Drought stress under vegetation canopy spectral image unmanned aerial vehicle patrol measurement method and system

Abstract

The application discloses a method and system for unmanned aerial vehicle patrol measurement of vegetation canopy spectral images under drought stress, and relates to the technical field of unmanned aerial vehicle remote sensing image processing. The method comprises the following steps: a multi-rotor unmanned aerial vehicle is used to carry a multispectral camera to collect multispectral images of a canopy in a research area; after radiation calibration and geometric correction, a spectral stress intensity is calculated pixel by pixel; according to the texture of a visible light channel, a canopy structure unevenness is calculated pixel by pixel, and a local spatial gradient thereof is extracted as a structure gradient; the spectral stress intensity and the structure gradient are coupled as two orthogonal measurement dimensions to form a pixel binary measurement; without relying on ground measurement calibration, the pixels are divided into four grades of normal, mild, moderate and severe, and a stress spatial distribution map is output. The application reverses the canopy structure unevenness into a second measurement dimension orthogonal to the spectrum, and can objectively measure the spatial distribution of stress intensity, the expansion rate and the spatial migration direction of a landscape scale without specific ground calibration of a species.

Description

Technical Field

[0001] This invention relates to the field of UAV remote sensing image processing technology, specifically to a method and system for UAV inspection and measurement of vegetation canopy spectral images under drought stress. Background Technology

[0002] In ecological and agroforestry research, objective, image-based measurements of the early response characteristics of vegetation canopies under drought stress are a crucial supporting link in assessing vegetation drought resistance and monitoring drought progression. In the early stages of drought stress, vegetation exhibits responses such as decreased leaf water content, reduced chlorophyll content, and changes in canopy structure. However, these changes in visible light morphology are not yet significant at this stage and are difficult to directly identify using conventional visible light aerial images. For a long time, researchers have primarily used portable chlorophyll meters and moisture meters to measure stress leaf by leaf within quadrats. This method is inefficient and lacks coverage of the spatial distribution of stress over large landscape scales, failing to meet the needs of continuous landscape-scale surveys and measurements.

[0003] With the development of multi-rotor drones and lightweight multispectral imaging technology, UAV low-altitude multispectral remote sensing has provided a high spatiotemporal resolution, high mobility, and low cost data acquisition method for landscape-scale vegetation monitoring, and has been widely used in vegetation status measurement. However, existing UAV multispectral vegetation stress measurement schemes have common limitations in their technical approaches.

[0004] Chinese invention patent CN115950838B discloses a method for rapid monitoring and discrimination of drought conditions in summer maize using unmanned aerial vehicles (UAVs) based on chlorophyll content. This method uses a multi-rotor UAV equipped with a multispectral camera to acquire canopy multispectral images and calculates the Normalized Difference Vegetation Index (NDVEI), Soil-Regulated Vegetation Index (SDEV), and Red-Edge NDVEI. It then uses ground-measured chlorophyll content at different growth stages to establish regression equations to retrieve chlorophyll content, and finally classifies plots into four levels—normal, mild drought, moderate drought, and severe drought—based on calibrated chlorophyll content thresholds. However, this method requires a co-registration-based NDEV threshold method to remove structurally uneven areas such as canopy gaps as background before retrieval, retaining only pure vegetation pixels. Furthermore, its classification is limited to a single spectral intensity dimension per pixel and relies on ground-measured chlorophyll content at each growth stage to establish a species-specific regression calibration model. Therefore, without contacting the ground and without species-specific calibration, this scheme cannot objectively measure the spatial distribution of drought stress intensity in the vegetation canopy at the landscape scale.

[0005] Furthermore, Chinese invention patent CN109459392A discloses a method for estimating aboveground biomass of rice based on UAV multispectral imagery. This method uses spectral vegetation index and texture feature parameters as independent variables in a stepwise multiple regression to estimate biomass. Although this scheme introduces texture features, these features are only used as parallel independent variables in the regression fitting with ground-based measurements. Ground-based data is still required to establish the regression model, and the measurement results are still dependent on ground calibration. Moreover, it does not address the spatial grading and temporal variation measurement of stress levels.

[0006] Therefore, existing technologies generally treat the spatial non-uniformity of canopy cover as background noise that interferes with the extraction of pure vegetation pixel spectra and discard it, and use the single spectral intensity dimension per pixel combined with ground calibration as the sole basis for stress measurement. The core bottleneck of this technical approach is that, without contacting the ground and without species-specific calibration, it is impossible to objectively and graphically measure the spatial distribution of drought stress intensity in vegetation canopy at the landscape scale. Summary of the Invention

[0007] To address the core technical bottlenecks in existing UAV multispectral vegetation stress measurements, which rely on removing canopy structural inhomogeneity as background noise and depend on species-specific ground calibration, thus failing to objectively measure the spatial distribution of stress at the landscape scale within the pure image domain, this invention provides a UAV survey measurement method and system for vegetation canopy spectral images under drought stress. By inverting the local spatial gradient of canopy structural inhomogeneity into a second measurement dimension orthogonal to spectral intensity and coupling it into a pixel-based binary measurement quantity on the same pixel, objective image measurement of the spatial distribution of drought stress intensity in vegetation canopy is achieved from the principle of positively complementary measurement dimensions without relying on ground-based field calibration.

[0008] The technical solution of this invention is as follows: A method for UAV-based surveying and measurement of vegetation canopy spectral images under drought stress, comprising the following steps: Step S1: A multi-rotor UAV equipped with a multispectral camera flies along a preset route and collects canopy multispectral images of the study area by the multispectral camera. Step S2: Perform radiometric calibration and geometric correction on the canopy multispectral image, and calculate the spectral stress intensity pixel by pixel based on the red edge channel and near-infrared channel of the corrected canopy multispectral image; Step S3: Based on the visible light channel texture of the corrected canopy multispectral image, calculate the canopy structural inhomogeneity pixel by pixel. The canopy structural inhomogeneity includes canopy coverage uniformity and canopy void distribution. Extract the local spatial gradient of the canopy structural inhomogeneity in the pixel neighborhood as the structural gradient quantity. Step S4: The spectral stress intensity and the structural gradient are used as two orthogonal measurement dimensions and coupled into a pixel binary measurement on the same pixel. Step S5: Without relying on ground-based measured targets, the pixels in the study area are divided into four levels: normal state, mild stress, moderate stress, and severe stress based on the pixel binary measurement, and a stress spatial distribution map is output.

[0009] This invention also provides a drone-based survey and measurement system for vegetation canopy spectral images under drought stress, comprising: an image acquisition module for controlling a multi-rotor drone equipped with a multispectral camera to fly along a preset route and acquire canopy multispectral images of the study area; a spectral measurement module for performing radiometric calibration and geometric correction on the canopy multispectral images and calculating the spectral stress intensity per pixel; a structural measurement module for calculating the canopy structural inhomogeneity per pixel based on the visible light channel texture of the canopy multispectral images and extracting its local spatial gradient as a structural gradient; a binary coupling module for coupling the spectral stress intensity and the structural gradient into a pixel binary measurement; and a graded output module for classifying the pixels of the study area into four levels—normal state, mild stress, moderate stress, and severe stress—based on the pixel binary measurement without relying on ground-based calibration and outputting a stress spatial distribution map.

[0010] The beneficial effects of this invention are as follows: First, by extracting the local spatial gradient of canopy structural inhomogeneity as a structural gradient quantity, and coupling it with spectral stress intensity as two orthogonal measurement dimensions on the same pixel to form a pixel-based binary measurement quantity, this invention achieves objective measurement of the spatial distribution of stress intensity without relying on ground-based calibration. The mechanism is as follows: the spectral stress intensity quantity is a zero-order intensity quantity reflecting the apparent degree of stress on the pixel itself, while the local spatial gradient of canopy structural inhomogeneity is a first-order spatial differential component reflecting the boundary and orientation of stressed patches. The two are orthogonal in measurement information and cannot be linearly derived from each other; therefore, the information dimension carried by the pixel-based binary measurement quantity composed of these two is strictly higher than either single dimension, and its spatial differential component naturally does not depend on the absolute radiation baseline of the image. Compared to the approach of removing structurally inhomogeneous areas such as canopy gaps as background and using only a single spectral intensity dimension in conjunction with ground chlorophyll calibration for classification, this invention does not require establishing a species-specific ground regression calibration model, and the classification basis is upgraded from one-dimensional to a two-dimensional orthogonal quantity.

[0011] Second, by determining the structural confidence based on the proportion of vegetation cover pixels in the pixel neighborhood and weighting it on the first-order spatial differential component, this invention suppresses spurious structural gradients in sparsely covered regions. The mechanism is as follows: in sparsely covered regions, the structural gradient is easily dominated by the edges of canopy gaps, generating artificially high gradients unrelated to stress. The proportion of vegetation cover pixels is an endogenous quantity characterizing the confidence level of the structural gradient in this region. By weighting the spatial differential component with a structural confidence that is positively correlated with the coverage proportion, the structural gradient contribution in low-confidence regions can be adaptively reduced without introducing any external calibration. Compared to the approach of using texture features as indiscriminate regression variables in contrasting documents, this invention adaptively suppresses spurious responses of the structural gradient in sparsely covered regions.

[0012] Third, by using the topological invariant of the canopy void as the relative registration reference for multiple sorties and performing baseline self-normalization on the binary measurements of pixels, this invention eliminates baseline drift caused by differences in flight altitude, illumination, and phenology between different sorties without contacting the ground. The mechanism is as follows: the connectivity morphology of the canopy void remains unchanged under scale changes, representing a stable structural feature independent of absolute radiance values. Using it as the relative registration reference allows for precise alignment of corresponding pixels from different sorties. Furthermore, by using the statistical values ​​of normal-state pixels within each sortie as the baseline for self-normalization of the binary measurements, the absolute horizontal differences between sorties are converted into relative offsets, thus avoiding misjudging baseline drift as forced extension.

[0013] Fourth, the aforementioned features generate a synergistic effect that exceeds the sum of their individual effects. Specifically: the spectral stress intensity on a single pixel can only provide the instantaneous stress intensity of that pixel, and the structural gradient on a single pixel can only provide the boundary intensity of that pixel; neither can characterize the spatial dynamics of stress when existing alone. However, when orthogonally coupled pixel-level binary measurements are aligned across multiple flights via relative registration with topological invariants and baseline self-normalization, spatiotemporal differencing can be performed to construct a stress front vector field. The divergence of this vector field can then be used to quantify the expansion rate of each stress region, and its principal direction can be used to quantify the spatial migration direction of each stress region. Thus, the combination of pixel-level instantaneous intensity and first-order gradient yields field-level expansion rates and migration directions—two types of spatiotemporal dynamic measurement results that single-pixel measurements cannot provide—achieving a nonlinear gain in measurement capability. Attached Figure Description

[0014] Figure 1 This is a flowchart of the method for UAV inspection and measurement of vegetation canopy spectral images under drought stress according to the present invention.

[0015] Figure 2 This is a schematic diagram of the structure of the UAV patrol and measurement system for vegetation canopy spectral images under drought stress according to the present invention. Detailed Implementation

[0016] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. The following embodiments are used to illustrate the present invention, but are not intended to limit the scope of protection of the present invention.

[0017] This embodiment provides a method for UAV-based surveying and measurement of vegetation canopy spectral images under drought stress, such as... Figure 1 As shown, the method includes steps S1 to S5. This embodiment uses a survey and measurement of the shrub and grass canopy in a semi-arid region as an example. The study area is flat and open. The purpose of the study is to obtain the spatial distribution of drought stress intensity and its dynamic expansion over time. All measurement outputs are strictly limited to image measurement parameters of canopy spectral characteristics and structural indices, which are used by downstream researchers for comprehensive analysis in conjunction with meteorological data and soil moisture content.

[0018] Step S1: Canopy Multispectral Image Acquisition. In step S1, a multi-rotor UAV equipped with a multispectral camera flies along a preset route, and the multispectral camera acquires canopy multispectral images of the study area.

[0019] The multi-rotor UAV is a quadcopter or hexcopter configuration, and its payload bay carries the multispectral camera. The multispectral camera has a visible light channel, an infrared edge channel, and a near-infrared channel, wherein the visible light channel further includes a blue light channel, a green light channel, and a red light channel. In this embodiment, the center wavelength of the infrared edge channel is 717nm, the center wavelength of the near-infrared channel is 842nm, and the center wavelengths of the blue, green, and red light channels of the visible light channel are 475nm, 560nm, and 668nm, respectively.

[0020] Before data acquisition, a pre-defined flight path is planned in the ground control software according to the boundary of the study area. In this embodiment, the flight altitude of the pre-defined flight path is set to 80m, corresponding to a ground resolution of approximately 4cm, with a forward overlap rate of 80% and a lateral overlap rate of 75%. The flight altitude is selected within the range of 50m to 120m. Too low a flight altitude will result in insufficient coverage area per sortie and reduced patrol efficiency, while too high a flight altitude will cause the ground resolution to exceed 10cm, making it impossible to distinguish structural details such as canopy gaps. Both the forward overlap rate and the lateral overlap rate are not less than 70% to ensure that adjacent flight strip images have sufficient homonymous features for subsequent stitching and registration. The multi-rotor UAV flies along the pre-defined flight path at a constant speed, and the multispectral camera triggers shooting according to a pre-defined equidistant exposure interval to acquire a multi-channel aerial image sequence covering the study area.

[0021] The multi-rotor UAV transmits the acquired multi-channel aerial photograph sequence to the processing unit via onboard storage or data link. The processing unit performs image stitching on the multi-channel aerial photograph sequence to obtain the canopy multispectral image covering the entire study area. The canopy multispectral image is a multi-channel raster image containing visible light channels, red edge channels, and near-infrared channels, with each pixel recording the corresponding surface radiance in each channel. To ensure the temporal comparability of the survey measurements, this embodiment repeats step S1 at fixed time intervals in the same study area to obtain multiple canopy multispectral images belonging to different sorties, with a time interval of 7 days between adjacent sorties.

[0022] Step S2: Calculation of spectral stress intensity. In step S2, radiometric calibration and geometric correction are performed on the canopy multispectral image, and the spectral stress intensity is calculated pixel by pixel based on the red-edge channel and near-infrared channel of the corrected canopy multispectral image.

[0023] First, radiometric calibration is performed on the canopy multispectral image. This involves converting the radiance of each channel into surface reflectance based on the standard reflectance plate image captured simultaneously during acquisition, thus eliminating the influence of differences in light intensity and camera response. Then, geometric correction is performed, registering the image to a geographic coordinate system based on airborne positioning and attitude data and ground control points, ensuring that each pixel corresponds to a specific geographical location. Both radiometric calibration and geometric correction are well-known preprocessing operations in the art.

[0024] On the corrected canopy multispectral image, various spectral indices reflecting vegetation chlorophyll content and leaf water content are calculated pixel-by-pixel. In this embodiment, the normalized vegetation index (NDI) is calculated based on red channel reflectance and near-infrared channel reflectance, the red-edge vegetation index is calculated based on red-edge channel reflectance and near-infrared channel reflectance, and the water stress index is calculated based on near-infrared channel reflectance and shortwave reflectance characteristics. The aforementioned NDI, red-edge vegetation index, and water stress index are all standard spectral indices known in the art, reflecting vegetation greenness, chlorophyll status, and water status, respectively.

[0025] The red-edge and near-infrared channels were chosen as the primary sources of spectral stress intensity because, in the early stages of drought stress, the decrease in chlorophyll content and leaf water content is preferentially reflected in the reflectance slope of the red-edge band and the reflectance intensity of the near-infrared band. At this time, the changes in reflectance in the visible light band are not yet significant and cannot be directly identified through conventional visible light aerial images. The normalized difference vegetation index (NDVI) is sensitive to vegetation greenness and cover status, the red-edge vegetation index is sensitive to subtle changes in chlorophyll content, and the water stress index is sensitive to leaf water content status. These three indices collectively characterize the apparent response of vegetation canopy under drought stress from different spectral mechanisms.

[0026] The three types of spectral indices are subjected to min-max normalization within the study area, mapping their values ​​to the interval between 0 and 1. Then, a weighted average is taken of the normalized indices to obtain the pixel-by-pixel spectral stress intensity. A higher spectral stress intensity value indicates a higher degree of apparent spectral stress in that pixel. The spectral stress intensity is the zero-order intensity reflecting the apparent stress level of the pixel itself, forming the first measurement dimension of subsequent pixel-based binary measurements. In this embodiment, the weighted average weights of the three indices are equal, each accounting for one-third. In other embodiments, the water stress index can be assigned a higher weight depending on the research objective to enhance the sensitivity to leaf water content status. It is important to emphasize that the calculation of the spectral stress intensity does not rely on any ground-measured chlorophyll content or leaf water content data; all inputs come from the reflectance of each channel of the canopy multispectral image. Therefore, the spectral stress intensity is a purely image-domain relative measurement.

[0027] Step S3: Calculation of canopy structural inhomogeneity and structural gradient. In step S3, based on the visible light channel texture of the corrected canopy multispectral image, the canopy structural inhomogeneity is calculated pixel by pixel. The canopy structural inhomogeneity includes canopy coverage uniformity and canopy void distribution, and the local spatial gradient of the canopy structural inhomogeneity in the pixel neighborhood is extracted as the structural gradient.

[0028] Unlike existing methods that remove structurally uneven areas such as canopy gaps as background, this step retains vegetation cover pixels and canopy gap pixels in the corrected canopy multispectral image, without removing the background from the canopy gap pixels. Specifically, pixels with a normalized vegetation index (NVI) higher than a set vegetation threshold are identified as vegetation cover pixels, and pixels with an NVI lower than the threshold are identified as canopy gap pixels. In this embodiment, the vegetation threshold is set to 0.3.

[0029] For each target pixel, take its side length as the center. The square window is used as the cell neighborhood, denoted as In this embodiment, the neighborhood radius Take 5 pixels. In the neighborhood of the stated pixel. Within, based on visible light channel texture statistics, two components are defined: one is canopy cover uniformity, which is represented by the complement of the coefficient of variation of the normalized vegetation index of vegetation cover pixels in the neighborhood of the pixel. The smaller the coefficient of variation, the more uniform the cover. The other is canopy void distribution, which is represented by the proportion of canopy void pixels in the neighborhood of the pixel.

[0030] In calculating canopy cover uniformity, this embodiment further employs a texture metric based on the gray-level co-occurrence matrix to characterize the visible light channel texture. Specifically, it calculates the contrast and angular second moment of the green light channel gray-level co-occurrence matrix within the pixel neighborhood. Contrast reflects the local variation in gray level within the neighborhood, while the angular second moment reflects the uniformity of gray level distribution within the neighborhood. The texture metric and the normalized vegetation index (NVI) coefficient of variation are normalized together, and the average value is used as the basis for measuring canopy cover uniformity. The reason for preserving and quantifying the spatial inhomogeneity of canopy structure in the early stages of drought stress is that, in addition to the spectral responses of leaf water content and chlorophyll content, vegetation under drought stress also exhibits structural changes in the canopy due to localized water loss, such as leaf curling, patchy cover, and increased gaps. These structural changes present a non-uniform distribution from the local outwards and represent another type of measurable information independent of the pixel-by-pixel spectral intensity. Existing methods generally use a co-registration normalized vegetation index thresholding method to remove structurally uneven regions such as canopy gaps as background, thus losing this type of measurable information. In contrast, this step preserves and quantifies the canopy gap distribution, making it a valid input for subsequent measurements. Based on this, the canopy structural inhomogeneity of the target pixel is calculated according to the following formula: ,in: coordinates The canopy structure inhomogeneity of the target pixel is a scalar with a value range of [0,1]. It is dimensionless and is calculated by this formula. It characterizes the degree to which the canopy structure in the neighborhood of the pixel deviates from a uniform closed canopy. The larger the value, the more inhomogeneous the structure. The row and column coordinates of the cell are integers, and the unit is pixels; The uniformity of canopy cover is a scalar quantity with a value range of [0,1]. It is dimensionless and is calculated from the complement of the normalized vegetation index variation coefficient of the vegetation cover pixels in the neighborhood of the pixel. The larger the value, the more uniform the cover. The distribution of canopy voids is a scalar with a value range of [0,1]. It is dimensionless and is calculated by the ratio of the number of canopy void pixels in the neighborhood of the given pixel to the total number of pixels in the neighborhood, representing the void ratio. To balance the weights, is a scalar with a value range of [0,1]. In this embodiment, it is set to 0.5. It is dimensionless and set empirically to balance the contribution of uneven coverage and the contribution of void ratio. An excessively large gap will weaken the effect of the gap ratio. If the value is too small, it weakens the effect of covering the degree of dispersion.

[0031] After obtaining the canopy structural inhomogeneity, its local spatial gradient in the pixel neighborhood is extracted as the structural gradient. In this embodiment, the Sobel operator is used to convolve the canopy structural inhomogeneity field in both the row and column directions to obtain the first-order partial derivatives in both directions. Then, the L2 norm is taken as the structural gradient, calculated according to the following formula: ,in: coordinates The structural gradient at point is a scalar, and its value range is . The unit is The value is calculated using this formula and represents the intensity of the stress patch boundary at that pixel. The gradient operator represents the spatial first-order partial derivative with respect to the field quantity; The non-uniformity of the canopy structure along the direction and The first-order partial derivative in the direction is a scalar with units of . It is obtained by convolution using the Sobel operator; It is the L2 norm, defined as the arithmetic square root of the sum of squares of its components; The maximum value of the structural gradient within the study area is given by , where is a scalar and the unit is . The values ​​were obtained from statistics of the study area. The structural gradient quantity is a first-order spatial differential component reflecting the boundary and orientation of the stressed patch, and it is orthogonal to the spectral stress intensity quantity in terms of measurement information, forming the second measurement dimension of the subsequent pixel binary measurement quantity.

[0032] Step S4: Coupling of structural confidence weighting with pixel binary measurement. In step S4, the spectral stress intensity and the structural gradient are used as two orthogonal measurement dimensions and coupled into a pixel binary measurement on the same pixel.

[0033] Because in areas with sparse vegetation cover, the structural gradient is easily dominated by the edge of the canopy gap, resulting in a falsely high response unrelated to stress, before coupling, the structural confidence is determined based on the proportion of vegetation cover pixels in the neighborhood of the target pixel corresponding to the spectral stress intensity. The structural confidence is positively correlated with the proportion of vegetation cover pixels in the neighborhood of the target pixel. This embodiment uses the following formula to calculate the structural confidence: ,in: coordinates The structural confidence level at a given location is a scalar with a value range of (0,1). It is dimensionless and is calculated using this formula. It characterizes the confidence level of the structural gradient of that pixel. The sigmoid function is defined as follows: This maps the real number independent variable to the interval (0,1); The independent variable of the sigmoid function is a scalar, dimensionless. Here, is the steepness coefficient, and is a scalar with a range of values. In this embodiment, 12 is chosen. This value is dimensionless and set empirically to control the width of the transition band around the structural confidence level. Too large a value will make the transition too steep, approaching a hard threshold. If the confidence level is too low, the distinction will be insufficient; The percentage of vegetation cover pixels in the neighborhood of a given pixel is a scalar with a value range of [0,1]. It is dimensionless and is calculated as the ratio of the number of pixels in the neighborhood whose normalized vegetation index is higher than the vegetation threshold to the total number of pixels in the neighborhood. It represents the sufficiency of vegetation cover. The coverage percentage threshold is a scalar with a value range of [0,1]. In this embodiment, it is set to 0.5. It is dimensionless and is set empirically. When the vegetation cover pixel percentage is lower than this threshold, the structure confidence drops rapidly. is a natural constant. The structural gradient is then weighted by the structural confidence level to obtain the weighted structural gradient, calculated according to the following formula: ,in: coordinates The weighted structure gradient at point is a scalar, and its value range is . The unit is The formula is used to calculate the structural gradient amount after suppressing the spurious response in the sparse cover region; The definition is the same as the aforementioned formula; The definition is the same as the aforementioned formula. Within the neighborhood of the target pixel, the spectral stress intensity is measured as the zero-order intensity component of the target pixel, and the weighted structural gradient is measured as the first-order spatial differential component of the canopy structural inhomogeneity along the neighborhood of the pixel. To make the two components comparable on a numerical scale, the spectral stress intensity and the weighted structural gradient are respectively subjected to minimum-maximum normalization within the study area to complete scale alignment. Then, the scale-aligned zero-order intensity component and the first-order spatial differential component are combined into two components of the pixel binary measurement, calculated according to the following formula: ,in: coordinates The pixel binary measurement at a given location is a two-dimensional vector with both components ranging from [0,1] and being dimensionless. It is calculated using this formula and represents the intensity-gradient binary expression of stress on that pixel. The zeroth-order intensity component after scale alignment is a scalar with a value range of [0,1], dimensionless, and derived from the spectral stress intensity quantity. Obtained through minimum-maximum normalization; The first-order spatial differential component after scale alignment is a scalar with a value range of [0,1], dimensionless, and is derived from the weighted structure gradient. Obtained through minimum-maximum normalization; The spectral stress intensity is defined as the same as in step S2, and is dimensionless; These are the minimum and maximum values ​​of the spectral stress intensity within the study area, respectively, and are scalars with no dimension. Let be the minimum and maximum values ​​of the weighted structural gradient within the study area, respectively, where is a scalar and the unit is . Simultaneously, a structure confidence layer composed of the structure confidence scores is output, with the structure confidence layer recording pixel by pixel. The values ​​are used by downstream researchers to assess the reliability of the structural gradient quantities of each pixel. The pixel binary measurement orthogonally couples the spectral stress intensity quantity with the structural gradient quantity, and the measurement information it carries has a dimension that is strictly higher than any single dimension.

[0034] Step S5: Stress Level Grading and Spatiotemporal Expansion Measurement Output. In Step S5, without relying on ground-based measured targets, the pixels in the study area are divided into four levels—normal state, mild stress, moderate stress, and severe stress—based on the aforementioned binary pixel measurements, and a stress spatial distribution map is output. This step further includes a sub-process of multi-flight time-series measurements to quantify the expansion rate and spatial migration direction of each stress level's region.

[0035] In single-flight grading, quantile statistics are performed on the binary measurements of all pixels in the study area. The boundary between adjacent quantile intervals is used as the relative grading boundary between two adjacent grades, thereby completing the relative grading without introducing any ground-based measured values. In this embodiment, the two components of the binary measurements of the pixels are combined into a comprehensive measurement value with equal weights. Based on the quantile of this comprehensive measurement value in the study area, the 25%, 50%, and 75% quantiles are taken as the grading boundaries between normal state and mild stress, mild stress and moderate stress, and moderate stress and severe stress. Thus, each pixel is assigned to one of the four grades, and a stress spatial distribution map is output according to geographic coordinates.

[0036] To quantify the expansion dynamics of each stress region over time, steps S1 to S4 are performed on the canopy multispectral images of the same study area collected by the multi-rotor UAV at different times to obtain the pixel binary measurement of each flight; then multi-flight registration and baseline self-normalization are performed.

[0037] Considering baseline drift caused by differences in flight altitude, illumination, and phenology among different flights, this step uses canopy void topological invariants as the relative registration benchmark. The canopy void topological invariants are the connectivity morphology characteristics of the canopy void distribution that remain unchanged under scale variations. Specifically, a continuous homogeneous filter is constructed for the canopy void indicator field of each flight, the birth and death scales of the void connectivity components are extracted, and the canopy void topological invariants are calculated according to the following formula: ,in: For the first The topological invariant of the canopy void for each flight is a scalar, and its value range is... Dimensionless, calculated by this formula, characterizing the total persistence of void connectivity morphology that remains stable under scale variations; superscript For the sequence number, It is a positive integer, ranging from 1 to... , Total number of frames; Indicates to From 1 to Summing each one individually; For the first The number of gap-connected components in the continuous cohomology of each sortie is a positive integer, calculated from the continuous cohomology; subscript The index of the gap connectivity component is an integer, ranging from 1 to 1. The first Flight number The birth and death scales of each interstitial connected component are scalars, dimensionless (normalized scales), and are obtained from continuous homogeneous filtering. This is an indicator function that takes the value 1 if the condition within the square brackets is true, and 0 otherwise. Here, is the duration threshold, and is a scalar with a value range of . In this embodiment, a value of 0.1 is used, which is dimensionless, to filter short-lifetime noise components. If the value is too large, effective structural features will be filtered out. If the value is too small, too much noise will be retained. The centroids of the gap connectivity components with durations higher than the duration threshold are used as relative registration control points. The nearest match between adjacent control points, weighted by duration, is used as the correspondence. The relative registration transformation that minimizes the control point alignment error is calculated using the following formula: ,in: For the first Relative to the number The optimal spatial registration transformation for each sortie is a 2x3 affine transformation matrix, where the rotation and scaling elements are dimensionless and the translation elements are in pixels. This matrix is ​​obtained by solving the formula and is used to register the sortie number of sorties. Image registration up to the first flight Flight coordinate system; This represents the transformation that minimizes the objective function. ; Indicates to From 1 to Summing each one individually; The number of continuous gap control points, a positive integer; subscript The control point number is an integer, ranging from 1 to... ; For the first Flight number The centroid coordinates of each persistent interstitial connected component are two-dimensional vectors, in pixels, and are determined by a continuity higher than that of the component. The centroid of the interstitial connected component is obtained; For the first In the flight and the first Flight number Each control point is weighted by its duration and the nearest matching control point number is mapped as an integer. For the first The first of the sorties The centroid coordinates of each control point are two-dimensional vectors, with units of pixels. This is the square of the L2 norm. After relative registration is completed, baseline self-normalization is performed on the pixel binary measurements of each registered flight. The baseline self-normalization is based on the pixel binary measurement statistics of the normal state files within each flight, and is calculated according to the following formula: ,in: For the first Flight coordinates The pixel binary measurement after baseline self-normalization is a two-dimensional vector, dimensionless, and is calculated by this formula. It represents the stress binary measurement after eliminating baseline drift. For the first Flight coordinates The binary measurement quantity of the pixel at that location is defined in the same step S4 and is dimensionless. For the first The mean vector of pixel binary measurements of pixels in the normal status file of a flight is a two-dimensional vector with each component taking values ​​in the range of [0,1]. It is dimensionless and is obtained by statistical analysis of all pixels in the normal status file of that flight. For the first The standard deviation vector of the pixel binary measurement of the normal status file of a flight is a two-dimensional vector, and the value range of each component is as follows: Dimensionless, obtained from the statistics of all pixels in the normal status range within this sortie; subscript 0 indicates the normal status range; division operations are performed element-by-element by vector components. After completing multi-sor registration and baseline self-normalization, spatiotemporal difference is performed on the binary measurements of the pixels in adjacent sorties to construct the stress front vector field. Each vector in the stress front vector field is determined by the spatial direction and change of the stress level in the neighborhood of the corresponding pixel, and is calculated according to the following formula: ,in: coordinates The stress front vector at a given location is a two-dimensional vector with units of . , calculated by this formula, represents the direction and speed of the stress level rising at that pixel; The change in the stress level of this pixel in adjacent flights is equal to the first... Flight frequency reduction The flight speed range is a scalar quantity, and its value ranges from integers. Dimensionless, obtained by the difference between the classification results of adjacent sorties; The stress level is a scalar quantity, taking integer values ​​of 0, 1, 2, and 3, corresponding to normal state, mild stress, moderate stress, and severe stress, respectively, and is dimensionless. Let be the time interval between adjacent flights, and be a scalar with a value range of . In this embodiment, 7 days is used, with the unit being days, which is determined by the aerial survey time. is the spatial unit direction vector for the forced gear shift, which is a two-dimensional vector, dimensionless, with a magnitude of 1, and is obtained by normalizing the gear field spatial gradient. Let be the spatial gradient of the stress gear field, which is a two-dimensional vector with units of . This is obtained by taking the first spatial partial derivative with respect to the gear field; The L2 norm is used. The expansion rate of each stress region is quantified by the divergence of the stress front vector field and calculated according to the following formula: ,in: coordinates The expansion rate at that point is a scalar, with units of . The value calculated by this formula represents the spatial expansion intensity of the stress region per unit time at the pixel. A positive value indicates that the stress region is expanding outward, and a negative value indicates that the stress region is contracting. The divergence operator represents the divergence of a vector field; The stress front vectors are respectively located at... direction and The directional component is a scalar, with units of . ; They are respectively along direction, along The first-order partial derivative of the direction. The expansion rate can be further converted into the area expansion per unit time by combining the ground resolution.

[0038] The spatial migration direction of each stress region is quantified by the principal direction of the stress front vector field. In this embodiment, the resultant vector direction of the stress front vectors within the connected region of a certain stress level is taken as the spatial migration direction of the stress region of that level, and is calculated according to the following formula: ,in: Let be the direction of spatial migration, and be a scalar with a range of values. The unit is deg, which is calculated by this formula and represents the overall spatial migration direction of the stress area at this level. This is the arctangent function for the four quadrants, which returns the azimuth angle based on the vertical and horizontal components, in deg. It is the set of all pixels within a connected region of a certain stress level; Indicates the region Summing all cells within the range; Coordinates The horizontal and vertical components of the stress front vector are scalars, with units of . Finally, the expansion rate and the spatial migration direction are superimposed on the stress spatial distribution map output. The stress spatial distribution map registers each pixel according to geographic coordinates and assigns distinguishing identifiers, calculates the area of ​​each stress region and its proportion in the study area, and superimposes the expansion rate and the spatial migration direction in the form of field quantities and vectors, outputting them together as a raster map with coordinates. It should be noted that all results output in this embodiment are strictly limited to image measurement parameters of canopy spectral characteristics and structural indicators, without providing scientific interpretations of vegetation physiological mechanisms and drought response patterns, intended for downstream researchers to use in conjunction with meteorological data and soil moisture content for comprehensive analysis.

[0039] Thus, this embodiment upgrades the monitoring of vegetation canopy under drought stress from ground-based leaf-by-leaf measurement to UAV multispectral imaging patrol measurement, and achieves objective image measurement of the spatial distribution of stress intensity and its expansion rate and spatial migration direction without relying on ground-based measurement targets.

[0040] This embodiment provides a drone-based survey and measurement system for vegetation canopy spectral images under drought stress, such as... Figure 2As shown, the system includes an image acquisition module, a spectral measurement module, a structural measurement module, a binary coupling module, and a hierarchical output module. These five modules are connected sequentially, each corresponding to steps S1 to S5 in the aforementioned method embodiment. The system can be deployed on a ground processing workstation and works collaboratively with the multi-rotor UAV via a data link. In terms of hardware deployment, the multispectral camera onboard the multi-rotor UAV records a multi-channel aerial photograph sequence via onboard storage and transmits the sequence to the ground processing workstation after landing or via a data link. The spectral measurement module, structural measurement module, binary coupling module, and hierarchical output module run on the ground processing workstation. The flight path planning unit of the image acquisition module operates before fieldwork, while the image stitching unit and its subsequent modules operate during the office processing stage, thus forming a complete inspection and measurement process from field acquisition to office measurement.

[0041] The image acquisition module is configured to control a multi-rotor UAV equipped with a multispectral camera to fly along a preset flight path, and the multispectral camera acquires canopy multispectral images of the study area. The image acquisition module includes a flight path planning unit and an image stitching unit. The flight path planning unit generates the preset flight path based on the boundary of the study area and sets the flight altitude, forward overlap rate, and lateral overlap rate. The image stitching unit stitches the multi-channel aerial image sequence acquired by the multi-rotor UAV to obtain the canopy multispectral image covering the entire study area. The specific processing logic of the image acquisition module is the same as step S1 in the aforementioned method embodiment. The multispectral camera has a visible light channel, an infrared edge channel, and a near-infrared channel. The flight altitude is selected within the range of 50m to 120m, and the forward overlap rate and lateral overlap rate are both not less than 70%. The image acquisition module outputs the canopy multispectral image to the spectral measurement module and the structural measurement module.

[0042] The spectral measurement module is configured to perform radiometric calibration and geometric correction on the canopy multispectral image, and calculate the spectral stress intensity pixel by pixel based on the red-edge channel and near-infrared channel of the canopy multispectral image. The specific processing logic of the spectral measurement module is the same as step S2 in the aforementioned method embodiment. It obtains the spectral stress intensity based on the normalized vegetation index, red-edge vegetation index, and water stress index through normalized weighting, and outputs the spectral stress intensity to the binary coupling module. At the same time, it provides the vegetation cover pixel ratio information to the binary coupling module to determine the structure confidence level.

[0043] The structure measurement module is configured to calculate canopy structural inhomogeneity pixel-by-pixel based on the visible light channel texture of the canopy multispectral image. The canopy structural inhomogeneity includes canopy cover uniformity and canopy void distribution, and the local spatial gradient of the canopy structural inhomogeneity within the pixel neighborhood is extracted as the structural gradient. The specific processing logic of the structure measurement module is the same as step S3 in the aforementioned method embodiment. It retains vegetation cover pixels and canopy void pixels without performing background removal on the canopy void pixels. First, it statistically analyzes the canopy cover uniformity and the canopy void distribution based on the visible light channel texture and synthesizes the canopy structural inhomogeneity. Then, it extracts the local spatial gradient using the Sobel operator to obtain the structural gradient, and outputs the structural gradient to the binary coupling module.

[0044] The binary coupling module is configured to couple the spectral stress intensity and the structural gradient as two orthogonal measurement dimensions onto the same pixel to form a pixel binary measurement. The specific processing logic of the binary coupling module is the same as step S4 in the aforementioned method embodiment. It first determines a structural confidence level positively correlated with the vegetation cover percentage within the pixel's neighborhood, then weights the structural gradient with the structural confidence level to obtain a weighted structural gradient. Next, it combines the spectral stress intensity and the weighted structural gradient after scale alignment to form the pixel binary measurement, and outputs a structural confidence layer composed of the structural confidence levels. The binary coupling module outputs the pixel binary measurement to the hierarchical output module.

[0045] The graded output module is configured to divide the pixels in the study area into four levels—normal state, mild stress, moderate stress, and severe stress—based on the pixel binary measurement, without relying on ground-based measured targets, and output a stress spatial distribution map. The specific processing logic of the grading output module is the same as step S5 in the aforementioned method embodiment. It performs quantile statistics on the binary measurements of all pixels in the study area and uses the boundaries of adjacent quantile intervals as relative grading boundaries to complete the single-flight grading. For multi-flight measurements, the grading output module further includes a topology registration unit, a baseline normalization unit, and a front measurement unit. The topology registration unit performs spatial registration on the images of each flight using the topological invariant of the canopy gap as the relative registration reference. The baseline normalization unit performs baseline self-normalization on the registered binary measurements using the binary measurements of pixels in the normal state range within each flight as the baseline. The front measurement unit performs spatiotemporal difference on the binary measurements of pixels after baseline self-normalization of adjacent flights to construct a stress front vector field. It quantifies the expansion rate of each stress region using the divergence of the stress front vector field and quantifies the spatial migration direction of each stress region using the principal direction of the stress front vector field. Finally, it superimposes the expansion rate and the spatial migration direction onto the stress spatial distribution. Figure 1 And output it.

[0046] It should be noted that the division of modules in the above system embodiment is only a logical functional division. In actual implementation, there may be other division methods. For example, multiple modules may be combined or integrated into another system, or some features may be omitted or not executed. Each module can be implemented in hardware or as a software functional unit. All results output by the system are strictly limited to image measurement parameters of canopy spectral characteristics and structural indicators, without any scientific interpretation of vegetation physiological mechanisms and drought response patterns.

[0047] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for UAV-based surveying and measurement of vegetation canopy spectral images under drought stress, characterized in that... include: A multi-rotor UAV equipped with a multispectral camera flies along a preset route, and the multispectral camera collects canopy multispectral images of the study area. Radiometric calibration and geometric correction are performed on the canopy multispectral image, and the spectral stress intensity is calculated pixel by pixel based on the red edge channel and near-infrared channel of the corrected canopy multispectral image; Based on the visible light channel texture of the corrected canopy multispectral image, the canopy structural inhomogeneity is calculated pixel by pixel. The canopy structural inhomogeneity includes canopy coverage uniformity and canopy void distribution. The local spatial gradient of the canopy structural inhomogeneity in the pixel neighborhood is extracted as the structural gradient quantity. The spectral stress intensity and the structural gradient are used as two orthogonal measurement dimensions and coupled into a pixel binary measurement on the same pixel. Without relying on ground-based measured targets, the pixels in the study area are divided into four levels—normal state, mild stress, moderate stress, and severe stress—based on the aforementioned binary pixel measurement, and a stress spatial distribution map is output.

2. The method for UAV-based surveying and measurement of vegetation canopy spectral images under drought stress according to claim 1, characterized in that, The pixel-by-pixel calculation of canopy structure inhomogeneity includes: retaining vegetation cover pixels and canopy void pixels in the corrected canopy multispectral image, without performing background removal on the canopy void pixels; using the canopy cover uniformity to characterize the dispersion of vegetation cover pixels in the neighborhood of the pixel, using the canopy void distribution to characterize the proportion and connectivity of canopy void pixels in the neighborhood of the pixel, and outputting the canopy structure inhomogeneity as a measurement independent of the spectral stress intensity.

3. The method for UAV-based surveying and measurement of vegetation canopy spectral images under drought stress according to claim 2, characterized in that, The coupling of the same pixel into a binary measurement includes: within the neighborhood of the target pixel, measuring the spectral stress intensity as the zero-order intensity component of the target pixel, and measuring the structural gradient as the first-order spatial differential component of the canopy structural inhomogeneity along the neighborhood of the pixel; and combining the zero-order intensity component and the first-order spatial differential component after scale alignment to form two components of the binary measurement.

4. The method for UAV-based surveying and measurement of vegetation canopy spectral images under drought stress according to claim 3, characterized in that, Also includes: The structural confidence level is determined based on the spectral stress intensity of the target pixel, and the structural confidence level is positively correlated with the proportion of vegetation cover pixels in the neighborhood of the pixel. The first-order spatial differential component is weighted by the structural confidence to obtain a weighted structural gradient, and the first-order spatial differential component in the pixel binary measurement is replaced by the weighted structural gradient; a structural confidence layer composed of the structural confidence is output.

5. The method for UAV-based surveying and measurement of vegetation canopy spectral images under drought stress according to claim 4, characterized in that, It also includes: extracting canopy void topological invariants from two or more canopy multispectral images of the same study area collected by the multi-rotor UAV at different times, wherein the canopy void topological invariants are the connectivity morphological features of the canopy void distribution that remain unchanged under scale changes; performing spatial registration of the images of each flight using the canopy void topological invariants as the relative registration reference; and performing baseline self-normalization on the pixel binary measurements of each flight after registration, wherein the baseline self-normalization uses the pixel binary measurement statistics of normal state pixels within each flight as the baseline.

6. The method for UAV-based surveying and measurement of vegetation canopy spectral images under drought stress according to claim 5, characterized in that, Also includes: Spatiotemporal difference is performed on the binary measurements of adjacent flights after baseline self-normalization to construct a stress front vector field. Each vector of the stress front vector field is determined by the spatial direction and change of the stress level in the neighborhood of the corresponding pixel. The expansion rate of each stress level is quantified by the divergence of the stress front vector field, and the spatial migration direction of each stress level is quantified by the principal direction of the stress front vector field. The expansion rate and the spatial migration direction are then superimposed on the stress spatial distribution map for output.

7. The method for UAV-based surveying and measurement of vegetation canopy spectral images under drought stress according to claim 1, characterized in that, The multispectral camera has a visible light channel, an infrared channel, and a near-infrared channel; the flight altitude of the preset flight path is 50m to 120m, the forward overlap rate and the lateral overlap rate are both not less than 70%, and the corresponding ground resolution is 2cm to 10cm.

8. The method for UAV-based surveying and measurement of vegetation canopy spectral images under drought stress according to claim 1, characterized in that, The classification into four levels—normal state, mild stress, moderate stress, and severe stress—includes: performing quantile statistics on the binary measurements of all pixels in the study area without introducing ground-based measured values, and using the boundary of adjacent quantile intervals as the relative classification boundary between two adjacent levels.

9. The method for UAV-based surveying and measurement of vegetation canopy spectral images under drought stress according to claim 1, characterized in that, The output stress spatial distribution map includes: registering each pixel according to geographic coordinates and assigning distinguishing labels, calculating the area of ​​each stress region and its proportion in the study area, and outputting it together with the stress spatial distribution map as a raster map with coordinates.

10. A drone-based surveying and measurement system for vegetation canopy spectral images under drought stress, used to implement the drone-based surveying and measurement method for vegetation canopy spectral images under drought stress as described in any one of claims 1-9, characterized in that, include: The image acquisition module is configured to control a multi-rotor UAV equipped with a multispectral camera to fly along a preset route, and the multispectral camera acquires canopy multispectral images of the study area. The spectral measurement module is configured to perform radiometric calibration and geometric correction on the canopy multispectral image, and to calculate the spectral stress intensity pixel by pixel based on the red edge channel and near-infrared channel of the canopy multispectral image; The structure measurement module is configured to calculate the canopy structural inhomogeneity pixel by pixel based on the visible light channel texture of the canopy multispectral image. The canopy structural inhomogeneity includes the canopy coverage uniformity and the canopy void distribution. The module also extracts the local spatial gradient of the canopy structural inhomogeneity in the pixel neighborhood as the structural gradient. The binary coupling module is configured to couple the spectral stress intensity and the structural gradient as two orthogonal measurement dimensions on the same pixel into a pixel binary measurement quantity. The hierarchical output module is configured to divide the pixels in the study area into four levels—normal state, mild stress, moderate stress, and severe stress—based on the pixel binary measurement, without relying on ground-based measured targets, and output a stress spatial distribution map.

Images