the new data, and α and β are weight factors;

[0080] S65. Based on the updated spatio-temporal correlation parameter P(t,x), generate the optimized grassland biomass monitoring results, and at the same time output the dynamic change trend chart to characterize the change law of the grassland biomass across spatio-temporal scales.

[0081] Optionally, the specific steps of S7 include:

[0082] S71. Map the optimized grassland biomass monitoring data to a high-resolution spatial grid, and calculate the spatial distribution value of the biomass for each grid cell:

[0083]

[0084] where B(x,y) is the biomass value of the grid cell (x,y), b i (x,y) is the original biomass value of the corresponding cell in the monitoring data, and w i is the optimized weight factor;

[0085] S72. Based on the distribution results, construct a high-resolution spatial distribution map of grassland biomass, and use the interpolation algorithm to fill the unobserved areas and maintain consistency with the observed data;

[0086] S73. Extract the dynamic change trend of grassland biomass in the time dimension, and calculate the overall change rate for the spatial distribution map at each time series point:

[0087]

[0088] where, R t is the overall change rate at the t-th moment, and B t (x i , y i ) and B t-1 (x i , y i ) are the biomass values at the current moment and the previous moment respectively;

[0089] S74. Conduct statistical analysis on the dynamic change trend data, generate distribution maps of the biomass growth rate and decline rate within the region, and record the abnormal change areas at the same time;

[0090] S75. Integrate the spatial distribution map and the analysis results of the dynamic change trend to generate a comprehensive assessment report of the grassland ecosystem.

[0091] Optionally, it includes the following modules:

[0092] Remote sensing data acquisition module: Collect multi-modal remote sensing data, including high-resolution drone images, multi-temporal satellite remote sensing data, and other auxiliary remote sensing data;

[0093] Data preprocessing module: Perform radiometric correction, geometric correction, and noise removal on the remote sensing data to generate a standardized remote sensing image dataset;

[0094] Feature extraction module: Extract spectral features, texture features, and geometric structure features from the standardized remote sensing data, and construct a high-dimensional feature matrix representing the state of the grassland ecosystem;

[0095] Dynamic analysis module: Design a dynamic spatio-temporal feature analysis framework based on the high-dimensional feature matrix, and use the time series change decomposition and spatial heterogeneity clustering methods to generate the spatio-temporal correlation parameters of grassland biomass;

[0096] Prediction modeling module: Combine deep learning algorithms and dynamic feature description sets to train a non-linear biomass prediction model to achieve accurate estimation of grassland biomass;

[0097] Partition Optimization Module: Through the joint analysis of high-resolution images and the results of the prediction model, using regional segmentation technology to stratify heterogeneous grassland areas, and generating refined zonal estimation results of grassland biomass;

[0098] Uncertainty Analysis Module: Through uncertainty propagation analysis and quantization optimization algorithms, evaluate the error sources of monitoring data and generate optimized biomass estimation results;

[0099] Collaborative Monitoring Module: Through the collaborative monitoring mechanism of drones and satellite remote sensing, fuse cross-temporal and cross-spatial scale data, dynamically update spatio-temporal correlation parameters, and optimize grassland biomass monitoring results;

[0100] Evaluation and Visualization Module: Generate high-resolution spatial distribution maps of grassland biomass and analysis results of dynamic change trends, and provide comprehensive evaluation reports on grassland ecosystems.

[0101] The beneficial effects of the present invention are:

[0102] By combining multi-modal remote sensing data fusion technology, deep learning neural network modeling, dynamic spatio-temporal feature analysis, and high-resolution image segmentation technology, the present invention realizes the accurate estimation and dynamic monitoring of grassland biomass, enabling the system to comprehensively respond to the complex dynamic changes of grassland ecosystems, especially effectively improving the monitoring accuracy and adaptability in heterogeneous grassland environments. At the same time, using uncertainty quantification analysis and optimization algorithms, it is possible to evaluate and optimize the uncertainty of multi-source data, thereby outputting more reliable monitoring results.

[0103] Through the collaborative monitoring mechanism of drones and satellite remote sensing, the present invention organically combines high-resolution and multi-temporal data, dynamically updates the spatio-temporal correlation parameters of grassland biomass, and significantly improves the real-time performance in the time dimension and the integrity in the space dimension of the monitoring results. This not only solves the contradiction between high resolution and large-scale coverage in traditional methods, but also makes the generation of spatial distribution maps and dynamic change trends of grassland biomass more accurate and efficient.

[0104] By constructing a dynamic feature description set and an optimized deep learning model, the present invention provides a full-process technical framework for grassland biomass monitoring from feature extraction to model prediction, and can describe the high-resolution spatial distribution and analyze the temporal dynamic trends of grassland biomass. At the same time, combined with automated regional segmentation and zonal estimation techniques, it captures complex local changes in grasslands, provides comprehensive and detailed grassland ecological assessments, and effectively supports scientific decision-making in precision agriculture and grassland ecological management. Description of the Drawings

[0105] The accompanying drawings are used to provide a further understanding of the present invention and form a part of the specification. They are used together with the embodiments of the present invention to explain the present invention and do not constitute a limitation to the present invention. In the drawings:

[0106] Figure 1 is the overall framework diagram of a method and system for remote sensing monitoring of grassland biomass proposed by the present invention;

[0107] Figure 2 is the construction flow chart of the grassland biomass prediction model based on dynamic spatio-temporal feature analysis in the present invention. Detailed implementation manners

[0108] Now, the present invention will be further described in detail with reference to the accompanying drawings. These drawings are all simplified schematic diagrams, only showing the basic structure of the present invention in a schematic manner, so they only show the components related to the present invention.

[0109] Refer to Figure 1-2 , a method and system for remote sensing monitoring of grassland biomass, including the following steps:

[0110] S1. Using multi-modal remote sensing data fusion technology, integrating unmanned aerial vehicle images, high-resolution satellite images and time-series remote sensing data, generating a cross-resolution and multi-temporal joint feature space, and extracting grassland spectral features, texture features and geometric structure features to construct a high-dimensional feature matrix representing the state of the grassland ecosystem;

[0111] S2. Based on the high-dimensional feature matrix and the dynamic spatio-temporal feature analysis framework, using time-series change decomposition and spatial heterogeneity clustering methods to generate spatio-temporal correlation parameters of grassland biomass and construct a description set of grassland dynamic change features;

[0112] S3. Using the description set of dynamic change features, combining with a deep neural network and a multi-task optimization algorithm to train a non-linear biomass prediction model to achieve high-precision grassland biomass estimation for the characteristics of multi-source data and the heterogeneous distribution of grasslands;

[0113] S4. Jointly analyzing the results of the prediction model with high-resolution images, applying regional segmentation technology to perform hierarchical processing on heterogeneous grassland areas to generate refined grassland biomass zonal estimation results to capture complex local changes in grasslands;

[0114] S5. Conducting uncertainty propagation analysis on the zonal estimation results and multi-modal data, using an uncertainty quantification optimization algorithm to evaluate the error sources of the monitoring data and output optimized grassland biomass estimation results;

[0115] S6. Through the collaborative monitoring mechanism of drones and satellite remote sensing, combine the zonal estimation results with cross-temporal and cross-spatial scale data, dynamically update the spatio-temporal correlation parameters, and optimize the real-time performance in the time dimension and the integrity in the space dimension of the grassland biomass monitoring results;

[0116] S7. Generate a high-resolution spatial distribution map of grassland biomass and an analysis result of dynamic change trends based on the optimized monitoring data, and form a comprehensive assessment of the state of the grassland ecosystem and the changes in grassland biomass.

[0117] In this embodiment, the specific steps of S1 are as follows:

[0118] S11. Obtain drone remote sensing image data, collect detailed information on the grassland surface through a multispectral sensor to generate high-resolution images, and at the same time obtain multi-temporal high-resolution satellite image data covering the study area and other remote sensing data;

[0119] S12. Perform data preprocessing on the drone images and satellite images, including radiometric correction, geometric correction, and noise removal, to generate a standardized image set, making the multi-source data comparable in the spectral, geometric, and time dimensions;

[0120] S13. Apply a data registration algorithm to spatially align remote sensing data with different resolutions and multi-temporal phases, and use the SIFT algorithm to perform geometric registration on the images to generate a remote sensing data set with cross-resolution and multi-temporal phases;

[0121] S14. Extract multi-modal features of the remote sensing data, including spectral features, texture features, and geometric structure features. The spectral features are calculated using the normalized difference vegetation index NDVI and the enhanced vegetation index EVI:

[0122]

[0123] where NIR represents the reflectance in the near-infrared band, R represents the reflectance in the red band, and B represents the reflectance in the blue band;

[0124] S15. Extract the texture features of the images, calculate the contrast, homogeneity, and energy indicators of the images using the gray-level co-occurrence matrix, and generate a multi-dimensional texture feature matrix;

[0125] S16. Extract the geometric structure features of the remote sensing images, combine the watershed segmentation algorithm and the edge detection algorithm to identify the grassland terrain boundaries and morphological structures, and generate a geometric feature description set related to the spatial distribution;

[0126] S17. Construct a high-dimensional feature matrix representing the state of the grassland ecosystem, and fuse the extracted spectral features, texture features, and geometric structure features into a joint feature vector.

[0127] In this embodiment, the specific steps of S2 are as follows:

[0128] S21. Use the time-series remote sensing data in the high-dimensional feature matrix to establish a time series decomposition model, and extract seasonal and trend change features through the weighted moving average method:

[0129]

[0130] Among them, T(t) is the trend component of the time series, X(t-i) is the observed value at the previous i moments, W(i) is the weighting factor, and n is the moving window length;

[0131] S22. Combine the spatial heterogeneity clustering method, and perform clustering processing on the spatial features in the high-dimensional feature matrix through the density-based spatial clustering algorithm to generate spatial partitions within the grassland area;

[0132] S23. Analyze the dynamic characteristics of each spatial partition, and construct a dynamic change model of grassland biomass with time and space dimensions as the core:

[0133] M(x,t) = F(S(x), T(t), P(x,t));

[0134] Among them, M(x,t) represents the dynamic characteristics at position x and time t, S(x) is the spatial feature, T(t) is the time series feature, and P(x,t) is the external influence factor;

[0135] S24. Through the adaptive time window method, dynamically update the spatio-temporal correlation parameters within the partition, and generate a description set of grassland dynamic characteristics according to the seasonal fluctuations and heterogeneous distribution changes of grassland biomass.

[0136] In this embodiment, the specific content of S3 includes:

[0137] S31. Divide the description set of dynamic change characteristics into a training data set and a validation data set, and through normalization processing, map the spatio-temporal correlation parameters and external environmental variables to a unified numerical range to adapt to the input requirements of the deep neural network;

[0138] S32. Construct a multi-layer deep neural network structure, including an input layer, a hidden layer, and an output layer. The input layer receives the key features in the description set of dynamic change characteristics. The hidden layer realizes non-linear feature transformation through the activation function, and the output layer generates the predicted value of grassland biomass:

[0139] y = f(W3·g(W2·g(W1·x + b1)+b2)+b3);

[0140] Among them, x is the input feature vector, W1, W2, W3 are weight matrices, b1, b2, b3 are bias vectors, g(·) is the activation function, and f(·) is the activation function of the output layer;

[0141] S33. Combine with the multi-task optimization algorithm to define the loss function for the grassland biomass prediction task. The loss function consists of the mean square error and the uncertainty quantification error:

[0142] L = α·MSE + β·UQE;

[0143] where α and β are task weight parameters, UQE is the uncertainty measure of the predicted value;

[0144] S34. Use the backpropagation algorithm to optimize the network weight and bias parameters, and adjust the network structure according to the gradient update principle of the loss function to gradually minimize the prediction error;

[0145] S35. Verify the optimized model, and calculate the accuracy indicators of the prediction results using the validation dataset, including the correlation coefficient and the relative error, to make the model applicable under the grassland heterogeneity distribution and multi-source data characteristics;

[0146] S36. Take the trained and verified deep neural network model as the final non-linear biomass prediction model to generate the estimated results of the spatial distribution of grassland biomass.

[0147] In this embodiment, the specific steps of S4 include:

[0148] S41. Conduct a spatial overlap analysis of the grassland biomass distribution results generated by the prediction model and the high-resolution remote sensing image to calibrate the spatial distribution consistency between the prediction results and the image;

[0149] S42. Based on the calibrated data, use the region segmentation technology to segment the heterogeneous grassland areas. Adopt the watershed algorithm to generate the initial segmentation regions according to the spectral differences and topographic change characteristics of the image;

[0150] S43. Optimize the initial segmentation regions, and combine the edge detection method and morphological operations to eliminate the over-segmented and under-segmented regions. The optimized segmentation regions are represented in vector form as:

[0151] Ri = {x|x∈Ω, g(x)>τi};

[0152] where R i is the i-th segmentation region, x is the pixel point, Ω is the global research region, g(x) is the pixel feature function within the region, and τ i is the segmentation threshold;

[0153] S44. Combine the segmentation regions, analyze the biomass distribution characteristics within each region, calculate the mean and standard deviation of the biomass within the region, and generate the zonal estimation parameters:

[0154]

[0155] Among them, μ i is the biomass mean of the i-th divided area, σ i is the standard deviation, y(x) is the predicted biomass value of pixel point x, |R i | is the number of pixels within the area;

[0156] S45. Combine the partition estimation parameters with the prediction model results to generate a refined grassland biomass partition estimation map, record the biomass distribution and statistical information of each area, and use it to capture the complex local change characteristics of the grassland.

[0157] In this embodiment, the S5 specifically includes:

[0158] S51. Integrate the partition estimation results with multi-modal remote sensing data, analyze the error distribution of each data source through the uncertainty propagation model, and use the error distribution function to describe the random characteristics of the error sources:

[0159]

[0160] Among them, e(x) is the error component, μ e is the error mean, is the error variance;

[0161] S52. Construct an uncertainty propagation formula, calculate the uncertainty measure of the partition biomass estimation result, and describe the uncertainty range through the propagation relationship of errors:

[0162]

[0163] Among them, U(y) is the output uncertainty, y is the biomass estimation value, x i is the input feature variable, σ i is the standard deviation of variable x i ;

[0164] S53. Use the uncertainty quantification optimization algorithm to adjust the estimation model parameters based on the optimization objective function:

[0165] min[L(y)+γ·U(y)];

[0166] Among them, L(y) is the error loss function between the estimation value and the actual observation value, and γ is the uncertainty weight parameter;

[0167] S54. Verify the optimized biomass estimation result, analyze the mean square error and standardized residual distribution within the area, and verify whether the uncertainty level of the optimization result meets the predetermined threshold condition;

[0168] S55. Output the optimized grassland biomass estimation results, including the estimated values and their corresponding uncertainty ranges.

[0169] In this embodiment, the specific steps of S6 are as follows:

[0170] S61. Collect multi-temporal satellite remote sensing data covering the target area to obtain the macroscopic dynamic change information of the grassland, and combine it with the high-resolution image data obtained by the unmanned aerial vehicle to generate a remote sensing data set across time and space scales.

[0171] S62. Perform spatial registration processing on the cross-scale data, and use the multi-scale registration algorithm to align the detailed information of the unmanned aerial vehicle image with the macroscopic coverage data of the satellite image. The registered data is represented by the joint resolution as:

[0172] D(x,t) = Fmerge(Dsatellite(x,t), D UAV (x));

[0173] where D(x,t) is the spatio-temporal joint remote sensing data, D satellite (x,t) is the satellite data, D UAV (x) is the unmanned aerial vehicle image data, and F merge is the data fusion function.

[0174] S63. Use the spatio-temporal scale compensation algorithm to dynamically balance the temporal and spatial resolutions D(x,t) in the joint remote sensing data, so as to ensure the consistency of the unmanned aerial vehicle and satellite data in the time dimension and the integrity in the spatial dimension.

[0175] S64. Combine the zonal estimation results, and optimize the spatio-temporal correlation parameters of the grassland biomass based on the dynamic update model. The updated correlation parameter matrix can be expressed as:

[0176] P(t,x) = α·Pprev(t,x) + β·Pnew(t,x);

[0177] where P(t,x) is the optimized correlation parameter matrix, P prev (t,x) is the previous parameter, P new (t,x) is the correlation parameter of the new data, and α and β are weight factors.

[0178] S65. Based on the updated spatio-temporal correlation parameter P(t,x), generate the optimized grassland biomass monitoring results, and at the same time output the dynamic change trend chart to characterize the change law of the grassland biomass across spatio-temporal scales.

[0179] In this embodiment, the specific steps of S7 are as follows:

[0180] S71. Map the optimized grassland biomass monitoring data to a high - resolution spatial grid, and calculate the spatial distribution value of biomass for each grid cell:

[0181]

[0182] Among them, B(x, y) is the biomass value of the grid cell (x, y), and b i (x, y) is the original biomass value of the corresponding cell in the monitoring data, and w i is the optimized weight factor;

[0183] S72. Based on the distribution result, construct a high - resolution grassland biomass spatial distribution map, and use the interpolation algorithm to fill the unobserved area and maintain consistency with the observed data;

[0184] S73. Extract the dynamic change trend of grassland biomass in the time dimension, and calculate the overall change rate for the spatial distribution map at each time series point:

[0185]

[0186] Among them, R t is the overall change rate at the t - th moment, B t (x i , y i ) and B t-1 (x i , y i ) are the biomass values at the current moment and the previous moment respectively;

[0187] S74. Conduct statistical analysis on the dynamic change trend data, generate distribution maps of biomass growth rate and decline rate within the region, and record the abnormal change regions;

[0188] S75. Integrate the spatial distribution map and the results of dynamic change trend analysis to generate a comprehensive assessment report of the grassland ecosystem.

[0189] In this embodiment, the following modules are included:

[0190] Remote sensing data acquisition module: Collect multi - modal remote sensing data, including high - resolution UAV images, multi - temporal satellite remote sensing data, and other auxiliary remote sensing data;

[0191] Data pre - processing module: Perform radiometric correction, geometric correction, and noise removal on the remote sensing data to generate a standardized remote sensing image dataset;

[0192] Feature extraction module: Extract spectral features, texture features, and geometric structure features from the standardized remote sensing data, and construct a high - dimensional feature matrix representing the state of the grassland ecosystem;

[0193] Dynamic Analysis Module: Design a dynamic spatio-temporal feature analysis framework based on a high-dimensional feature matrix, and adopt time-series change decomposition and spatial heterogeneity clustering methods to generate spatio-temporal correlation parameters of grassland biomass;

[0194] Prediction Modeling Module: Combine deep learning algorithms and dynamic feature description sets to train a non-linear biomass prediction model to achieve accurate estimation of grassland biomass;

[0195] Zoning Optimization Module: Through the joint analysis of high-resolution images and the results of the prediction model, use regional segmentation technology to perform hierarchical processing on heterogeneous grassland areas to generate refined zonal estimation results of grassland biomass;

[0196] Uncertainty Analysis Module: Through uncertainty propagation analysis and quantization optimization algorithms, evaluate the error sources of monitoring data and generate optimized biomass estimation results;

[0197] Collaborative Monitoring Module: Through the collaborative monitoring mechanism of drones and satellite remote sensing, fuse cross-scale spatio-temporal data, dynamically update spatio-temporal correlation parameters, and optimize grassland biomass monitoring results;

[0198] Evaluation and Visualization Module: Generate high-resolution spatial distribution maps of grassland biomass and dynamic change trend analysis results, and provide a comprehensive evaluation report on the grassland ecosystem.

[0199] Example 1:

[0200] To verify the feasibility of the present invention in implementation, the present invention is applied to a certain grassland area in Hainan Tibetan Autonomous Prefecture, Qinghai Province. To solve the problems of insufficient data fusion ability, low model accuracy, and insufficient response to grassland dynamic changes in traditional grassland biomass monitoring, a grassland biomass remote sensing monitoring method and system proposed by the present invention are used to dynamically monitor and evaluate the research area. This monitoring combines multi-modal remote sensing data, including drone high-resolution images and multi-temporal satellite remote sensing data, and realizes accurate monitoring of the grassland ecosystem through deep learning model optimization, biomass dynamic change analysis, and spatial distribution evaluation technology.

[0201] In this scenario, multi-modal remote sensing data of the study area was collected first. The UAV image data was collected in early July 2023, covering a total area of 120 square kilometers within the study area, with a resolution of 0.1 meter, mainly recording the detailed features of the grassland, such as vegetation coverage rate and regional texture features. At the same time, multi-temporal satellite remote sensing images from May to August 2023 with a resolution of 10 meters were obtained to record the macroscopic changes of the grassland. All data was processed through spatial registration and dynamic spatio-temporal scale compensation techniques to generate a remote sensing data set across resolutions and time phases. Through the dynamic spatio-temporal feature analysis module of the present invention, a dynamic change model of grassland biomass was constructed, and key spatio-temporal correlation parameters were extracted to provide basic data for subsequent model optimization and monitoring result analysis.

[0202] After data processing, a deep learning neural network was used to estimate and model the grassland biomass. The input features of the model included spectral features, texture features, and geometric structure features, with a total of 12,000 sample points. After 50 rounds of training, the final mean square error on the validation set was 4.2 g / m 2 , and the correlation coefficient reached 0.92. By comparing with the traditional linear regression model, the accuracy of the model of the present invention was improved by about 25%, verifying the effectiveness of the combination of dynamic feature modeling and deep learning.

[0203] Based on the partition optimization technology, taking advantage of the high resolution of UAV images, the grassland area was refined and partitioned according to grassland heterogeneity and terrain features. The partition results showed that the study area could be divided into 8 main partitions, including grassland dense areas, sparse areas, and mixed terrains. The average biomass distribution of the partitions generated by the region segmentation technology is as follows. In the dynamic trend analysis, the change of grassland biomass was analyzed by combining the spatio-temporal update model, and it was found that the grassland biomass gradually increased from May to July and reached the peak in August. The dynamic change trend within the study area showed that the biomass increase rate in the high-coverage area was relatively fast, while the biomass change in the sparse area was relatively small.

[0204] Table 1 Average Biomass Distribution Table of Partitions (Unit: g / m 2 )

[0205]

[0206]

[0207] Table 2 Dynamic Change Trend Table of Grassland Biomass (Unit: g / m 2 )

[0208] Time Average biomass Biomass growth rate (%) May 2023 240.3 - June 2023 275.6 14.7 July 2023 320.8 16.4 August 2023 345.2 7.6

[0209] As can be seen from the above table, the present study finally generated a high-resolution spatial distribution map and a dynamic change trend map of grassland biomass, and comprehensively evaluated the state of the grassland ecosystem. The results show that the method of the present invention can effectively solve the contradiction problem between spatial resolution and coverage in traditional monitoring methods, while improving the accuracy of biomass estimation and the dynamic response ability. In the practical application of the study area, the comprehensive evaluation results of the system have provided key data support for grassland ecological protection and precision agriculture.

[0210] In summary, the present invention has successfully solved the limitations of traditional remote sensing monitoring methods for grassland biomass in terms of single data, low model accuracy, and insufficient dynamic response by introducing multi-modal remote sensing data fusion, dynamic spatio-temporal feature analysis, deep learning modeling, and regional optimization techniques, effectively improving the accuracy of grassland biomass monitoring, dynamically capturing the change trend of grassland biomass, and at the same time realizing the refined analysis of high-resolution spatial distribution.

[0211] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and should be covered by the protection scope of the present invention.

Claims

1. A remote sensing monitoring method for grassland biomass, characterized in that It includes the following steps: S1. Using the multi-modal remote sensing data fusion technology, integrating unmanned aerial vehicle (UAV) images, satellite images and time-series remote sensing data, generating a joint feature space with cross-resolution and multi-temporal characteristics, extracting grassland spectral features, texture features and geometric structure features, and constructing a high-dimensional feature matrix representing the state of the grassland ecosystem; S2. Based on the high-dimensional feature matrix and the dynamic spatio-temporal feature analysis framework, using the time-series change decomposition and spatial heterogeneity clustering methods, generating the spatio-temporal correlation parameters of grassland biomass, and constructing a description set of grassland dynamic change characteristics; S3. Using the description set of dynamic change characteristics, combining the deep neural network and the multi-task optimization algorithm, training a non-linear biomass prediction model to achieve high-precision grassland biomass estimation for the characteristics of multi-source data and the heterogeneous distribution of grasslands; S4. Jointly analyzing the results of the prediction model and the images, applying the region segmentation technology to perform hierarchical processing on the heterogeneous grassland areas, and generating refined grassland biomass zonal estimation results to capture the complex local changes of grasslands; The specific content of S4 includes: S41. Conducting a spatial overlap analysis of the grassland biomass distribution results generated by the prediction model and the remote sensing images to calibrate the spatial distribution consistency between the prediction results and the images; S42. Based on the calibrated data, using the region segmentation technology to segment the heterogeneous grassland areas, and adopting the watershed algorithm to generate the initial segmentation areas according to the spectral differences and terrain change characteristics of the images; S43. Optimizing the initial segmentation areas, combining the edge detection method and morphological operations to eliminate the over-segmented and under-segmented areas, and the optimized segmentation areas are represented in vector form as: R i = {x | x ∈ Ω, g(x) > τ i}; Among them, R i is the i-th segmented region, x is a pixel point, Ω is the global research region, g(x) is the pixel feature function within the region, and τ i is the segmentation threshold; S44. Combining the segmentation areas, analyzing the biomass distribution characteristics within each area, calculating the mean and standard deviation of the biomass within the area, and generating zonal estimation parameters; where μ i is the mean biomass of the i-th segmented region, σ i is the standard deviation, y(x) is the predicted biomass value of pixel point x, and |R i | is the number of pixels in the region; S45. Jointly generating a refined grassland biomass zonal estimation map with the zonal estimation parameters and the results of the prediction model, recording the biomass distribution and statistical information of each area, and being used to capture the complex local change characteristics of grasslands; S5. Conducting uncertainty propagation analysis on the zonal estimation results and multi-modal data, using the uncertainty quantification optimization algorithm to evaluate the error sources of the monitoring data, and outputting the optimized grassland biomass estimation results; S6. Through the collaborative monitoring mechanism of UAV and satellite remote sensing, combining the zonal estimation results with the cross-temporal and cross-spatial scale data, dynamically updating the spatio-temporal correlation parameters, and optimizing the real-time performance in the time dimension and the integrity in the spatial dimension of the grassland biomass monitoring results; S7. Generating a grassland biomass spatial distribution map and dynamic change trend analysis results based on the optimized monitoring data, and forming a comprehensive assessment of the state of the grassland ecosystem and the changes in grassland biomass.

2. The remote sensing monitoring method for grassland biomass according to claim 1, wherein The specific content of S1 includes: S11. Obtaining UAV remote sensing image data, collecting the grassland surface detail information through a multi-spectral sensor to generate images, and simultaneously obtaining multi-temporal satellite image data and other remote sensing data covering the study area; S12. Perform data preprocessing on the UAV images and satellite images, including radiometric correction, geometric correction, and noise removal, to generate a standardized image set, enabling the comparability of multi-source data in the spectral, geometric, and temporal dimensions; S13. Apply a data registration algorithm to spatially align remote sensing data with different resolutions and multi-temporal phases. Use the SIFT algorithm to perform geometric registration on the images, generating a remote sensing data set across resolutions and multi-temporal phases; S14. Extract multi-modal features of the remote sensing data, including spectral features, texture features, and geometric structure features. The spectral features are calculated using the Normalized Difference Vegetation Index (NDVI) and the Enhanced Vegetation Index (EVI): where NIR represents the reflectance of the near-infrared band, R represents the reflectance of the red band, and B represents the reflectance of the blue band; S15. Extract the texture features of the images. Calculate the contrast, homogeneity, and energy metrics of the images using the gray-level co-occurrence matrix, generating a multi-dimensional texture feature matrix; S16. Extract the geometric structure features of the remote sensing images. Combine the watershed segmentation algorithm and the edge detection algorithm to identify the grassland terrain boundaries and morphological structures, generating a geometric feature description set related to the spatial distribution; S17. Construct a high-dimensional feature matrix representing the state of the grassland ecosystem, and fuse the extracted spectral features, texture features, and geometric structure features into a joint feature vector.

3. The remote sensing monitoring method of grassland biomass according to claim 1, characterized in that, The specific steps of S2 are as follows: S21. Use the multi-temporal remote sensing data in the high-dimensional feature matrix to establish a time series decomposition model, and extract the seasonal and trend change features through the weighted moving average method: where T(t) is the trend component of the time series, X(t - i) is the observation value at the previous i moments, W(i) is the weighting factor, and n is the length of the moving window; S22. Combine the spatial heterogeneity clustering method, and perform clustering processing on the spatial features in the high-dimensional feature matrix through the density-based spatial clustering algorithm to generate spatial partitions within the grassland area; S23. Analyze the dynamic features of each spatial partition, and construct a dynamic change model of grassland biomass with time and space dimensions as the core: M(x, t) = F(S(x), T(t), P(x, t)); where M(x, t) represents the dynamic features at position x and time t, S(x) is the spatial feature, T(t) is the time series feature, and P(x, t) is the external influencing factor; S24. Through the adaptive time window method, dynamically update the spatio-temporal correlation parameters within the partition, and generate a grassland dynamic feature description set according to the seasonal fluctuations and heterogeneous distribution changes of grassland biomass.

4. A remote sensing monitoring method for grassland biomass according to claim 1, characterized in that, The specific steps of S3 are as follows: S31. Divide the dynamic change feature description set into a training data set and a validation data set. Through normalization processing, map the spatio-temporal correlation parameters and external environmental variables to a unified numerical range to adapt to the input requirements of the deep neural network; S32. Construct a multi-layer deep neural network structure, including an input layer, a hidden layer, and an output layer. The input layer receives the key features in the dynamic change feature description set, the hidden layer realizes non-linear feature transformation through the activation function, and the output layer generates the predicted value of grassland biomass: y = f(W3 · g(W2 · g(W1 · x + b1) + b2) + b3); Among them, x is the input feature vector, W1, W2, and W3 are weight matrices, b1, b2, and b3 are bias vectors, g(·) is the activation function, and f(·) is the activation function of the output layer; S33. Combine the multi-task optimization algorithm to define the loss function for the grassland biomass prediction task. The loss function consists of the mean squared error and the uncertainty quantification error: L = α·MSE + β·UQE; where α and β are task weight parameters, UQE is the uncertainty measure of the predicted value; S34. Use the backpropagation algorithm to optimize the network weights and bias parameters, and adjust the network structure according to the gradient update principle of the loss function to gradually minimize the prediction error; S35. Verify the optimized model, and calculate the accuracy indicators of the prediction results using the validation dataset, including the correlation coefficient and the relative error, to make the model applicable under the grassland heterogeneity distribution and multi-source data characteristics; S36. Take the deep neural network model that has completed training and validation as the final non-linear biomass prediction model to generate the estimated results of the spatial distribution of grassland biomass.

5. A remote sensing monitoring method for grassland biomass according to claim 1, characterized in that The specific steps of S5 are as follows: S51. Integrate the partition estimation results with multi-modal remote sensing data, analyze the error distribution of each data source through the uncertainty propagation model, and use the error distribution function to describe the random characteristics of the error sources: where e(x) is the error component, and μ e is the error mean, and is the error variance; S52. Construct an uncertainty propagation formula to calculate the uncertainty measure of the partition biomass estimation results, and describe the uncertainty range through the propagation relationship of the errors: Among them, U(y) is the output uncertainty, y is the biomass estimation value, and x i is the input feature variable, and σ i is the standard deviation of the variable x i . S53. Use the uncertainty quantification optimization algorithm to adjust the estimation model parameters based on the optimization objective function: min[L(y) + γ·U(y)]; Among them, L(y) is the error loss function between the estimated value and the actual observed value, and γ is the uncertainty weight parameter; S54. Verify the optimized biomass estimation results, analyze the mean squared error and standardized residual distribution within the region, and verify whether the uncertainty level of the optimization results meets the predetermined threshold conditions; S55. Output the optimized grassland biomass estimation results, including the estimated value and its corresponding uncertainty range.

6. The remote sensing monitoring method of grassland biomass according to claim 1, characterized in that, The specific steps of S6 are as follows: S61. Collect multi-temporal satellite remote sensing data covering the target area to obtain the macroscopic dynamic change information of the grassland, and combine the image data obtained by the unmanned aerial vehicle to generate a remote sensing dataset across time and space scales; S62. Perform spatial registration processing on the cross-scale data, and use the multi-scale registration algorithm to align the detailed information of the unmanned aerial vehicle images with the macroscopic coverage data of the satellite images. The registered data is represented by the joint resolution as: D(x,t) = F merge (D satellite (x,t), D UAV (x)); Among them, D(x,t) is the spatio-temporal joint remote sensing data, D satellite (x,t) is the satellite data, D UAV (x) is the UAV image data, F merge is the data fusion function; S63. Use the spatio-temporal scale compensation algorithm to dynamically balance the temporal and spatial resolutions D(x,t) in the joint remote sensing data to make the consistency of the unmanned aerial vehicle and satellite data in the temporal dimension and the integrity in the spatial dimension; S64. Combine the partition estimation results to optimize the spatio-temporal correlation parameters of the grassland biomass based on the dynamic update model. The updated correlation parameter matrix can be expressed as: P(t, x) = α·P prev (t, x) + β·P new (t, x); Among them, P(t, x) is the optimized correlation parameter matrix, P prev (t, x) is the previous parameter, P new (t, x) is the correlation parameter of the newly added data, and α and β are weight factors; S65. Based on the updated spatio-temporal correlation parameter P(t,x), generate the optimized grassland biomass monitoring results, and at the same time output the dynamic change trend chart to characterize the change law of the grassland biomass across spatio-temporal scales.

7. A remote sensing monitoring method for grassland biomass according to claim 1, characterized in that, The specific steps of S7 are as follows: S71. Map the optimized grassland biomass monitoring data to a spatial grid, and calculate the spatial distribution value of biomass for each grid cell: Among them, B(x, y) is the biomass value of the grid cell (x, y), and b i (x, y) is the original biomass value of the corresponding cell in the monitoring data, and w i is the optimized weight factor; S72. Based on the distribution results, construct a spatial distribution map of grassland biomass, and use interpolation algorithms to fill in the unobserved areas and maintain consistency with the observed data; S73. Extract the dynamic change trend of grassland biomass in the time dimension, and calculate the overall change rate for the spatial distribution map at each time series point: where R t is the overall change rate at the t-th moment, B t (x i , y i ) and B t-1 (x i , y i ) are the biomass values at the current moment and the previous moment, respectively; S74. Conduct statistical analysis on the dynamic change trend data, generate distribution maps of biomass growth rate and decline rate within the region, and record the abnormal change areas simultaneously; S75. Integrate the spatial distribution map and the results of dynamic change trend analysis to generate a comprehensive assessment report of the grassland ecosystem.

8. A remote sensing monitoring system for grassland biomass, characterized in that, It includes the following modules: Remote sensing data acquisition module: Collect multi-modal remote sensing data, including UAV images, multi-temporal satellite remote sensing data, and other auxiliary remote sensing data; Data preprocessing module: Perform radiometric correction, geometric correction, and noise removal on the remote sensing data to generate a standardized remote sensing image dataset; Feature extraction module: Extract spectral features, texture features, and geometric structure features from the standardized remote sensing data, and construct a high-dimensional feature matrix representing the state of the grassland ecosystem; Dynamic analysis module: Design a dynamic spatio-temporal feature analysis framework based on the high-dimensional feature matrix, and use time series change decomposition and spatial heterogeneity clustering methods to generate spatio-temporal correlation parameters of grassland biomass; Prediction modeling module: Combine deep learning algorithms and dynamic feature description sets to train a non-linear biomass prediction model to achieve accurate estimation of grassland biomass; Zoning optimization module: Through the joint analysis of images and prediction model results, use region segmentation technology to perform hierarchical processing on heterogeneous grassland areas, and generate refined zonal estimation results of grassland biomass; The zoning optimization module specifically includes: Perform spatial overlap analysis on the grassland biomass distribution results generated by the prediction model and the remote sensing image to calibrate the spatial distribution consistency between the prediction results and the image; Based on the calibrated data, use region segmentation technology to segment heterogeneous grassland areas, and adopt the watershed algorithm to generate initial segmentation regions according to the spectral differences and terrain change characteristics of the image; Optimize the initial segmentation regions, combine edge detection methods and morphological operations to eliminate over-segmented and under-segmented regions, and the optimized segmentation regions are represented in vector form as: R i = {x | x ∈ Ω, g(x) > τ i}; where R i is the i-th segmented region, x is the pixel, Ω is the global research region, g(x) is the pixel feature function within the region, and τ i is the segmentation threshold; Combine the segmentation regions, analyze the biomass distribution characteristics within each region, calculate the mean and standard deviation of biomass within the region, and generate zonal estimation parameters: Among them, μ i is the biomass mean of the i-th segmented region, σ i is the standard deviation, y(x) is the predicted biomass value of pixel point x, |R i | is the number of pixels in the region; Jointly generate a refined zonal estimation map of grassland biomass with the zonal estimation parameters and the prediction model results, record the biomass distribution and statistical information of each region, and use it to capture the complex local change characteristics of the grassland; Uncertainty analysis module: Evaluate the error sources of the monitoring data through uncertainty propagation analysis and quantization optimization algorithms, and generate optimized biomass estimation results; Collaborative monitoring module: Integrate cross-scale spatio-temporal data through the collaborative monitoring mechanism of UAVs and satellite remote sensing, dynamically update spatio-temporal correlation parameters, and optimize grassland biomass monitoring results; Assessment and Visualization Module: Generate the spatial distribution map of grassland biomass and the analysis results of dynamic change trends, and provide a comprehensive assessment report on the grassland ecosystem.

Images