A disease wood occurrence index estimation method and device based on multi-scale cooperation of unmanned aerial vehicles and satellites, a computer device, and a storage medium

Abstract

The application discloses a disease wood occurrence index estimation method and device based on multi-scale cooperation of unmanned aerial vehicles and satellites, computer equipment and a storage medium. The method comprises the following steps: acquiring unmanned aerial vehicle data and Sentinel-2 data in the same period; extracting disease wood area and plant number based on the unmanned aerial vehicle image, and constructing a disease wood data set of the unmanned aerial vehicle scale; aggregating the result to the Sentinel-2 pixel scale, calculating the disease wood area proportion, plant number, disease probability mean value and structure degradation index in each pixel, and constructing a spatial aggregation data set; then, the response relationship between the disease indexes and the Sentinel-2 spectral characteristics is established, and a regional scale disease wood occurrence index inversion model is constructed; finally, the model is applied to obtain wide-range continuous disease wood occurrence index estimation results. The application cooperates the advantages of high-precision identification of the unmanned aerial vehicle and wide-range coverage of the satellite, and improves the precision and efficiency of regional disease monitoring.

Description

Technical Field

[0001] This invention belongs to the field of remote sensing monitoring and forest resource survey technology, specifically involving a method, device, computer equipment and storage medium for estimating diseased tree occurrence indicators based on multi-scale collaboration of UAV remote sensing and satellite remote sensing, which is applicable to forest pest and disease monitoring and resource assessment. Background Technology

[0002] The occurrence and spread of forest diseases and pests significantly impact the stability of forest ecosystems and the safety of forestry production. Timely and accurate surveys of the area and number of diseased trees are crucial for forest pest and disease control and resource management. Traditional ground survey methods, relying mainly on manual field surveys and plot surveys, are not only labor-intensive and costly but also struggle to achieve rapid and continuous monitoring over large areas, making them unsuitable for the demands of modern, refined forest management.

[0003] With the development of remote sensing technology, forest disease monitoring methods based on satellite remote sensing imagery have been gradually applied. Among them, medium- and high-resolution optical satellites such as Sentinel-2 are widely used for monitoring forest health at regional and even national scales due to their advantages such as high temporal resolution, free access, and multispectral configuration. Existing technologies typically identify and statistically analyze disease-affected areas by analyzing vegetation indices or spectral variation characteristics. However, the spatial resolution of Sentinel-2 satellite imagery is typically 10 to 20 meters. In complex forest environments, a single satellite pixel often contains multiple trees and various land cover types, easily leading to significant pixel mixing effects. Especially in the early stages of disease outbreaks or when the disease severity is low, the proportion of diseased trees within a pixel is low, and their spectral anomalies are easily masked by the spectral signals of healthy vegetation, making it difficult to effectively identify diseased trees in satellite imagery. This situation means that when relying solely on medium-resolution satellite imagery for disease monitoring, the area and number of diseased trees are often systematically underestimated, making it difficult to meet the needs of refined management.

[0004] To improve the accuracy of disease monitoring, some existing technologies attempt to incorporate high-resolution satellite or aerial remote sensing imagery. However, high-resolution commercial satellite imagery is costly to acquire and significantly affected by imaging cycles and cloud cover, making it difficult to apply continuously over large areas and in multiple time phases. Furthermore, even with high-resolution imagery, it remains difficult to accurately distinguish between healthy and diseased trees at the individual tree scale, resulting in limited accuracy in tree number estimation.

[0005] Due to its advantages of flexibility, maneuverability, and high spatial resolution, unmanned aerial vehicle (UAV) remote sensing is increasingly being used in detailed forest surveys. Using centimeter- to sub-meter-level images acquired by UAVs, the canopy morphology of individual trees and their disease characteristics can be clearly identified, enabling accurate identification, area calculation, and tree count of diseased trees. However, UAV remote sensing has limited coverage, and its operational efficiency is constrained by flight time and airspace conditions, making it difficult to directly use for continuous monitoring at regional or larger scales.

[0006] While existing technologies utilize either satellite remote sensing or UAV remote sensing for forest disease monitoring, these methods are often used independently, lacking an effective multi-scale collaborative mechanism. Particularly at the satellite scale, there is a lack of a technical method to interpret and correct disease information within satellite pixels using the detailed identification results from UAVs. This makes it difficult to address the spectral signal weakening caused by diseased trees within individual pixels, thus limiting the accuracy of medium-resolution satellites in estimating the area and number of diseased trees.

[0007] Therefore, how to maintain the advantages of satellite remote sensing for large-scale monitoring while making full use of the ability of UAV remote sensing in the fine identification of diseased trees, and to construct a multi-scale collaborative method for estimating the area and number of diseased trees, has become a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0008] To address the shortcomings of existing technologies, this invention provides a method, apparatus, computer equipment, and storage medium for estimating diseased tree occurrence indicators based on multi-scale collaboration between UAVs and satellites.

[0009] To achieve the above-mentioned objectives, the technical solution adopted by the present invention is as follows:

[0010] A method for estimating diseased tree occurrence indicators based on multi-scale collaboration between UAVs and satellites includes the following steps:

[0011] S1. Acquire and preprocess data from drones and Sentinel-2 images during the same period. Based on drone images, accurately extract the area and number of diseased trees and construct a drone-scale diseased tree dataset.

[0012] S2 aggregates the UAV results to the Sentinel-2 pixel scale, calculates the proportion of diseased tree area, number of trees, mean disease probability, and structural degradation index within the pixel, and constructs a spatially aggregated dataset of satellite pixel-UAV diseased tree information.

[0013] S3. Construct a satellite pixel spectral feature dataset, establish the response relationship between diseased tree occurrence indicators and Sentinel-2 spectrum and spectral index, and based on the response relationship, construct a regional-scale inversion model for diseased tree occurrence indicators.

[0014] Further, in step S1, data from the concurrent UAV and Sentinel-2 are acquired and preprocessed, including the following steps:

[0015] Low-altitude aerial photography was conducted using drones equipped with multispectral sensors to collect images of sample areas with different degrees of disease occurrence (mild, moderate, and severe) within the study area, with image resolution reaching centimeter to sub-meter levels.

[0016] Preprocessing of UAV imagery, including radiometric and geometric corrections, ensures spatial consistency with satellite imagery.

[0017] Sentinel-2 Level-2A data of the target area was acquired using Google Earth Engine (GEE) and the time of drone aerial photography was to be as synchronous as possible.

[0018] Furthermore, the step of accurately extracting the area and number of diseased trees based on UAV imagery and constructing a UAV-scale diseased tree dataset includes the following sub-steps:

[0019] By utilizing the abnormal color and texture characteristics of diseased trees, the canopy of diseased trees is manually interpreted and delineated to distinguish between diseased and healthy trees. The canopy range of diseased trees is extracted, and a UAV-scale dataset of diseased trees containing the area and number of diseased trees is constructed.

[0020] Furthermore, the process of aggregating the UAV results to the Sentinel-2 pixel scale, calculating the proportion of diseased tree area, number of trees, mean disease probability, and structural degradation index within each pixel, and constructing a spatially aggregated dataset of satellite pixel-UAV diseased tree information includes the following sub-steps:

[0021] S21, using the 10m resolution pixel of Sentinel-2 as the basic spatial unit, superimposes the recognition results of the UAV-scale diseased tree dataset onto the corresponding spatial range;

[0022] S22, For each Sentinel-2 pixel, based on the area and number of diseased trees extracted from the UAV-scale diseased tree dataset, calculate the proportion of diseased tree area, the number of diseased trees, the mean disease probability, and the structural degradation index within it.

[0023] S23. Based on the calculated proportion of diseased area, number of diseased trees, mean disease probability, and structural degradation index, a spatial aggregation dataset of satellite pixel-UAV diseased tree information is constructed.

[0024] Further, in step S3, establishing the response relationship between the disease occurrence index and the Sentinel-2 spectrum and spectral index includes:

[0025] The main response indices corresponding to the diseased area ratio were determined to be the enhanced vegetation index and the vegetation red edge index.

[0026] The main response indices corresponding to the number of diseased plants were determined to be the normalized differential moisture index and the vegetation red edge index.

[0027] The main response indices corresponding to the mean disease probability and structural degradation index were determined to be the enhanced vegetation index, the vegetation red edge index, the normalized differential moisture index, and the normalized differential chlorophyll index.

[0028] Further, in step S3, the construction of a regional-scale satellite-based continuous estimation inversion regression mapping model for diseased trees includes the following sub-steps:

[0029] Using Sentinel-2 band reflectance and vegetation index as independent variables, and the proportion of diseased tree area and number of trees obtained by UAV aggregation as dependent variables, we explored the sensitivity of different band combinations and indices to disease information.

[0030] A regression model for the proportion of diseased tree area was constructed. The proportion of diseased tree area mainly reflects the change in the coverage of healthy vegetation within a pixel, and its spectral response is mainly based on the Enhanced Vegetation Index (EVI) and the Red Edge Vegetation Index (NDRE).

[0031] The regression model of diseased tree number reflects canopy structure fragmentation and water loss. Its spectral response is mainly based on the normalized differential moisture index (NDMI) and supplemented by the normalized differential chlorophyll index (NDRE).

[0032] A mean disease probability regression model was constructed. The mean disease probability mainly reflects the average risk level of the overall pixels being in a diseased state. Its spectral response is mainly based on EVI, NDRE, NDCI, and NDMI.

[0033] A structural degradation index regression model was constructed. The structural degradation index mainly reflects the degree of destruction, thinning and decline in integrity of the pixel forest canopy structure. Its spectral response is mainly composed of EVI, NDRE, NDCI and NDMI.

[0034] The model is written in matrix form, and the regression coefficients are estimated using the ordinary least squares (OLS) method to obtain the values ​​of each coefficient.

[0035] Furthermore, in step S3, the model accuracy and applicability are evaluated, including the following sub-steps:

[0036] Based on the estimated regression coefficients, regression hypothesis verification and variable validity analysis are performed to test the significance of the regression coefficients. Each regression coefficient is tested to determine whether the corresponding variable significantly affects the disease index. Through multicollinearity diagnosis, the variance inflation factor (VIF) is calculated to determine whether collinearity holds.

[0037] The accuracy of the model is evaluated, and the coefficient of determination is determined. The root mean square error (RMSE) measures the model's interpretability; it measures the magnitude of the estimation error.

[0038] To evaluate model stability, K-fold cross-validation is performed. The sample set is randomly divided into K parts, and K-1 parts are used for training and 1 part for validation each time. This is repeated K times, and the stability of each fold is calculated. , Based on the above To evaluate the stability of the model; to conduct coefficient stability analysis and compare different compromise regression coefficients;

[0039] Residual analysis and model rationality verification are performed to test the normality and heteroscedasticity of the residuals and output the final model.

[0040] This invention also discloses a device for estimating the occurrence index of diseased trees based on multi-scale collaboration between UAVs and satellites, used to execute the above-mentioned method for estimating the occurrence index of diseased trees, specifically including:

[0041] The acquisition module is configured to acquire a spatially aggregated dataset of satellite pixel-UAV-based diseased tree information for the target area during the same period.

[0042] The spectral response module was configured to use Sentinel-2 band reflectance and vegetation index as independent variables, and the proportion of diseased tree area and number of trees obtained by UAV aggregation as dependent variables, to obtain a sensitivity relationship model of different indices to disease information.

[0043] The estimation module is configured to apply the optimal model to Sentinel-2 images of the entire study area, generate a spatial distribution map of the area ratio and number of diseased trees at the regional scale, and summarize the estimated results of the total area and total number of diseased trees.

[0044] The present invention also discloses a computer device, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, wherein when the computer-executable instructions are executed by the processor, the steps of the above-described method for estimating the occurrence index of diseased trees are implemented.

[0045] The present invention also discloses a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the above-described method for estimating the occurrence index of diseased trees.

[0046] Compared with the prior art, the advantages of the present invention are as follows:

[0047] 1. This invention effectively overcomes the pixel mixing effect of medium-resolution satellite imagery by establishing a collaborative observation and information transmission mechanism between the fine scale of UAVs and the regional scale of satellites. Utilizing the precise background information of diseased trees extracted from high-resolution UAV imagery, the disease status within satellite pixels is quantitatively calibrated, thereby significantly improving the accuracy of large-scale diseased tree identification relying solely on satellite remote sensing. This solves the technical problem that diseased trees, especially those with mild diseases, are easily masked by healthy signals in satellite imagery, leading to missed detections or underestimations.

[0048] 2. This invention creatively achieves the complementary advantages of high-precision UAV point sampling and low-cost satellite area coverage. This method only requires conducting UAV surveys in typical areas to establish a training model, which can then be extended to massive amounts of historical and current Sentinel-2 imagery. This significantly expands the monitoring range while maintaining accuracy, reducing the cost and operational difficulty of achieving continuous dynamic monitoring of the area, and providing a feasible path for operationalization.

[0049] 3. This invention breaks through the limitations of traditional methods that only estimate the area of ​​disease. By aggregating information from unmanned aerial vehicles (UAVs), it can simultaneously derive multi-dimensional indicators such as the number of diseased trees, the average probability of disease, and the canopy structure degradation index. This provides a richer data foundation for assessing the distribution density, intensity, and degree of damage to the ecosystem structure of forest diseases, enabling a more comprehensive and in-depth quantitative assessment of their impact on forest health.

[0050] 4. The inversion model constructed in this invention has a clear physical mechanism basis, and there is an interpretable spectral response relationship between the feature variables used in the model and the disease indicators. This not only enhances the reliability and stability of the model, but also, because the selected spectral features such as red edge and shortwave infrared are relatively sensitive to changes in vegetation physiology, this scheme has potential application value in early disease monitoring and early warning.

[0051] 5. The UAV platform, Sentinel-2 satellite data, and Google Earth engine upon which this invention relies are all mature and open technological resources. The entire methodology is clear and highly standardized. Therefore, this solution is easy to reproduce and promote in different forest regions, possessing strong universality and high operability, making it convenient for deployment and application in actual forest resource management and pest and disease control. Attached Figure Description

[0052] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0053] Figure 1 This is a flowchart of a method for estimating diseased tree occurrence indicators based on multi-scale collaboration between UAVs and satellites, provided by an embodiment of the present invention.

[0054] Figure 2 This is a schematic diagram of a device for estimating the occurrence index of diseased trees based on multi-scale collaboration between UAVs and satellites, provided in an embodiment of the present invention.

[0055] Figure 3 This is a structural block diagram of a computing device provided in an embodiment of the present invention. Detailed Implementation

[0056] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0057] This invention provides a method for estimating the area and number of diseased trees based on multi-scale collaboration between UAVs and satellites. It also relates to a device for estimating the area and number of diseased trees based on multi-scale collaboration between UAVs and satellites, a computing device, a computer-readable storage medium, and a computer program, which will be described in detail in the following embodiments.

[0058] Figure 1 The flowchart illustrates a method for estimating the area and number of diseased trees based on multi-scale collaboration between UAVs and satellites, according to an embodiment of this specification, and specifically includes the following steps.

[0059] Step S1: Obtain concurrent UAV and Sentinel-2 data, and construct a UAV-scale diseased tree dataset based on the UAV data.

[0060] The specific steps include:

[0061] S11 employs a drone equipped with a multispectral sensor for low-altitude aerial photography, collecting sample areas within the study area with varying degrees of disease occurrence (mild, moderate, and severe), achieving image resolution from centimeter to sub-meter. Preprocessing of the drone images, including radiometric and geometric corrections, ensures spatial consistency with satellite imagery. Sentinel-2 Level-2A data within the target area is acquired using Google Earth Engine (GEE) and timed as synchronously as possible with the drone aerial photography.

[0062] S12 utilizes the abnormal color and texture characteristics of diseased trees to manually interpret and delineate the canopy of diseased trees, and constructs a drone-scale dataset of diseased trees that includes the area and number of diseased trees.

[0063] S2: Aggregate the UAV results to the Sentinel-2 pixel scale, calculate the proportion of diseased tree area, the mean probability of disease in the number of trees, and the structural degradation index within the pixel, and construct a spatially aggregated dataset of satellite pixel-UAV diseased tree information.

[0064] The specific steps include:

[0065] S21, using Sentinel-210m pixels as the basic unit, superimpose the results of UAV-damaged trees within its spatial range. For each satellite pixel, superimpose the UAV-damaged tree identification results within its spatial coverage range; construct a 10m aggregated target variable form, as shown in Table 1 below;

[0066] Table 1: Aggregate Target Variables

[0067] UAV output 10m pixel aggregation variable Disease / Health Binary Values Percentage of diseased area ( ) Single tree disease label Number of diseased plants ( ) Mean probability of disease Mean probability of disease ( ) Canopy damage Structural degradation index ( )

[0068] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0069] Calculate the area ratio of diseased trees within the satellite pixel, where the area ratio is the ratio of the area of ​​diseased trees to the total area of ​​the pixel; count the number of diseased trees contained within the satellite pixel; calculate the mean disease probability and structural degradation index, and establish a spatial aggregation dataset of satellite pixel-UAV diseased tree information.

[0070] S211, the proportion of diseased wood area, in a single Sentinel-2 pixel Within the coverage area, the proportion of UAV defective pixels is statistically analyzed.

[0071]

[0072] in, This represents the proportion of UAV-damaged pixels to the total area of ​​Sentinel-2 pixels. Represents UAV pixels; This represents a disease marker, where 0 indicates healthy and 1 indicates disease. It is the UAV pixel area; It is the area of ​​a Sentinel-2 pixel.

[0073] S212, the number of diseased trees is based on the results of single-tree identification by the UAV, counting the number of diseased trees within Sentinel-2 pixels:

[0074]

[0075] For the first Number of diseased trees within each Sentinel-2 pixel; This refers to the number of diseased trees that have fallen into this Sentinel-2 pixel. Indicating the first at the drone scale Are the trees in the forest diseased or damaged?

[0076] S213, the average probability of the disease. The "average risk level" represents the overall state of damage to pixels, and is particularly sensitive to early and mild damage.

[0077] UAV outputs a disease probability map, aggregates the probability mean, performs probability distribution statistics, and constructs a feature vector.

[0078] (1)

[0079] (2)

[0080] represent Mean probability of disease within each Sentinel-2 pixel; It is the first in the scale of drones The probability of disease in each disease unit (pixel or monocular); It is the number of UVA cells falling into this Sentinel-2 pixel.

[0081] S214, the structural degradation index ( This is used to quantitatively characterize the degree of forest canopy structure damage, thinning, and integrity decline at the Sentinel-2 pixel scale. UAV canopy binarization is used, and the proportion of voids within a 10m pixel is calculated; a larger value indicates more severe structural degradation. At the UAV scale, based on the identification results of diseased trees and canopy structure information, unit-level structural degradation is constructed.

[0082]

[0083] Representing the The structural degradation index of a Sentinel-2 pixel; It is the first in the scale of drones The amount of structural degradation of a single canopy unit or individual tree; It is the number of UVA cells falling into this Sentinel-2 pixel.

[0084] S3: Construct a Sentinel-2 satellite pixel spectral feature dataset, establish the response relationship between the satellite pixel-UAV diseased tree information spatial aggregation dataset and the Sentinel-2 satellite spectrum, construct a regional-scale satellite diseased tree continuous estimation inversion regression mapping model, and evaluate its accuracy and applicability.

[0085] The specific steps include:

[0086] S31. Based on Sentinel-2Level-2A data, the Enhanced Vegetation Index (EVI), Vegetation Red Edge Index (NDRE), Normalized Differential Moisture Index (NDMI), and Normalized Differential Chlorophyll Index (NDCI) were calculated to construct a Sentinel-2Level-2A spectral dataset of sensitive characteristics of diseased trees. The calculation formulas and changes in the representation of each vegetation index are shown in Table 2 below.

[0087] Table 2: Vegetation indices sensitive to diseased trees

[0088] EVI Canopy growth vitality NDRE Changes in vegetation physiological activity and chlorophyll content NDMI Changes in leaf water content and water stress in vegetation NDCI Changes in vegetation chlorophyll concentration and nutrient status

[0089] S32. Based on the above aggregated target variables, construct 10m pixel aggregated variables (proportion of diseased area, number of diseased trees, mean distribution of disease probability, and structural degradation index) and diseased tree sensitivity indices (enhanced vegetation index, vegetation red edge index, normalized differential moisture index, and normalized differential chlorophyll index) to construct the Sentinel-2Level-2A spectral dataset.

[0090] Canopy structural damage is mainly manifested as leaf and branch shedding, increased canopy voids, and xylem exposure, with its spectral response primarily reflected in the shortwave infrared band. The B11 and B12 bands of Sentinel-2 are highly sensitive to changes in canopy moisture content and structural continuity, serving as the primary source of spectral information characterizing the degree of canopy damage. The red-edge band provides auxiliary indication of canopy thinning and early structural changes, while the traditional NDVI index can only indirectly reflect structural degradation and is insufficient to characterize the degree of canopy damage independently.

[0091] The proportion of diseased area reflects changes in the degree of healthy vegetation coverage within a pixel, and its spectral response is mainly reflected in the red edge band and the NDVI index. As the proportion of diseased area increases, the area of ​​healthy vegetation coverage decreases, causing the red edge position to shift towards shorter wavelengths and the overall NDVI value to decrease.

[0092] The number of diseased plants reflects the degree of canopy structure fragmentation at the pixel scale. Its increase is usually accompanied by an increase in the number of canopy voids and enhanced xylem exposure. This type of change is mainly reflected in the Sentinel-2 shortwave infrared band, corresponding to a significant increase in reflectivity in the B11 and B12 bands.

[0093] The mean disease probability characterizes the overall intensity of disease occurrence at the pixel scale, and its spectral response is mainly reflected in the NDVI and red edge bands; while the disease probability distribution reflects the spatial heterogeneity and patchy characteristics of diseases within the pixel, and has higher sensitivity to the shortwave infrared band.

[0094] Table 3: Comparison Table of Aggregate Variables and Response Spectra

[0095] UAV aggregate variables Essential meaning Sentinel-2 Master Response Second Response proportion of diseased area Coverage Change Red-Edge, NDVI NIR Number of diseased plants Degree of structural fragmentation SWIR (B11 / B12) Red-Edge Mean probability of disease Physiological degeneration intensity NDVI, Red-Edge — Disease probability distribution Spatial heterogeneity SWIR Red-Edge

[0096] The calculation formulas and the changes in the vegetation indices are shown in Table 4 below;

[0097] Table 4: Feature-Label Pairs

[0098] Sentinel-2 characteristics Main response index Auxiliary Response Index 10m pixel aggregation variable NIR+RED+BULE EVI, NDRE NDCI Red-Edge+ NDMI, NDRE EVI SWIR EVI, NDRE, NDCI, NDMI — SWIR+Red-Edge+ +Red EVI, NDRE, NDCI, NDMI —

[0099] S33. Based on the above-mentioned establishment of the response relationship between the satellite pixel-UAV diseased tree information spatial aggregation dataset and the Sentinel-2 satellite spectrum, a regional-scale satellite diseased tree continuous estimation inversion regression mapping model is constructed, and its accuracy and applicability are evaluated.

[0100] S331, a regression model for the proportion of diseased tree area was constructed. The proportion of diseased tree area mainly reflects the change in healthy vegetation coverage within a pixel, and its spectral response is mainly represented by EVI and NDRE. The regression model is constructed as follows:

[0101]

[0102]

[0103] S332, a regression model for the number of diseased trees, where the number of diseased trees primarily reflects canopy structure fragmentation and water loss. Its spectral response is mainly based on NDMI, supplemented by NDRE. The regression model is constructed as follows:

[0104]

[0105]

[0106] S332, construct a regression model for the mean probability of disease. The mean probability of disease mainly reflects the average risk level of pixels being in a diseased state, and its spectral response is mainly represented by EVI, NDRE, NDCI, and NDMI. The regression model is constructed as follows:

[0107]

[0108]

[0109] S333, a regression model for the structural degradation index is constructed. The structural degradation index mainly reflects the degree of damage, thinning, and decline in integrity of the forest canopy structure in pixels. Structural degradation mainly affects canopy continuity and porosity (NDMI), the sensitivity of red-edge reflection to structural disturbances (NDRE, NDCI), and overall cover decline (EVI). Its spectral response is mainly dominated by EVI, NDRE, NDCI, and NDMI. The regression model is constructed as follows:

[0110]

[0111]

[0112] in, This demonstrates a negative correlation between decreased moisture content and increased structural degradation.

[0113] S334 establishes a quantitative relationship model between Sentinel-2 spectral information and disease status at the Sentinel-2 10m pixel scale, enabling the estimation of the proportion of diseased tree area, the number of diseased trees, the mean disease probability, and the structural degradation index. Dependent variables are defined for different disease characterization targets. Proportion of diseased wood area , number of diseased trees Mean probability of disease Structural degradation index Taking the Sentinel-2 disease-sensitive vegetation index as an example, the independent variable vector X is as follows, and each index is calculated at the Sentinel-2 10m pixel scale:

[0114]

[0115] Constructing a spatially aggregated dataset of satellite pixel-drone diseased tree information:

[0116]

[0117] M represents the number of valid Sentinel-2 pixel samples, with each sample containing 4 explanatory variables and 1 dependent variable.

[0118] S335, using a multiple linear regression model to describe the relationship between Sentinel-2 spectra and disease indicators:

[0119]

[0120] For the intercept term, ~ For regression coefficients, This is the random error term.

[0121] The above model is rewritten in matrix form, and the regression coefficients are estimated. Taking the regression model of the number of diseased trees as an example, the specific steps are as follows:

[0122] Rewrite the model in matrix form:

[0123]

[0124] Y is the dependent variable, and X is the independent variable. Let be the random error, where:

[0125]

[0126] The regression coefficients were estimated using ordinary least squares (OLS), and the results were obtained. , , , and .

[0127]

[0128] This represents the transpose of matrix X, i.e., rows transformed into columns and columns transformed into rows.

[0129] S336, based on the estimated regression coefficients, performs regression hypothesis testing and variable validity analysis as follows:

[0130] To test the significance of the regression coefficients, a t-test was performed on each regression coefficient to determine whether the corresponding variable significantly affects the disease index.

[0131]

[0132] Let be the estimated value of the i-th regression coefficient. This represents the standard error of the coefficient.

[0133] Diagnosis of multicollinearity, calculation of variance inflation factor (VIF):

[0134]

[0135] If VIF < 5, collinearity is acceptable; if VIF > 10, variables need to be removed or combined.

[0136] S336, Evaluate the accuracy of the model and determine the coefficients of determination. The root mean square error (RMSE) measures the model's interpretability; it measures the magnitude of the estimation error, and its specific formula is as follows:

[0137]

[0138]

[0139] M is the number of samples, i.e., the number of Sentinel-2 pixels. It is the true value of the j-th sample. These are model predictions. This represents the average of the true values ​​of all samples.

[0140] S337, Evaluate model stability; the specific steps are as follows:

[0141] K-fold cross-validation randomly divides the sample set into K parts, and uses... One training sample, one validation sample, repeated K times, calculate the result of each fold. , .

[0142] Based on the above To evaluate the model's stability, the stability index formula is as follows:

[0143]

[0144] In K-fold cross-validation, the standard deviation of the RMSE for each fold is given. It is the average RMSE of each fold in the K-fold cross-validation. <0.2 indicates that the model is stable; If the value is greater than 0.3, the model is unstable.

[0145] Stability analysis of coefficients, comparing different compromise regression coefficients, using the following formula:

[0146]

[0147] This represents the maximum value that the i-th regression coefficient takes in all K-fold cross-validations. This represents the minimum value of the i-th regression coefficient across all K-fold cross-validations. A smaller value indicates that the variable's contribution is stable.

[0148] S338, Residual Analysis and Model Reasonableness Validation: Testing for residual normality and heteroscedasticity. The residual calculation formula is as follows:

[0149]

[0150] S339, output the final model, the formula is as follows:

[0151]

[0152] The above is a schematic scheme of a diseased tree occurrence index estimation device based on multi-scale collaboration between UAVs and satellites according to this embodiment. It should be noted that the technical solution of this diseased tree occurrence index estimation device based on multi-scale collaboration between UAVs and satellites belongs to the same concept as the technical solution of the aforementioned method for estimating diseased tree occurrence indexes based on multi-scale collaboration between UAVs and satellites. Details not described in detail in the technical solution of the diseased tree occurrence index estimation device based on multi-scale collaboration between UAVs and satellites can be found in the description of the technical solution of the aforementioned method for estimating diseased tree occurrence indexes based on multi-scale collaboration between UAVs and satellites.

[0153] Figure 3 A schematic diagram of a computing device 300 according to another embodiment of the present invention is shown. The computing device 300 includes at least a storage unit 310 and a processing unit 320, which are connected for data communication via a bus 330; a database 350 is also provided for data storage and management.

[0154] In some embodiments, the computing device 300 may further include an access module 340 for supporting data exchange and communication via one or more networks 360. The network 360 may be any form of communication network or combination thereof, such as a Public Switched Telephone Network (PSTN), a Local Area Network (LAN), a Wide Area Network (WAN), a Personal Area Network (PAN), and the Internet. The access module 340 may be configured to include various wired or wireless interfaces, such as a Network Interface Card (NIC), a wireless LAN interface compliant with the IEEE 802.11 standard, a Wi-MAX interface, an Ethernet interface, a Universal Serial Bus (USB) interface, a cellular communication interface, a Bluetooth interface, and a Near Field Communication (NFC) interface.

[0155] In this embodiment, the functional modules in the computing device 300, and Figure 3 Other modules not explicitly shown can be connected via a bus or other communication mechanisms. It should be understood that... Figure 3 The structure shown is merely an example and does not constitute a limitation on the scope of protection of this invention. Those skilled in the art can appropriately add, delete, or replace the modules according to specific application requirements.

[0156] The computing device 300 can take various terminal forms, including mobile devices and fixed devices. For example, it can be a tablet computer, personal digital assistant (PDA), portable computer, laptop computer, netbook, smartphone, or wearable device (such as smartwatch, smart glasses, etc.), or it can be a desktop computer, personal computer, or server equipment, wherein the server can be a mobile server or a fixed server.

[0157] The processing unit 320 executes computer-executable instructions stored in the storage unit 310 to implement the various steps of the above-described method for estimating the occurrence index of diseased trees based on multi-scale collaboration between UAVs and satellites. It should be noted that the computing device scheme in this embodiment is based on the same technical concept as the aforementioned method scheme; for parts not detailed herein, please refer to the relevant descriptions in the corresponding method embodiments.

[0158] Another embodiment of the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, can realize the steps of the above-described method for estimating the occurrence index of diseased trees based on multi-scale collaboration between UAVs and satellites. This storage medium solution and the aforementioned method solution belong to the same technical concept; details not elaborated herein can be found in the corresponding method embodiments.

[0159] Furthermore, the present invention also provides a computer program that, when run on a computer, enables the computer to execute the steps of the above-described method for estimating the occurrence index of diseased trees. This program is also based on the technical concept of the above method, and details not described in detail can be found in the method embodiments.

[0160] The above embodiments are merely illustrative examples of the present invention. Other implementations are possible within the scope of the claims. In some cases, the steps defined in the claims may be performed in a different order than in the embodiments, while still achieving the desired effect; the processes in the accompanying drawings are not required to be performed in the shown order or sequential order. Furthermore, in some embodiments, parallel processing or multi-tasking methods may be employed.

[0161] The computer instructions include computer program code, which may take the form of source code, object code, executable files, or intermediate code. Computer-readable media can be any entity or device capable of carrying the aforementioned program code, such as storage media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), and software distribution media. In different jurisdictions, the scope of computer-readable media may be appropriately adjusted according to applicable laws and patent practices; for example, in some jurisdictions, electrical carrier signals or electrical signals may not be included.

[0162] It should be noted that, for ease of expression, the method embodiments are described as a combination of several steps in the description of the embodiments. However, those skilled in the art should understand that the execution order of the steps is not strictly limited, and in some cases the order can be adjusted or the steps can be executed in parallel. Furthermore, the embodiments listed in the specification are preferred solutions, and not all of the steps or modules involved are essential components.

[0163] Each embodiment has its own emphasis in its description. For parts that are not described in detail, please refer to the relevant content of other embodiments.

[0164] In summary, the above embodiments are only used to illustrate the principles and applications of the present invention and do not constitute a limitation thereof. Those skilled in the art can make various modifications and substitutions without departing from the spirit and substance of the present invention, and all such equivalent changes should fall within the protection scope of the claims of the present invention.

Claims

1. A method for estimating diseased tree occurrence indicators based on multi-scale collaboration between UAVs and satellites, characterized in that, Including the following steps: S1. Acquire and preprocess data from drones and Sentinel-2 images during the same period. Based on drone images, accurately extract the area and number of diseased trees and construct a drone-scale diseased tree dataset. S2 aggregates the UAV results to the Sentinel-2 pixel scale, calculates the proportion of diseased tree area, number of trees, mean disease probability, and structural degradation index within the pixel, and constructs a spatially aggregated dataset of satellite pixel-UAV diseased tree information. S3. Construct a satellite pixel spectral feature dataset, establish the response relationship between diseased tree occurrence indicators and Sentinel-2 spectrum and spectral index, and based on the response relationship, construct a regional-scale inversion model for diseased tree occurrence indicators.

2. The method for estimating the occurrence index of diseased trees according to claim 1, characterized in that, In step S1, data from the concurrent UAV and Sentinel-2 are acquired and preprocessed, including the following steps: Low-altitude aerial photography was conducted using drones equipped with multispectral sensors to collect images of sample areas with different degrees of disease occurrence (mild, moderate, and severe) within the study area, with image resolution reaching centimeter to sub-meter levels. Preprocessing of UAV imagery, including radiometric and geometric corrections, ensures spatial consistency with satellite imagery. Sentinel-2 Level-2A data within the target area was acquired using Google Earth Engine (GEE) and the drone aerial photography time was kept as close as possible to the time of the drone aerial photography.

3. The method for estimating the occurrence index of diseased trees according to claim 2, characterized in that, The process of accurately extracting the area and number of diseased trees from UAV imagery and constructing a UAV-scale diseased tree dataset includes the following sub-steps: By utilizing the abnormal color and texture characteristics of diseased trees, the canopy of diseased trees is manually interpreted and delineated to distinguish between diseased and healthy trees. The canopy range of diseased trees is extracted, and a UAV-scale dataset of diseased trees containing the area and number of diseased trees is constructed.

4. The method for estimating the occurrence index of diseased trees according to claim 3, characterized in that, The process of aggregating UAV results to the Sentinel-2 pixel scale, calculating the proportion of diseased tree area, number of trees, mean disease probability, and structural degradation index within each pixel, and constructing a spatially aggregated dataset of satellite pixel-UAV diseased tree information includes the following sub-steps: S21, using the 10m resolution pixel of Sentinel-2 as the basic spatial unit, superimposes the recognition results of the UAV-scale diseased tree dataset onto the corresponding spatial range; S22, For each Sentinel-2 pixel, based on the area and number of diseased trees extracted from the UAV-scale diseased tree dataset, calculate the proportion of diseased tree area, the number of diseased trees, the mean disease probability, and the structural degradation index within it. S23. Based on the calculated proportion of diseased area, number of diseased trees, mean disease probability, and structural degradation index, a spatial aggregation dataset of satellite pixel-UAV diseased tree information is constructed.

5. The method for estimating the occurrence index of diseased trees according to claim 4, characterized in that, In step S3, establishing the response relationship between diseased tree occurrence indicators and Sentinel-2 spectrum and spectral index includes: The main response indices corresponding to the diseased area ratio were determined to be the enhanced vegetation index and the vegetation red edge index. The main response indices corresponding to the number of diseased plants were determined to be the normalized differential moisture index and the vegetation red edge index. The main response indices corresponding to the mean disease probability and structural degradation index were determined to be the enhanced vegetation index, the vegetation red edge index, the normalized differential moisture index, and the normalized differential chlorophyll index.

6. The method for estimating the occurrence index of diseased trees according to claim 5, characterized in that, In step S3, the construction of a regional-scale satellite-based continuous estimation inversion regression mapping model for diseased trees includes the following sub-steps: Using Sentinel-2 band reflectance and vegetation index as independent variables, and the proportion of diseased tree area and number of trees obtained by UAV aggregation as dependent variables, we explored the sensitivity of different band combinations and indices to disease information. A regression model of the proportion of diseased tree area was constructed. The proportion of diseased tree area mainly reflects the change in the coverage of healthy vegetation within the pixel. Its spectral response is mainly based on the enhanced vegetation index (EVI) and the vegetation red edge index (NDRE). The regression model for the number of diseased trees reflects canopy structure fragmentation and water loss. Its spectral response is mainly based on the normalized differential moisture index (NDMI), supplemented by the normalized differential chlorophyll index (NDRE). A mean disease probability regression model was constructed. The mean disease probability mainly reflects the average risk level of the overall pixels being in a diseased state. Its spectral response is mainly based on EVI, NDRE, NDCI, and NDMI. A structural degradation index regression model was constructed. The structural degradation index mainly reflects the degree of destruction, thinning and decline in integrity of the pixel forest canopy structure. Its spectral response is mainly composed of EVI, NDRE, NDCI and NDMI. The model is written in matrix form, and the regression coefficients are estimated using ordinary least squares to obtain the values ​​of each coefficient.

7. The method for estimating the occurrence index of diseased trees according to claim 6, characterized in that, Step S3 evaluates the model's accuracy and applicability, including the following sub-steps: Based on the estimated regression coefficients, regression hypothesis verification and variable validity analysis are performed to test the significance of the regression coefficients. Each regression coefficient is tested to determine whether the corresponding variable significantly affects the disease index. By diagnosing multicollinearity, the variance inflation factor is calculated to determine whether multicollinearity is valid. The accuracy of the model is evaluated, and the coefficient of determination is determined. The root mean square error (RMSE) measures the magnitude of the estimation error, which in turn measures the model's interpretability. To evaluate model stability, K-fold cross-validation is performed. The sample set is randomly divided into K parts, and K-1 parts are used for training and 1 part for validation each time. This is repeated K times, and the stability of each fold is calculated. , Based on the above To evaluate the stability of the model; to conduct coefficient stability analysis and compare different compromise regression coefficients; Residual analysis and model rationality verification are performed to test the normality and heteroscedasticity of the residuals and output the final model.

8. A device for estimating the occurrence index of diseased trees based on multi-scale collaboration between UAVs and satellites, used to execute the method for estimating the occurrence index of diseased trees according to any one of claims 1 to 7, characterized in that, include: The acquisition module is configured to acquire a spatially aggregated dataset of satellite pixel-UAV-based diseased tree information for the target area during the same period. The spectral response module was configured to use Sentinel-2 band reflectance and vegetation index as independent variables, and the proportion of diseased tree area and number of trees obtained by UAV aggregation as dependent variables, to obtain a sensitivity relationship model of different indices to disease information. The estimation module is configured to apply the optimal model to Sentinel-2 images of the entire study area, generate a spatial distribution map of the area ratio and number of diseased trees at the regional scale, and summarize the estimated results of the total area and total number of diseased trees.

9. A computer device, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, they implement the steps of the method for estimating the occurrence index of diseased trees according to any one of claims 1 to 7.

10. A computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the method for estimating the disease occurrence index of any one of claims 1 to 7.

Images