Method and system for monitoring grassland shrubs based on remote sensing of unmanned aerial vehicle

Abstract

The invention relates to the technical field of grassland ecological monitoring and remote sensing application, in particular to a method and system for monitoring grassland shrubs based on unmanned aerial vehicle remote sensing, and the method comprises the steps: determining a fluctuation coefficient representing the physiological stability of the shrubs based on an unmanned aerial vehicle multispectral image and ground actual measurement data; determining fractal dimensions representing boundaries of shrub communities on the basis of multiple high-resolution orthoimages of the unmanned aerial vehicle; based on the unmanned aerial vehicle multispectral image and ground actual measurement data, determining a soil moisture spatial heterogeneity entropy representing the shrub local hydrological environment reconstruction intensity; solving an invasion and occupation risk index; predicting to obtain a state transition probability; obtaining a system instability factor; judging the relationship between the system instability factor and a preset intervention threshold; if the system instability factor is greater than the preset intervention threshold, generating an intervention instruction; if not, generating a maintenance instruction; according to the method, the practical guiding significance of a prediction result is ensured, and the accuracy and robustness of an evaluation system are integrally improved.

Description

Technical Field

[0001] This invention relates to the field of grassland ecological monitoring and remote sensing application technology, specifically a method and system for monitoring grassland shrubs based on UAV remote sensing. Background Technology

[0002] In current grassland ecosystem management, monitoring of shrub expansion generally relies on static analysis of single remote sensing images;

[0003] This approach is limited to describing the macroscopic morphology of shrubs, failing to comprehensively assess the inherent physiological stability of shrubs, their interaction with the local hydrological environment, and neglecting key latent driving factors such as microtopography and underground mycorrhizal networks. This static analysis method based on a single dimension results in insufficient monitoring depth, and the assessment results often lag behind the actual ecological processes, enabling only passive responses. Its core flaw lies in the lack of forward-looking early warning capabilities for the risks of shrub expansion, failing to provide scientific decision support for formulating precise and proactive intervention strategies. Therefore, how to construct a comprehensive monitoring and decision-making system that integrates multi-source data, can dynamically assess current risks, and accurately predict future evolutionary trends has become an urgent technical problem to be solved. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for monitoring grassland shrubs, aiming to overcome the shortcomings of existing technologies that rely solely on single static data, have limited monitoring dimensions, provide delayed assessment results, and lack forward-looking early warning capabilities. This allows for a comprehensive and dynamic assessment of the current expansion risk of shrubs and accurate prediction of their future evolutionary trends, providing decision support for proactive and scientific ecological intervention management. Specifically, the technical solution of this invention is as follows:

[0005] A method for monitoring grasslands and shrublands based on UAV remote sensing includes:

[0006] S1, Based on UAV multispectral imagery and ground-based measured data, the fluctuation coefficient characterizing the physiological stability of shrubs was determined;

[0007] S2, Based on UAV multi-resolution orthophotos, determine the fractal dimension that characterizes the geometric complexity of shrub community boundaries;

[0008] S3. Based on UAV multispectral imagery and ground-based measured data, the spatial heterogeneity entropy of soil moisture, which characterizes the intensity of local hydrological environment modification in shrublands, was determined.

[0009] S4 combines the fluctuation coefficient, fractal dimension, and spatial heterogeneity entropy of soil moisture, and normalizes them according to the preset reference benchmark value. Then, it calculates the encroachment risk index through the preset linear weighted model.

[0010] S5, based on the analysis of a high-precision digital elevation model, calculates the local gradient of the terrain humidity index;

[0011] S6. Based on the gene sequence analysis of soil samples, the mycorrhizal network connectivity index is calculated.

[0012] S7 combines the trespass risk index, the local gradient of the topographic humidity index, and the mycorrhizal network connectivity index, and uses a pre-defined convolutional neural network model to predict the state transition probability.

[0013] S8, the system instability factor is obtained by weighted summation of the encroachment risk index and the state transition probability;

[0014] S9, Determine the relationship between the system instability factor and the preset intervention threshold:

[0015] If the system instability factor is greater than the preset intervention threshold, an intervention instruction is generated to initiate a preset level of intervention measures.

[0016] If the system instability factor is less than or equal to the preset intervention threshold, a maintenance instruction is generated to maintain the current monitoring state.

[0017] Optional, linearly weighted models include:

[0018] The fractal dimension and the spatial heterogeneity entropy of soil moisture were treated as positively correlated terms, while the fluctuation coefficient was treated as a negatively correlated term.

[0019] Optionally, the reference baseline value is obtained by measuring and statistically analyzing native shrub communities that have been identified as long-term stable within the study area.

[0020] Optionally, before making predictions using a convolutional neural network model, the following steps are also included:

[0021] The local gradient of the topographic humidity index and the mycorrhizal network connectivity index are normalized based on the maximum and minimum values ​​in the model training dataset.

[0022] Optionally, the convolutional neural network model is trained using a historical dataset that includes normalized input variables and whether patch fusion events occurred in the corresponding period as labels.

[0023] Optionally, a weighted summation of the encroachment risk index and the state transition probability can be performed, including:

[0024] Based on the preset management objectives, weighting coefficients are assigned to the encroachment risk index and the state transition probability.

[0025] Optionally, intervention instructions are used to trigger blocking interventions.

[0026] A system for monitoring grasslands and shrublands based on UAV remote sensing includes:

[0027] The physiological state analysis unit is used to determine the fluctuation coefficient characterizing the physiological stability of shrubs based on UAV multispectral imagery and ground-based measured data.

[0028] Boundary morphology analysis unit is used to determine the fractal dimension, which characterizes the geometric complexity of shrub community boundaries, based on UAV multispectral imagery.

[0029] The environmental interaction analysis unit is used to determine the spatial heterogeneity entropy of soil moisture, which characterizes the intensity of local hydrological environment modification in shrublands, based on UAV multispectral imagery and ground-based measured data.

[0030] The comprehensive risk assessment unit combines the fluctuation coefficient, fractal dimension, and spatial heterogeneity entropy of soil moisture, and normalizes them according to a preset reference benchmark value. Through a preset linear weighted model, it calculates the encroachment risk index.

[0031] The latent factor solution unit is used to calculate the local gradient of the topographic moisture index based on the analysis of the high-precision digital elevation model, and to calculate the mycorrhizal network connectivity index based on the gene sequence analysis of soil samples.

[0032] The future state prediction unit is used to combine the encroachment risk index, the local gradient of the topographic humidity index, and the mycorrhizal network connectivity index to predict the state transition probability through a pre-set convolutional neural network model.

[0033] The system instability decision unit is used to calculate the system instability factor by weighted summation of the invasion risk index and the state transition probability.

[0034] The instruction generation unit is used to determine the relationship between the system instability factor and the preset intervention threshold. If the system instability factor is greater than the preset intervention threshold, an intervention instruction is generated. If the system instability factor is less than or equal to the preset intervention threshold, a maintenance instruction is generated.

[0035] Compared with the prior art, the present invention has the following beneficial effects:

[0036] 1. This invention achieves comprehensive and in-depth integration of monitoring dimensions. It not only analyzes the external morphology of shrubs, but also integrates their internal physiological stability, ability to modify the local hydrological environment, and previously neglected implicit driving factors such as micro-topography and underground mycorrhizal networks. It constructs a multi-dimensional comprehensive evaluation system from micro-physiology to macro-environmental interaction, which significantly improves the depth and breadth of monitoring.

[0037] 2. This invention realizes the transformation from passive response to proactive early warning. By introducing a prediction model based on machine learning, this method can not only assess the current risk of shrub encroachment, but also predict the probability of its future evolution into a critical expansion state. This enables managers to identify and pay attention to shrub patches with high expansion potential in advance, realizing a fundamental shift from post-event remediation to pre-event prevention.

[0038] 3. This invention constructs a closed-loop management system from monitoring to decision-making. By weighted integration of current risks and future trends, a top-level decision indicator is formed, and intervention or maintenance instructions are automatically generated based on this indicator, directly triggering hierarchical management measures. This opens up the complete link from data analysis to management action, providing operable intelligent decision support for precise and efficient ecosystem management.

[0039] 4. This invention ensures the scientific rigor and reliability of the risk assessment results. The risk assessment model is constructed based on clear ecological logic, and the reference benchmark values ​​used for its normalization process are derived from measured statistics of long-term stable communities within the region, thus possessing objectivity and regional adaptability. Simultaneously, the prediction model is directly linked to key ecological events, ensuring the practical guiding significance of the prediction results and improving the overall accuracy and robustness of the assessment system. Attached Figure Description

[0040] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0041] Figure 1 This is a flowchart of the method of the present invention;

[0042] Figure 2 This is a structural diagram of the system of the present invention. Detailed Implementation

[0043] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0044] Example 1:

[0045] Please see Figure 1 A method for monitoring grasslands and shrublands based on UAV remote sensing includes:

[0046] S1, Based on UAV multispectral imagery and ground-based measured data, the fluctuation coefficient characterizing the physiological stability of shrubs was determined;

[0047] S2, Based on UAV multi-resolution orthophotos, determine the fractal dimension that characterizes the geometric complexity of shrub community boundaries;

[0048] S3. Based on UAV multispectral imagery and ground-based measured data, the spatial heterogeneity entropy of soil moisture, which characterizes the intensity of local hydrological environment modification in shrublands, was determined.

[0049] S4 combines the fluctuation coefficient, fractal dimension, and spatial heterogeneity entropy of soil moisture, and normalizes them according to the preset reference benchmark value. Then, it calculates the encroachment risk index through the preset linear weighted model.

[0050] S5, based on the analysis of a high-precision digital elevation model, calculates the local gradient of the terrain humidity index;

[0051] S6. Based on the gene sequence analysis of soil samples, the mycorrhizal network connectivity index is calculated.

[0052] S7 combines the trespass risk index, the local gradient of the topographic humidity index, and the mycorrhizal network connectivity index, and uses a pre-defined convolutional neural network model to predict the state transition probability.

[0053] S8, the system instability factor is obtained by weighted summation of the encroachment risk index and the state transition probability;

[0054] S9, Determine the relationship between the system instability factor and the preset intervention threshold:

[0055] If the system instability factor is greater than the preset intervention threshold, an intervention instruction is generated to initiate a preset level of intervention measures.

[0056] If the system instability factor is less than or equal to the preset intervention threshold, a maintenance instruction is generated to maintain the current monitoring state.

[0057] This embodiment provides a method for monitoring grassland and shrubland based on UAV remote sensing, aiming to establish a complete closed-loop management process from data collection, risk assessment, future prediction to decision intervention, so as to achieve accurate, dynamic and forward-looking monitoring of grassland and shrubland expansion;

[0058] The specific steps of this method include:

[0059] To quantify the stability of energy utilization efficiency of individual shrubs over a complete growing season, it is necessary to determine the fluctuation coefficient characterizing the physiological stability of shrubs based on UAV multispectral imagery and ground-based measured data. In the context of this invention, physiological stability is considered a key indicator of whether a shrub is in a resting state; an unstable physiological state often precedes its entry into an active expansion phase. Specifically, the fluctuation coefficient here refers to the fluctuation coefficient of the canopy photosynthetically active radiation absorptivity. Its function is to quantitatively assess the stability of shrub canopy capture and utilize solar energy throughout the entire observation period. To achieve this, data is collected by a drone equipped with a multispectral sensor at a preset time frequency throughout the entire growing season in the target area. The acquired multispectral images undergo preprocessing such as radiometric calibration, atmospheric correction, and geometric correction, and vegetation index algorithms are used, for example, by establishing the Normalized Difference Vegetation Index (NDVI) and ground-measured data. The linear regression model between them was inverted to obtain the results for each observation time point. Photosynthetically active radiation absorptivity instantaneous value The volatility coefficient is calculated using the following formula. That is, the coefficient of variation of the time series of photosynthetically active radiation absorptivity in the canopy:

[0060] in: It is the dimensionless fluctuation coefficient calculated in this step; Using UAV multispectral imagery at specific times Instantaneous obtained by inversion value; It refers to all of the entire observation period The value is obtained by averaging. Arithmetic mean; It is based on the total number of time-series sampling points pre-set according to the length of the complete growing season of the target area and the frequency of UAV observation;

[0061] To quantify the morphological characteristics of shrub patch boundaries at the mesoscopic structural level, it is necessary to determine the fractal dimension, which characterizes the geometric complexity of shrub community boundaries, based on UAV multispectral imagery. This invention employs fractal dimension because it can more profoundly reveal the degree of boundary fragmentation and tortuosity, and this morphological complexity is directly related to the expansion vitality of shrub communities. This fractal dimension specifically refers to the boundary fractal dimension calculated using the box-counting method. Its function is to quantitatively describe the geometric complexity of shrub patch boundaries. To achieve this, high-resolution orthophotos acquired by UAVs are processed, and the vector boundaries of shrub patches are accurately extracted using supervised classification or object-oriented methods. The box-counting method is used to calculate these boundaries, and the theoretical formula is as follows:

[0062] In practical calculations, this value is expressed in a log-log coordinate system for a series of discrete grid sizes. and the minimum number of squares required to completely cover the boundary. The result obtained by linear regression fitting is the negative of the slope of the line. In the formula, It is the dimensionless fractal dimension calculated in this step, and its theoretical value is between 1 and 2; The range of values ​​is determined by the spatial resolution of the UAV imagery; It is achieved through image processing algorithms in a given... The minimum number of squares required to cover the boundary as determined by the following statistics;

[0063] To quantify the ability of shrubs to redistribute water in the surrounding microenvironment from a macro-environmental interaction perspective, it is necessary to determine the spatial heterogeneity entropy of soil moisture, characterizing the intensity of local hydrological environment modification by shrubs, based on UAV multispectral imagery and ground-based measured data. The root system and canopy of shrubs can form a resource-enriching "fertility island" effect; this step measures the intensity of this effect using information entropy. This spatial heterogeneity entropy of soil moisture refers to the entropy value calculated using Shannon entropy theory. Its function is to quantitatively assess the complexity of the spatial distribution of soil moisture in the shrub-grass transition zone. To achieve this, a ring-shaped area extending outward from the edge of shrub patches at a specific distance is defined as the shrub-grass transition zone. High-density soil moisture content data in this area is obtained through UAV thermal infrared remote sensing inversion or by deploying a soil moisture sensor network within the transition zone. The dynamic range of soil moisture is then divided into... Each of the discrete humidity levels is statistically analyzed and falls into the first category. The probability of each level Calculated using the following formula :

[0064] in: This is the spatial heterogeneity entropy of soil moisture calculated in this step, and its unit is bits; It is the number of humidity levels preset based on the dynamic range of soil moisture in the study area; This data was obtained by frequency statistics of all soil moisture measurements within the ecotone. The probability of each humidity level;

[0065] To construct a comprehensive and quantifiable risk assessment index, it is necessary to combine the volatility coefficient, fractal dimension, and soil moisture spatial heterogeneity entropy, and normalize them according to a preset reference benchmark value. Then, through a preset linear weighted model, the encroachment risk index is calculated. It is a dimensionless index for assessing the current risk of shrub patch expansion. Its function is to unify multi-source heterogeneous indicators under an evaluation system, providing a direct basis for risk level classification. This step is achieved through a linear weighted model, which will... , , Weighted combinations are performed; to eliminate dimensional differences, a preset reference value representing the stable state of the system is used. , , After normalization, the model takes the following form:

[0066] in: The first step is calculated from this step. Risk index of shrub patch encroachment; These are the weighting coefficients, which sum to 1 and can be determined through sensitivity analysis based on historical data or expert scoring.

[0067] To incorporate micro-topography as a latent driving factor, it is necessary to analyze and calculate the local gradient of the topographic humidity index (TWI) based on a high-precision digital elevation model. Here, the local gradient of the topographic humidity index refers to the spatial rate of change of the TWI. Its function is to quantify the driving force of micro-topography on water redistribution; this value is obtained by hydrological analysis calculation of a high-precision digital elevation model (DEM).

[0068] To incorporate the latent factor of underground biological networks, it is necessary to calculate the mycorrhizal network connectivity index based on gene sequence analysis of soil samples. Here, the mycorrhizal network connectivity index refers to a biological index characterizing the developmental degree and connectivity of underground arbuscular mycorrhizal fungal networks. Its function is to reveal the potential ability of shrubs to acquire and share resources through underground mycorrhizal networks; this value is obtained by collecting soil samples from the target area and performing gene sequence analysis on them, and can be specifically defined as the ratio of the abundance of gene sequences of dominant AMF species in the target area to the abundance of the background in the area.

[0069] To predict future system evolution trends from current risk assessment, it is necessary to combine the invasive risk index, the local gradient of the topographic moisture index, and the mycorrhizal network connectivity index, and use a pre-defined convolutional neural network model to predict the state transition probability; this state transition probability refers to the shrub patch... In the next moment The probability of the current state evolving into a critical expansion state Its function is to quantitatively predict the risk of a catastrophic phase transition in the future; this probability is calculated through a pre-defined convolutional neural network (CNN) model, with the invasion risk index at the current moment as its input. And two normalized latent factors and Its nonlinear mapping relationship is expressed as:

[0070] in: This represents a trained CNN model. and These are the normalized latent factors;

[0071] To construct a top-level decision-making indicator that integrates current risks and future trends, it is necessary to perform a weighted summation of the encroachment risk index and the state transition probability to obtain the system instability factor; this system instability factor is a comprehensive dimensionless indicator. It does this by analyzing the current risk index of encroachment. With future state transition probability The weighted sum is used as the final decision-making basis for triggering tiered intervention measures. The calculation formula is:

[0072] in and These are weighting coefficients that can be adjusted according to management objectives; in a preferred embodiment, to more comprehensively assess system risk, the influence of spatial neighborhood can be introduced into the calculation of the system instability factor, for example, the modified system instability factor. It can be represented as: ,in Adjacent plaques The risk index of encroachment, and It is related to plaque and The spatial weighting coefficients are inversely proportional to the distance between them, which makes the model's evaluation results more consistent with the spatial clustering risk in reality;

[0073] Make decisions and execute actions to determine the relationship between system instability factors and preset intervention thresholds; if the calculated... Greater than a preset intervention threshold The system determines that the shrub patch is at an unacceptable risk of instability and then generates an intervention command to initiate a preset level of intervention measures; this threshold Risk tolerance can be set based on historical data analysis and management; conversely, if... Less than or equal to Then, a maintenance command is generated to maintain the current monitoring status;

[0074] Through the above steps, this invention constructs a multi-dimensional, multi-scale integrated monitoring and decision-making framework. Compared with existing technologies that rely solely on static analysis of single remote sensing images, this invention integrates physiological, structural, and environmental interactions, as well as implicit topographic and biological factors, thereby improving the depth and accuracy of monitoring. By introducing state transition probability prediction, monitoring is transformed from a passive response to an active early warning system. Finally, through system instability factors and intervention commands, a closed loop from monitoring to management is achieved, providing scientific and operable technical support for grassland ecosystem management.

[0075] Example 2:

[0076] Linear weighted models, including:

[0077] The fractal dimension and the spatial heterogeneity entropy of soil moisture were treated as positively correlated terms, while the fluctuation coefficient was treated as a negatively correlated term.

[0078] This embodiment further defines the specific technical solution of the linear weighted model based on the method described in Embodiment 1;

[0079] In this model In the middle, the fractal dimension and the spatial heterogeneity entropy of soil moisture Treated as a positive correlation term; its underlying logic lies in higher fractal dimensions. This implies that shrub boundaries are more irregular, associated with active outward expansion, and increase the risk of encroachment; similarly, higher spatial heterogeneity entropy of soil moisture. This indicates that the shrubland has successfully transformed the local hydrological environment, forming a resource-rich fertile island. This strong environmental control is the basis for its further expansion, but it also increases the risks.

[0080] At the same time, the volatility coefficient It is treated as a negatively correlated term; the technical consideration is that... Characterizing physiological stability, a lower The value indicates that the energy utilization efficiency of the shrub is stable and in a steady state, corresponding to a lower risk of expansion; therefore, in the model... The negative sign design accurately reflects the contribution of physiological stability to the inhibition risk; by clearly defining the positive and negative correlations of each parameter, this linear weighted model establishes an assessment logic with clear ecological significance, ensuring the accuracy and reliability of the risk assessment results.

[0081] Example 3:

[0082] The reference baseline value was obtained by measuring and statistically analyzing native shrub communities that were identified as long-term stable within the study area.

[0083] This embodiment further limits the source of the reference benchmark value in the method described in Embodiment 1 to ensure the objectivity of the normalization process;

[0084] In terms of parameters , , The reference value used for normalization , , The data were obtained by measuring and statistically analyzing native shrub communities within the study area that were identified as long-term stable. To achieve this, ecological surveys and historical remote sensing image analysis were used to identify and delineate multiple native shrub communities within the study area that had not shown significant changes in area and morphology over a long period and were generally considered to be in a stable successional stage. Using the same method as described above, these stable samples were measured multiple times, and their values ​​were calculated. , and Value; perform statistical analysis on the measurement results of all stable samples, such as taking their arithmetic mean or median, to obtain the final reference baseline value;

[0085] This method establishes a data-driven, ecologically significant, stable reference system, making the reference benchmark highly adaptable to different regions. This improves the objectivity of the normalization process and the universality of the encroachment risk index. To ensure the robustness of the model, after determining the final reference benchmark, validity verification is required. If any reference benchmark value falls below a preset minimum threshold to prevent computational instability... Then the reference value should be set to the minimum threshold. This can avoid abnormal fluctuations in the calculation results of the encroachment risk index due to an excessively small denominator.

[0086] Example 4:

[0087] Before making predictions using a convolutional neural network model, the following steps are also included:

[0088] The local gradient of the topographic humidity index and the mycorrhizal network connectivity index are normalized based on the maximum and minimum values ​​in the model training dataset.

[0089] The convolutional neural network model is trained using a historical dataset that includes normalized input variables and whether patch fusion events occurred in the corresponding period as labels.

[0090] This embodiment is a further optimization of the input data processing and training method of the convolutional neural network model based on the method described in Embodiment 1;

[0091] In a preferred embodiment of the present invention, the specific structure of the convolutional neural network model can be configured as follows: an input layer, used to receive data from the invasion risk index. Local gradient of normalized topographic humidity index and mycorrhizal network connectivity index The system constructs a three-dimensional feature vector; this is followed by two one-dimensional convolutional layers. The first convolutional layer has 16 kernels of size 2, and the second convolutional layer has 32 kernels of size 2, both using ReLU as the activation function. A global max-pooling layer is then connected to extract the features. Finally, a fully connected output layer, containing a single neuron and using the Sigmoid activation function, outputs a value between 0 and 1, representing the state transition probability. ;

[0092] Before making predictions using a convolutional neural network model, the local gradient of the topographic humidity index is also included. and mycorrhizal network connectivity index Based on the maximum and minimum values ​​in the model training dataset, normalization is performed. This step aims to eliminate differences in the dimensions and numerical ranges of different latent variables input into the CNN model, preventing variables with larger numerical ranges from dominating the model training process. The maximum-minimum normalization method is used, and its calculation formula is as follows:

[0093] as well as

[0094] in: and It is a dimensionless input variable whose value is scaled to between 0 and 1 after normalization; and It is all samples in the model training dataset. The maximum and minimum values; and It is all samples in the model training dataset. The system needs to perform a robustness check during normalization. If the maximum value of an input variable equals its minimum value in the entire training dataset (i.e., the denominator is zero), then the normalized results of all variables in that dataset will be normalized. or The constant value is uniformly set to a preset constant, such as 0.5, to avoid calculation errors and ensure the stability of the model;

[0095] Furthermore, this convolutional neural network model is trained using a historical dataset labeled with normalized input variables and whether patch fusion events occurred at the corresponding time points. Patch fusion events refer to the phenomenon where two or more independent shrub patches merge into a larger patch due to expansion, which is a key node in the phase transition of the system. The training dataset is constructed as follows: historical data from the past few years in the study area are collected, and for each shrub patch at a historical time point, its input variables at that time are calculated. , and Meanwhile, by comparing the remote sensing images at this time point and the next time point, it is determined whether the patch participated in the patch fusion event during this period; if fusion occurs, the label of the sample is recorded as 1; if it does not occur, the label is recorded as 0; a large number of such sample pairs are collected and used for supervised learning training of the CNN model.

[0096] Normalization ensures that the model can fairly learn the importance of each input feature, improving training efficiency and prediction accuracy. Simultaneously, by directly linking the model's prediction objective to the key ecological event of patch fusion, the model's output... It is a direct prediction of the probability of a specific, observable, and ecologically significant critical event, which greatly enhances the practical value and decision-making guidance significance of the prediction model.

[0097] Example 5:

[0098] The weighted sum of the embezzlement risk index and the state transition probability includes:

[0099] Based on the preset management objectives, weighting coefficients are assigned to the encroachment risk index and the state transition probability.

[0100] This embodiment is a further refinement of the technical content of the weighted summation of the invasion risk index and the state transition probability in the method described in Embodiment 1;

[0101] In calculating the instability factor of the system At that time, weighting coefficient and The allocation is flexibly set based on preset management objectives; these preset management objectives refer to the strategic direction set by ecosystem managers based on specific conservation tasks, management budgets, risk preferences, and other factors. For example, if the management objective is prevention-first, aiming to identify and intervene in patches with high potential for future expansion as early as possible, then the probability of state transition representing future trends can be used. Assign a larger weighting factor Conversely, if management resources are limited and the goal is to prioritize addressing patches that have already shown clear signs of expansion, then an encroachment risk index can be used to represent the current state. Assign a larger weighting factor By introducing this weight adjustment mechanism, the system instability factor It is no longer a rigid technical output, but a decision-making tool that can be dynamically coupled with actual management needs, enhancing the applicability and flexibility of this method in real-world management scenarios.

[0102] Example 6:

[0103] Intervention instructions are used to trigger blocking interventions.

[0104] This embodiment further illustrates the specific types of intervention measures triggered by the intervention command generated in the method described in Embodiment 1; when the system instability factor Exceeding the preset intervention threshold The generated intervention command is used to trigger a disruptive intervention. This disruptive intervention is a proactive physical or biological management measure aimed at directly interrupting or reversing the shrub encroachment process, such as selective physical removal of high-risk patches, planned burning, or hydrological control. The disruptive intervention can be preset to different levels based on its intensity and scope. For example, a Level 1 intervention could be set to isolate or biologically control the risky patches; a Level 2 intervention to selective physical removal; and a Level 3 intervention to planned burning or regional hydrological control. The intervention command may include information such as the geographical coordinates of the high-risk patches, the intervention level, and the type of recommended measures. The intervention level is determined by the system instability factor. Exceeding the intervention threshold The extent to which this is determined; this embodiment clarifies the ultimate purpose and execution method of the intervention instructions, and constructs a complete closed-loop management system from monitoring, evaluation, prediction to decision-making and action, ensuring that technological outputs can be directly transformed into effective ecological management actions.

[0105] Example 7:

[0106] Please see Figure 2 A system for monitoring grasslands and shrublands based on UAV remote sensing includes:

[0107] The physiological state analysis unit is used to determine the fluctuation coefficient characterizing the physiological stability of shrubs based on UAV multispectral imagery and ground-based measured data.

[0108] Boundary morphology analysis unit is used to determine the fractal dimension, which characterizes the geometric complexity of shrub community boundaries, based on UAV multispectral imagery.

[0109] The environmental interaction analysis unit is used to determine the spatial heterogeneity entropy of soil moisture, which characterizes the intensity of local hydrological environment modification in shrublands, based on UAV multispectral imagery and ground-based measured data.

[0110] The comprehensive risk assessment unit combines the fluctuation coefficient, fractal dimension, and spatial heterogeneity entropy of soil moisture, and normalizes them according to a preset reference benchmark value. Through a preset linear weighted model, it calculates the encroachment risk index.

[0111] The latent factor solution unit is used to calculate the local gradient of the topographic moisture index based on the analysis of the high-precision digital elevation model, and to calculate the mycorrhizal network connectivity index based on the gene sequence analysis of soil samples.

[0112] The future state prediction unit is used to combine the encroachment risk index, the local gradient of the topographic humidity index, and the mycorrhizal network connectivity index to predict the state transition probability through a pre-set convolutional neural network model.

[0113] The system instability decision unit is used to calculate the system instability factor by weighted summation of the invasion risk index and the state transition probability.

[0114] The instruction generation unit is used to determine the relationship between the system instability factor and the preset intervention threshold. If the system instability factor is greater than the preset intervention threshold, an intervention instruction is generated. If the system instability factor is less than or equal to the preset intervention threshold, a maintenance instruction is generated.

[0115] This embodiment provides a system for monitoring grasslands and shrublands based on UAV remote sensing. This system is the technical carrier for implementing the aforementioned method, and includes:

[0116] The physiological state analysis unit is configured to receive time-series UAV multispectral imagery and ground-based measured data, and incorporates an fAPAR inversion algorithm and fluctuation coefficient. The calculation module is used to output the data for each shrub patch. value;

[0117] The boundary morphology analysis unit is configured to receive high-resolution UAV orthophotos and includes built-in image segmentation, edge extraction, and box-counting dimension calculation modules to calculate and output the fractal dimension of each patch boundary. ;

[0118] The environmental interactive analysis unit is configured to receive soil moisture data and includes built-in humidity level classification and Shannon entropy. The calculation module is used to output the cross-zones of each patch. value;

[0119] The integrated risk assessment unit is connected to the aforementioned three units and receives data. , and The system internally stores reference benchmark values ​​and incorporates a linear weighted model for normalization and weighted calculations, ultimately outputting an invasion risk index for each patch. ;

[0120] The latent factor calculation unit comprises two sub-modules, one configured to receive high-precision digital elevation model data to calculate the local gradient of the topographic moisture index. Another configuration is to receive gene sequence analysis data from soil samples to calculate the mycorrhizal network connectivity index. ;

[0121] The future state prediction unit, which is connected to the comprehensive risk assessment unit and the implicit factor calculation unit, receives data. , and As input, its core is a pre-trained convolutional neural network model, the specific structure of which can be as described in Example 4, including a one-dimensional convolutional layer, a pooling layer, and a fully connected output layer, used to output the state transition probability of each patch. ;

[0122] The system instability decision-making unit receives and And according to the preset weighting coefficient By performing a weighted summation, the system instability factor can be calculated. ;

[0123] The instruction generation unit, which is the final output of the system, receives the system instability factor. Compare it with the preset intervention threshold stored internally. Compare the results and generate intervention or maintenance instructions based on the comparison.

[0124] These units are interconnected through data interfaces and internal buses, working together to form an automated and intelligent monitoring and decision-making system. Through modular design, the system clearly divides complex methods and processes into functionally independent units, realizing the integration from data processing to decision command generation, and providing an efficient, reliable, and deployable solution for grassland ecological management.

[0125] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for monitoring grasslands and shrublands based on unmanned aerial vehicle (UAV) remote sensing, characterized in that... ,include: S1, Based on UAV multispectral imagery and ground-based measured data, the fluctuation coefficient characterizing the physiological stability of shrubs was determined; S2, Based on UAV multi-resolution orthophotos, determine the fractal dimension that characterizes the geometric complexity of shrub community boundaries; S3. Based on UAV multispectral imagery and ground-based measured data, the spatial heterogeneity entropy of soil moisture, which characterizes the intensity of local hydrological environment modification in shrublands, was determined. S4 combines the fluctuation coefficient, fractal dimension, and spatial heterogeneity entropy of soil moisture, and normalizes them according to the preset reference benchmark value. Then, it calculates the encroachment risk index through the preset linear weighted model. S5, based on the analysis of a high-precision digital elevation model, calculates the local gradient of the terrain humidity index; S6. Based on the gene sequence analysis of soil samples, the mycorrhizal network connectivity index is calculated. S7 combines the trespass risk index, the local gradient of the topographic humidity index, and the mycorrhizal network connectivity index, and uses a pre-defined convolutional neural network model to predict the state transition probability. S8, the system instability factor is obtained by weighted summation of the encroachment risk index and the state transition probability; S9, Determine the relationship between the system instability factor and the preset intervention threshold: If the system instability factor is greater than the preset intervention threshold, an intervention instruction is generated to initiate a preset level of intervention measures. If the system instability factor is less than or equal to the preset intervention threshold, a maintenance instruction is generated to maintain the current monitoring state.

2. The method for monitoring grasslands and shrublands based on UAV remote sensing according to claim 1, characterized in that... Linear weighted models include: The fractal dimension and the spatial heterogeneity entropy of soil moisture were treated as positively correlated terms, while the fluctuation coefficient was treated as a negatively correlated term.

3. The method for monitoring grasslands and shrublands based on UAV remote sensing according to claim 1, characterized in that... The reference baseline value was obtained by measuring and statistically analyzing native shrub communities that were identified as long-term stable within the study area.

4. The method for monitoring grasslands and shrublands based on UAV remote sensing according to claim 1, characterized in that... Before making predictions using a convolutional neural network model, the following steps are also included: The local gradient of the topographic humidity index and the mycorrhizal network connectivity index are normalized based on the maximum and minimum values ​​in the model training dataset.

5. A method for monitoring grasslands and shrublands based on UAV remote sensing according to claim 4, characterized in that... The convolutional neural network model is trained using a historical dataset that includes normalized input variables and whether patch fusion events occurred in the corresponding period as labels.

6. The method for monitoring grasslands and shrublands based on UAV remote sensing according to claim 1, characterized in that... The weighted sum of the encroachment risk index and the state transition probability includes: Based on the preset management objectives, weighting coefficients are assigned to the encroachment risk index and the state transition probability.

7. A method for monitoring grasslands and shrublands based on UAV remote sensing according to claim 1, characterized in that... Intervention instructions are used to trigger blocking interventions.

8. A system for monitoring grassland and shrubland based on UAV remote sensing, comprising the method for monitoring grassland and shrubland based on UAV remote sensing as described in any one of claims 1-7, characterized in that... ,include: The physiological state analysis unit is used to determine the fluctuation coefficient characterizing the physiological stability of shrubs based on UAV multispectral imagery and ground-based measured data. Boundary morphology analysis unit is used to determine the fractal dimension, which characterizes the geometric complexity of shrub community boundaries, based on UAV multispectral imagery. The environmental interaction analysis unit is used to determine the spatial heterogeneity entropy of soil moisture, which characterizes the intensity of local hydrological environment modification in shrublands, based on UAV multispectral imagery and ground-based measured data. The comprehensive risk assessment unit combines the fluctuation coefficient, fractal dimension, and spatial heterogeneity entropy of soil moisture, and normalizes them according to a preset reference benchmark value. Through a preset linear weighted model, it calculates the encroachment risk index. The latent factor solution unit is used to calculate the local gradient of the topographic moisture index based on the analysis of the high-precision digital elevation model, and to calculate the mycorrhizal network connectivity index based on the gene sequence analysis of soil samples. The future state prediction unit is used to combine the encroachment risk index, the local gradient of the topographic humidity index, and the mycorrhizal network connectivity index to predict the state transition probability through a pre-set convolutional neural network model. The system instability decision unit is used to calculate the system instability factor by weighted summation of the invasion risk index and the state transition probability. The instruction generation unit is used to determine the relationship between the system instability factor and the preset intervention threshold. If the system instability factor is greater than the preset intervention threshold, an intervention instruction is generated. If the system instability factor is less than or equal to the preset intervention threshold, a maintenance instruction is generated.

Images