Forest vegetation health index construction and diagnosis method based on multi-source remote sensing data

Abstract

The invention discloses a forest vegetation health index construction and diagnosis method based on multi-source remote sensing data, and the method achieves the fusion of a feature level and a decision level of data through selecting sample plots containing different health levels, collecting the multi-source remote sensing data of a satellite, an unmanned plane and a ground sensor, and constructing a multi-source data fusion network. Indexes reflecting forest health are extracted from the fused data, a hierarchical clustering method is adopted for screening and standardization processing, a health index model is constructed, and a machine learning algorithm is utilized for optimizing the weight. And finally, calculating a forest health index based on the optimized model, dividing health levels, and generating a forest health level distribution map. The method is suitable for large-area forest vegetation health diagnosis, has the characteristics of high efficiency, accuracy and comprehensiveness, can remarkably improve the scientificity and rationality of forest health monitoring, and can be expanded and applied to the fields of fire / insect pest early warning, carbon sink evaluation and the like.

Description

Technical Field

[0001] The present invention belongs to the field of forest vegetation health diagnosis, and in particular relates to a forest vegetation health index construction and diagnosis method based on multi-source remote sensing data. Background Art

[0002] Forest vegetation health refers to the state and ability of a forest, as a structure, to maintain its healthy existence and renewal and provide essential ecological services. A healthy forest should possess excellent self-renewal capabilities, be resilient to external disturbances, and be able to meet the reasonable needs of human and economic and social development. Forest vegetation health diagnosis involves assessing the productivity, structural elements, and resistance to disturbances of forest ecosystems through a series of scientific methods and technical means, employing a specific indicator system.

[0003] Traditional forest vegetation health diagnostic methods typically rely on field surveys conducted by trained forestry staff to assess the external health or damage of a forest. However, this approach is highly uncertain because its results rely on the observer's experience and subjective perception. Furthermore, traditional methods are complex in terms of time, cost, and human effort, making them suitable only for small-scale sampling and inadequate for large-scale forest health monitoring.

[0004] With the development of remote sensing technology, it has demonstrated advantages in monitoring vegetation activity, such as macroscopic, rapid, and repeatable observations, and has become an important means of monitoring vegetation growth and health. Currently, researchers are using remote sensing technology to construct the Forest-Vegetation Health Index (FHI). By extracting spectral or structural features from sensor data, FHI objectively, quantitatively, and repeatably obtains forest vegetation health indicators at multiple spatial scales for modeling. However, satellite-based remote sensing still has shortcomings in terms of temporal and spatial resolution and is approaching its feasibility limits. In recent years, drone technology has demonstrated significant value in assessing forest structure, species composition, and health. For example, drone-based lidar can quantify multiple forest structural parameters, and passive drone multispectral imaging technology has also been used to assess forest vegetation health and detect forest disturbance patterns. Furthermore, ground stress wave detection can directly identify cavities within trees, and combined with lidar data, it can form integrated internal and external monitoring.

[0005] Despite this, existing forest vegetation health diagnosis still faces many problems and limitations: The connotation and understanding of forest vegetation health are not unified, the selection of evaluation indicators is one-sided, and the selection of indicators at different evaluation scales is inconsistent. There is a lack of standardized forest vegetation health assessment processes and standardized data preprocessing workflows such as multi-source data fusion, which affects the consistency and operability of satellite remote sensing data, drone sensor data, and ground sensor data. Forest vegetation health assessment models and indices are becoming increasingly diverse, but lack promotion and transplantation, and have not been widely applied in actual forest health assessments. Therefore, a forest vegetation health index construction and diagnosis method based on multi-source remote sensing data is urgently needed to overcome the shortcomings of existing technologies and achieve efficient, accurate, and comprehensive assessments of forest vegetation health. Summary of the Invention

[0006] The present invention proposes a forest vegetation health index construction and diagnosis method based on multi-source remote sensing data to solve the problems existing in the above-mentioned prior art.

[0007] To achieve the above objectives, the present invention provides a method for constructing and diagnosing a forest vegetation health index based on multi-source remote sensing data, comprising the following steps:

[0008] Select forest sample plots that include healthy, sub-healthy, and unhealthy levels, and collect multi-source remote sensing data from the sample plots, including satellite data, drone data, and ground sensor data;

[0009] Constructing a multi-source data fusion network, performing feature extraction and decision-level fusion on the multi-source remote sensing data in sequence to obtain fused data;

[0010] Extracting evaluation indicators reflecting forest health from the fused data, screening the evaluation indicators using a hierarchical clustering method, and standardizing the screened indicators;

[0011] A forest vegetation health index model is constructed based on the screened indicators, the weights of the model are optimized using a machine learning algorithm, and the health grade thresholds are determined using a boxplot method;

[0012] The forest health levels in the area to be diagnosed are divided according to the grading threshold, and a forest health level distribution map is generated.

[0013] Optionally, select forest plots including:

[0014] Select a square plot and arrange shrub and herb plots in diagonal lines within the plot;

[0015] Field measurements were conducted on trees, shrubs and herbs in the sample plots, and data including diameter at breast height, tree height, crown width, species name, number of plants, height, cover and frequency were collected.

[0016] Optionally, the multi-source remote sensing data collected from the sample site includes:

[0017] Obtain macroscopic spectral data including visible light, near infrared and thermal infrared bands through satellite remote sensing;

[0018] Acquire medium-resolution canopy data using drones equipped with multispectral, hyperspectral, and LiDAR sensors;

[0019] The three-dimensional structural parameters of vegetation are obtained through ground-based LiDAR scanners, and wireless sensor networks are deployed to collect temperature and humidity data in the forest microenvironment;

[0020] A three-dimensional stress wave detection device is used to scan the wave velocity inside the trunk of the sample tree to detect the cavity inside the tree.

[0021] Optionally, the obtained fused data includes:

[0022] Design feature extraction networks for different types of data. Use 1D-CNN and LSTM to extract one-dimensional data features, 2D-CNN and LSTM to extract two-dimensional data features, and 3D-CNN and LSTM to extract three-dimensional data features.

[0023] The extracted different data features are spliced ​​and input into a feature fusion network with three fully connected layers;

[0024] The spatiotemporal resolution of multi-source data is unified through the decision-level fusion network, and the final fusion result is output.

[0025] Optionally, screening the evaluation indicators includes:

[0026] Pearson correlation coefficients were calculated for all indicators and converted into distance metrics;

[0027] Clustering indicators through hierarchical clustering algorithm makes indicators in the same category highly correlated;

[0028] From each category, an indicator with strong representativeness and clear forestry significance is selected as the forest vegetation health evaluation indicator.

[0029] Optionally, the step of generating a forest health grade distribution map includes:

[0030] Construct pyramid slices based on the fused data;

[0031] Calculating the forest vegetation health index value of each slice respectively, determining a grading threshold value according to the forest vegetation health index value, and dividing the health level according to the grading threshold value;

[0032] The results were generalized to the entire study area to generate a forest health grade distribution map.

[0033] The present invention also provides a computer device, comprising a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method.

[0034] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of the method when executed by a processor.

[0035] Compared with the prior art, the present invention has the following advantages and technical effects:

[0036] By integrating multi-source remote sensing data from satellites, drones, and ground sensors, this method avoids extensive manual fieldwork, significantly reducing fieldwork workload and human uncertainty. This method is suitable for diagnosing the health of large-scale forest vegetation and is particularly well-suited for assessing complex forest ecosystems. Furthermore, by integrating multi-source data fusion with a deep learning model, this method eliminates scale effects and differences in spatiotemporal resolution between different data sources, enabling a more accurate reflection of forest vegetation health, improving the spatiotemporal resolution of monitoring, and more sensitively reflecting the response of forest ecosystems to various disturbances. Furthermore, this method employs a hierarchical clustering method from unsupervised classification, combined with the Pearson correlation coefficient as a distance metric, to select the most representative indicators. The weights are then optimized using a machine learning algorithm to construct a Forest Health Index (FHI). This method avoids the subjectivity inherent in traditional methods due to human factors and ensures the scientific and rationality of the health index. The boxplot method is used to calculate health grading thresholds, further improving the accuracy of health grading. This method is not only applicable to forest health grading but can also be expanded to areas such as early warning of fires and pests, and carbon sink assessment, providing a more comprehensive decision-making support tool for forest resource management and ecological protection. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The accompanying drawings, which constitute part of this application, are intended to provide a further understanding of this application. The exemplary embodiments and descriptions of this application are intended to explain this application and do not constitute an improper limitation on this application. In the accompanying drawings:

[0038] Figure 1 is a flow chart of a method according to an embodiment of the present invention;

[0039] Figure 2 A sample plot layout diagram for an embodiment of the present invention;

[0040] Figure 3 Schematic diagram of a multi-source data fusion model according to an embodiment of the present invention;

[0041] Figure 4 This is a hierarchical clustering result diagram of an embodiment of the present invention. DETAILED DESCRIPTION

[0042] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0043] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0044] Example 1

[0045] like Figure 1 As shown, this embodiment provides a method for constructing and diagnosing a forest vegetation health index based on multi-source remote sensing data, including the following steps:

[0046] Select forest plots with healthy, sub-healthy, and unhealthy levels, and collect multi-source remote sensing data from the plots, including satellite data, drone data, and ground sensor data;

[0047] Construct a multi-source data fusion network, perform feature extraction and decision-level fusion on multi-source remote sensing data in sequence, and obtain fused data;

[0048] Extract evaluation indicators reflecting forest health from the fused data, use hierarchical clustering method to screen the evaluation indicators, and standardize the screened indicators;

[0049] A forest vegetation health index model was constructed based on the selected indicators. The model weights were optimized using a machine learning algorithm, and the health grade thresholds were determined using a boxplot method.

[0050] According to the grading threshold, the forest health levels in the diagnosis area are divided and a forest health level distribution map is generated.

[0051] It can be divided into the following steps:

[0052] In step S1, sample plot selection should also be based on differences in stand conditions, site conditions, human disturbance, and other factors. It is required to include several representative 20×20m sample plots of healthy, sub-healthy, and unhealthy species, different stand structures, and different developmental stages. The assessment of forest vegetation health should include multiple aspects, such as forest vitality, forest structure, and forest resistance.

[0053] Step S2: Use drones and ground sensors to acquire multispectral, hyperspectral, and point cloud data in the sample plots, measure all sample trees (DBH > 5 cm) in the sample plots, and collect satellite remote sensing data for the entire study area.

[0054] Step S3: Design feature extraction networks for remote sensing data from different sources, use CNN and LSTM networks to learn features, and perform decision-level fusion on the input fully connected network to obtain the final result;

[0055] Step S4: As many indicators as possible that can reflect forest health are obtained from multi-source remote sensing data. These indicators should be able to reflect the structural integrity, physiological activity, mechanical stability, etc. of the forest. Unsupervised clustering methods are used for screening. The screened indicators should also be standardized, for example (positive correlation indicator), (negative correlation indicator), (neutral index), where S′ is the normalized index value; S represents the value of a certain index of the plot; S max 、S min Represent the maximum and minimum values ​​of the plot, respectively;

[0056] Step S5: construct the forest vegetation health index FHI=f(α1, α2, α3...α n ) and use certain mathematical methods to fit the indicators, where α i Represents the evaluation index obtained after screening, i∈(1,n), f can be a linear evaluation model Nonlinear evaluation model Among them, W i is the weight, which can be determined by machine learning classification algorithm or reinforcement learning method. In order to make the final result within [0-1], the sum of all weights should be 1; the inclusion index and fitting equation of FHI can be adjusted and changed according to actual conditions.

[0057] Step S6, based on the selected healthy, sub-healthy and unhealthy forest samples, calculate and use box plots to represent the distribution of each FHI data of samples with different health levels, that is, T i-1,i =(X i-1 +X i ) / 2, where T i-1,i Respectively represent the thresholds between adjacent health levels; X i-1 With X i Respectively represent the 75th percentile value and the 25th percentile value between adjacent health levels; i can be 1, 2, 3 to represent healthy, sub-healthy and unhealthy respectively;

[0058] In step S7, a decision support system is constructed using the trained FHI index to map forest health levels. First, the fused data from step S3 is used to construct pyramid slices (1m×1m×n_bands). The FHI value for each slice is then calculated and classified into health levels based on the values. This process is repeated to ultimately produce a forest health map for the entire study area. The definition of a plot's health level should be based on a comprehensive evaluation of four aspects: forest structural integrity, tree physiological activity, environmental suitability, and forest resilience.

[0059] like Figure 2 As shown, the plot selection method mentioned in step S1 includes the following steps:

[0060] Step S21: Each tree plot is further divided into 10 small plots, including 5 5×5 m square plots of shrub layer and 5 circular plots with a radius of 1 m;

[0061] Step S22: For the tree layer, all trees in the plot (with a diameter at breast height ≥ 5 cm) should be numbered according to the relevant technical standards of the "Technical Specifications for Continuous Inventory of National Forest Resources" and measured using tools such as a diameter at breast height ruler and a height gauge. The tree measurement factors involved include tree species, diameter at breast height, tree height, and crown width.

[0062] Step S23: For the shrub layer and herb layer, record the species name, number of plants, etc. of all shrubs (tree height < 2m) and herbs in the sample plot, use diagonal intercept sampling or visual inspection method to record height, coverage, frequency and other information, and further calculate the required forest vegetation health evaluation indicators.

[0063] like Figure 3 As shown, the multi-source data fusion network mentioned in step S3 includes the following steps:

[0064] Step S31: Design different feature extraction networks for different types of data. For example, 1D-CNN and LSTM networks are used to learn features for one-dimensional data such as time and numerical values; 2D-CNN and LSTM networks are used to extract features for two-dimensional data such as satellite and UAV remote sensing images; and 3D-CNN and LSTM networks are used to extract features for three-dimensional data such as point clouds and stress waves.

[0065] Step S32: After sufficient feature extraction is performed on each of the three parts, the extracted feature information is spliced ​​together, and then a preliminary fusion result is obtained through a feature fusion module based on a three-layer fully connected network;

[0066] In step S33, in order to eliminate scale differences and unify spatiotemporal resolution, it is also necessary to dynamically adjust the weights according to the performance of the test set and the validation set through a decision-level weighted fusion strategy.

[0067] like Figure 4 As shown, the method for screening the indicators mentioned in step S4 includes the following steps:

[0068] Step S41, using the Pearson correlation coefficient (person correlation coefficient) As the sample distance, where X i and Y i are the i-th observation values ​​of the two variables, and are the means of the two variables respectively;

[0069] Step S42: Calculate the Pearson correlation coefficient of all indicators and calculate the correlation coefficient of all indicators by d(X, Y) = 1-ρ X,Y , the Pearson correlation coefficient is converted into a distance metric so that the distance corresponding to the indicator with high correlation is small;

[0070] Step S43: Using a hierarchical clustering algorithm, iteratively merge the closest clusters until a predetermined number of clusters or a distance threshold is reached. Throughout the clustering process, each merged cluster and its distance are recorded, ultimately constructing a dendrogram to visually display the hierarchical structure of the data.

[0071] In step S44, according to the principle of maximizing the intra-category correlation coefficient and minimizing the inter-category correlation coefficient, it is ensured that the data points within the same category are highly correlated, while the correlation between data points in different categories is low. This principle helps to improve the interpretability and predictive ability of the model. Ultimately, only one typical indicator with strong representativeness and clear silvicultural significance is selected from each category as the forest vegetation health indicator.

[0072] In this embodiment, the application scenarios of FHI can be further expanded to early warning of fire / insect pests, carbon sink assessment, etc.

[0073] This embodiment further provides a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method.

[0074] This embodiment further provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of the method when executed by a processor.

[0075] The above are merely preferred embodiments of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A method for constructing and diagnosing forest vegetation health index based on multi-source remote sensing data, characterized in that: The following steps are involved: Select forest sample plots that include healthy, sub-healthy, and unhealthy levels, and collect multi-source remote sensing data from the sample plots, including satellite data, drone data, and ground sensor data; Constructing a multi-source data fusion network, performing feature extraction and decision-level fusion on the multi-source remote sensing data in sequence to obtain fused data; Extracting evaluation indicators reflecting forest health from the fused data, screening the evaluation indicators using a hierarchical clustering method, and standardizing the screened indicators; A forest vegetation health index model is constructed based on the screened indicators, the weights of the model are optimized using a machine learning algorithm, and the health grade thresholds are determined using a boxplot method; The forest health levels in the area to be diagnosed are divided according to the grading threshold, and a forest health level distribution map is generated.

2. The method according to claim 1, characterized in that The selected forest plots include: Select a square plot and arrange shrub and herb plots in diagonal lines within the plot; Field measurements were conducted on trees, shrubs and herbs in the sample plots, and data including diameter at breast height, tree height, crown width, species name, number of plants, height, cover and frequency were collected.

3. The method according to claim 1, characterized in that The multi-source remote sensing data of the sample site includes: Obtain macroscopic spectral data including visible light, near infrared and thermal infrared bands through satellite remote sensing; Acquire medium-resolution canopy data using drones equipped with multispectral, hyperspectral, and LiDAR sensors; The three-dimensional structural parameters of vegetation are obtained through ground-based LiDAR scanners, and wireless sensor networks are deployed to collect temperature and humidity data in the forest microenvironment; A three-dimensional stress wave detection device is used to scan the wave velocity inside the trunk of the sample tree to detect the cavity inside the tree.

4. The method according to claim 1, wherein The fused data includes: Design feature extraction networks for different types of data. Use 1D-CNN and LSTM to extract one-dimensional data features, 2D-CNN and LSTM to extract two-dimensional data features, and 3D-CNN and LSTM to extract three-dimensional data features. The extracted different data features are spliced ​​and input into a feature fusion network with three fully connected layers; The spatiotemporal resolution of multi-source data is unified through the decision-level fusion network, and the final fusion result is output.

5. The method according to claim 1, characterized in that The evaluation indicators for screening include: Pearson correlation coefficients were calculated for all indicators and converted into distance metrics; Clustering indicators through hierarchical clustering algorithm makes indicators in the same category highly correlated; From each category, an indicator with strong representativeness and clear forestry significance is selected as the forest vegetation health evaluation indicator.

6. The method according to claim 1, characterized in that The steps of generating a forest health grade distribution map include: Construct pyramid slices based on the fused data; Calculating the forest vegetation health index value of each slice respectively, determining a grading threshold value according to the forest vegetation health index value, and dividing the health level according to the grading threshold value; The results were generalized to the entire study area to generate a forest health grade distribution map.

7. A computer device comprising a memory, a processor, and a computer program stored in the memory, wherein: The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 6.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images