Vegetation stress continuous monitoring-oriented hyperspectral data quality adaptive evaluation method

Abstract

The embodiment of the invention provides a vegetation stress continuous monitoring-oriented hyperspectral data quality adaptive evaluation method, and relates to the technical field of intelligent agriculture, and the method comprises the steps: obtaining single-time-phase hyperspectral data, and executing stress sensitive wave band static stability analysis; a stress detectability index is output through virtual stress end member forgery; and carrying out weighted fusion to generate a stress detection confidence thermodynamic diagram. Carrying out incremental diagnosis on the output time sequence deviation degree; backtracking historical data of the growing season to generate a time sequence confidence thermodynamic diagram; executing stress evolution tracking after splicing, and outputting an evolution thermodynamic diagram and staged separability; and three-dimensionally fusing the static index, the time sequence deviation degree and the separability, and outputting a space-time fusion stress confidence thermodynamic diagram. The paradox problem of false qualification of data quality and actual missing detection of stress characteristics caused by a general quality evaluation method in the prior art is solved. And the effects of realizing visual early warning in the vegetation stress development stage and providing a priori criterion for early intervention are achieved.

Description

Technical Field

[0001] This invention relates to the field of smart agriculture technology, and in particular to an adaptive evaluation method for the quality of hyperspectral data for continuous monitoring of vegetation stress. Background Technology

[0002] Existing hyperspectral data quality assessment methods have fundamental flaws in early-stage vegetation stress detection scenarios. The core problem lies in the severe disconnect between the assessment framework and the weak spectral response characteristics of vegetation.

[0003] Specifically, the general approach mechanically uses apparent indicators such as the overall signal-to-noise ratio, which not only fails to identify local noise residues in stress-sensitive areas, but also completely ignores the spectral discernibility requirements of key early stress indicators such as sub-nanometer red-edge shifts and changes in weak absorption characteristics.

[0004] Especially when the overall signal-to-noise ratio of the data meets the standard, the slight red edge displacement cannot be accurately extracted due to residual atmospheric noise or system errors in the red edge area. This creates a paradox between the falsely qualified data quality and the actual missed detection of stress characteristics, ultimately causing the evaluation results to be completely decoupled from the early stress detection rate, forming a blind spot that restricts the reliability of precision agriculture decision-making.

[0005] It should be noted that the information disclosed in this background section is intended only to enhance the understanding of the overall background of the invention and should not be construed as an admission or in any way implying that the information constitutes prior art known to those skilled in the art. Summary of the Invention

[0006] To address the aforementioned deficiencies or improvement needs of existing technologies, this invention provides an adaptive evaluation method for hyperspectral data quality in continuous monitoring of vegetation stress. This method resolves the paradox of existing general quality assessment methods failing to identify residual local noise in stress-sensitive areas due to their disconnect from the weak spectral response characteristics of vegetation, resulting in falsely satisfactory data quality and missed detection of actual stress characteristics. It achieves visualized early warning of vegetation stress development stages at the data quality level and directly maps the detectability probability of different stress stages using spatiotemporal fusion heatmaps, providing a priori criteria for early intervention. The specific technical solution is as follows:

[0007] This invention provides an adaptive evaluation method for the quality of hyperspectral data for continuous monitoring of vegetation stress, the method comprising:

[0008] After acquiring single-temporal hyperspectral data of the target region, stress-sensitive band quality analysis is performed, and a real-time static stability index is output. Stress identification of the single-temporal hyperspectral data is evaluated using virtual stress end-member forgery, and a real-time stress detectability index is output. After weighted fusion of the real-time static stability index and the real-time stress detectability index, spatial mapping reconstruction is performed, and a stress detection confidence heatmap is output. Incremental quality diagnosis is performed on the stress detection confidence heatmap, and temporal deviation is output. The single-temporal hyperspectral data is then used to... Starting from the timestamp of spectral data acquisition, a historical reconstruction of the stress confidence history of the target region is performed to obtain a historical time-series confidence heatmap. After stitching the historical time-series confidence heatmap and the stress detection confidence heatmap, stress stage evolution situation tracking is performed to output a stage evolution heatmap. Based on the stress evolution type identified in the stage evolution heatmap, the stage separability of progressive stress intensity is generated. The static stability index, temporal deviation, and stage separability are fused in three dimensions to output a spatiotemporal fusion stress confidence heatmap.

[0009] In one implementation, after splicing the historical time-series confidence heatmap and the stress detection confidence heatmap, stress stage evolution situation tracking is performed, and a stage evolution heatmap is output. The following processing is also performed:

[0010] After spatiotemporally aligning the historical time-series confidence heatmap and the stress detection confidence heatmap, resolution normalization is performed to obtain a time-normalized confidence heatmap. Based on a preset stress stage classification threshold, the pixel values ​​of the time-normalized confidence heatmap are mapped to multi-level stress state codes. Using the multi-level stress state codes as dynamic state carriers, the transition direction of the time-normalized confidence heatmap is compared pixel by pixel to identify the stress evolution type, and an evolution state matrix is ​​output. The spatiotemporal aggregation intensity mapping of the same type of evolution pixels is performed on the evolution state matrix to output the stage evolution heatmap.

[0011] In one implementation, incremental quality diagnosis is performed on the stress detection confidence heatmap to output the time series deviation, and the following processing is also performed:

[0012] Based on the single-phase hyperspectral data acquisition timestamp, historical hyperspectral data of the same phenological period in the target area are retrieved to construct a multi-phase confidence heatmap; based on the multi-phase confidence heatmap, incremental quality diagnosis of the stress detection confidence heatmap is performed, and the temporal deviation is output.

[0013] In one implementation, after acquiring single-temporal hyperspectral data of the target region, stress-sensitive band quality analysis is performed, and a real-time static stability index is output. The following processing is also performed:

[0014] Pure vegetation cover pixels are extracted from the single-phase hyperspectral data based on the vegetation index; stress-sensitive band data are extracted from the pure vegetation cover pixels with reference to a preset vegetation stress-sensitive band mapping table, wherein the stress-sensitive band data includes red-edge region data, weak absorption band data, and photosynthetic feature point data; multi-threaded parallel stability quantization is performed on the red-edge region data, weak absorption band data, and photosynthetic feature point data to output the red-edge stability index, weak absorption signal-to-noise ratio index, and continuum stability index, which constitute the real-time static stability index.

[0015] In one implementation, incremental quality diagnosis of the stress detection confidence heatmap is performed based on the multi-temporal confidence heatmap, and the time series deviation is output. The following processing is also performed:

[0016] Based on the multi-temporal confidence heatmap, a dynamic threshold interval for historical sensitive indicators is constructed; the red-edge stability index is calculated relative to the dynamic threshold interval for historical sensitive indicators; the weak absorption signal-to-noise ratio index is calculated relative to the dynamic threshold interval for historical sensitive indicators; after fusing the red-edge deviation and the weak absorption deviation, the continuum stability index is used for spatial autocorrelation correction, and the temporal deviation is output.

[0017] In one implementation, the stress identifiability evaluation of the single-temporal hyperspectral data is performed by forging virtual stress end-members, outputting a real-time stress detectability index, and the following processing is also performed:

[0018] A preset NDVI baseline is used to extract continuous pixel regions of healthy vegetation from the single-phase hyperspectral data. The average reflectance spectrum of these regions is calculated to output a baseline healthy endmember spectral curve. A stress simulation is performed on the baseline healthy endmember spectral curve based on a preset perturbation scale, outputting red-edge stress-type and water-stress-type endmember spectral curves. Using the baseline healthy endmember spectral curve as a reference, the spectral distinguishability of the red-edge stress-type and water-stress-type endmember spectral curves is quantified. The system outputs the red-edge stress type normalized difference index, the water stress type normalized difference index, the red-edge stress type spectral angular distance, and the water stress type spectral angular distance; based on the stress type mapping, it fuses the red-edge stress type normalized difference index, the water stress type normalized difference index, the red-edge stress type spectral angular distance, and the water stress type spectral angular distance to output the red-edge stress type detectability index and the water stress type detectability index; by comparing the red-edge stress type detectability index and the water stress type detectability index, it filters and outputs the real-time stress detectability index.

[0019] In one implementation, after weighted fusion of the real-time static stability index and the real-time stress detectability index, spatial mapping restoration is performed to output a stress detection confidence heatmap, and the following processing is also performed:

[0020] The red-edge stability index, weak absorption signal-to-noise ratio index, and continuum stability index in the real-time static stability index are weighted and fused using preset weights to output a static stability value. The static stability value and the real-time stress detectability index are adaptively weighted and fused to output a vegetation stress detection quality index. The vegetation stress detection quality index is mapped to the vegetation pixel spatial location of the single-temporal hyperspectral data, and spatial interpolation is performed to fill in the spatial location of non-vegetation pixels to obtain a vegetation stress quality distribution map. After normalization processing of the vegetation stress quality distribution map, pseudo-color mapping is performed to output the stress detection confidence heatmap.

[0021] In one implementation, the static stability index, temporal deviation, and stage separability are fused in three dimensions to output a spatiotemporal fused stress confidence heatmap, and the following processing is also performed:

[0022] The static stability index is normalized globally to generate a static stability normalized value; the temporal quality score is calculated based on the temporal deviation; the stage separability is mapped to a separability score; real-time adaptive weights are configured according to the stress dominance type of the stage evolution heatmap, and the static stability normalized value, temporal quality score, and separability score are weighted and fused to obtain a spatiotemporal fusion quality index; after mapping the real-time confidence level of the spatiotemporal fusion quality index according to the multi-level confidence threshold, pseudo-color rendering is performed according to the real-time confidence level to output the spatiotemporal fusion stress confidence heatmap.

[0023] Beneficial effects of the embodiments of the present invention:

[0024] In the solution provided by this invention, after acquiring single-temporal hyperspectral data of the target region, stress-sensitive band quality analysis is performed to output a real-time static stability index; stress identification evaluation of the single-temporal hyperspectral data is performed by forging virtual stress endmembers to output a real-time stress detectability index; after weighted fusion of the real-time static stability index and the real-time stress detectability index, spatial mapping restoration is performed to output a stress detection confidence heatmap; incremental quality diagnosis is performed on the stress detection confidence heatmap to output temporal deviation. Starting from the timestamp of the single-phase hyperspectral data acquisition, a historical reconstruction of the stress confidence history of the target area is performed to obtain a historical time-series confidence heatmap. After stitching the historical time-series confidence heatmap and the stress detection confidence heatmap, stress stage evolution situation tracking is performed to output a stage evolution heatmap. Based on the stress evolution type identified in the stage evolution heatmap, the stage separability of progressive stress intensity is generated. The static stability index, temporal deviation, and stage separability are fused in three dimensions to output a spatiotemporal fusion stress confidence heatmap. This achieves the effect of visually warning the development stage of vegetation stress at the data quality level, and directly mapping the detectability probability of different stress stages with a spatiotemporal fusion heatmap, providing a priori criteria for early intervention. Of course, implementing any product or method of this invention does not necessarily require achieving all of the above advantages simultaneously. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art 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.

[0026] Figure 1 A schematic diagram of the process for the adaptive evaluation method of hyperspectral data quality for continuous monitoring of vegetation stress provided by the present invention is shown.

[0027] Figure 2 This paper illustrates a flowchart of the generation stage evolution heatmap in the hyperspectral data quality adaptive evaluation method for continuous monitoring of vegetation stress provided by the present invention. Detailed Implementation

[0028] To facilitate understanding of the present invention, a more complete description of the invention will be given below with reference to the accompanying drawings, which illustrate preferred embodiments of the invention. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein; rather, these embodiments are provided to enable a more thorough and complete understanding of the disclosure of the invention.

[0029] Furthermore, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0030] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," and "counterclockwise," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0031] Unless otherwise expressly stated, throughout the specification and claims, the term "comprising" or its variations such as "including" or "comprises" shall be understood to include the stated elements or components without excluding other elements or other components.

[0032] The present invention provides an adaptive evaluation method for the quality of hyperspectral data for continuous monitoring of vegetation stress. This method addresses the paradox of existing general quality assessment methods failing to identify local noise residues in stress-sensitive areas due to their disconnect from the weak spectral response characteristics of vegetation, resulting in falsely qualified data quality and actual missed detection of stress characteristics.

[0033] Example: See Figure 1 The flowchart of the adaptive evaluation method for hyperspectral data quality for continuous monitoring of vegetation stress provided in this embodiment of the invention includes:

[0034] P100: After acquiring single-phase hyperspectral data of the target area, perform stress-sensitive band quality analysis and output real-time static stability index.

[0035] In one implementation, after acquiring single-temporal hyperspectral data of the target region, stress-sensitive band quality analysis is performed, and real-time static stability indices are output. Step P100 may further include:

[0036] P110: Extract pure vegetation cover pixels from the single-phase hyperspectral data based on the vegetation index.

[0037] P120: Refer to the preset vegetation stress-sensitive band mapping table, extract stress-sensitive band data from the pure vegetation cover pixels, wherein the stress-sensitive band data includes red edge region data, weak absorption band data and photosynthetic feature point data.

[0038] P130: Perform multi-threaded parallel stability quantization on the red-edge region data, weak absorption band data, and photosynthetic feature point data, and output the red-edge stability index, weak absorption signal-to-noise ratio index, and continuum stability index to form the real-time static stability index.

[0039] Specifically, the target area is a vegetation-covered area such as farmland or forest. Single-temporal hyperspectral data of the target area is obtained through hyperspectral data acquisition technology including but not limited to airborne hyperspectral imagers or spaceborne hyperspectral sensors. The single-temporal hyperspectral data is a cube image containing full-spectrum reflectance data from visible light to short-wave infrared bands at the current time.

[0040] Dynamic segmentation thresholds are constructed based on spectral characteristics such as normalized difference vegetation index. An adaptive threshold is set by the ratio of red light to near-infrared reflectance, and continuous vegetation patches that meet the leaf area index criteria are iteratively screened.

[0041] Furthermore, high-purity photosynthetic active pixels are directly extracted from continuous vegetation patches, and topographic elevation data and historical vegetation cover maps are simultaneously integrated during the processing to eliminate interference from abnormal areas such as cloud cover and snow cover. Finally, the pure vegetation cover pixels with continuous spatial distribution and consistent spectral response are output, providing a standardized input carrier for subsequent stress-sensitive band analysis.

[0042] A pre-constructed vegetation stress sensitive band mapping table is provided. The specific mapping rules of the vegetation stress sensitive band mapping table include: locking the 680-750nm wavelength range in the red edge region to capture the red-edge blue shift phenomenon caused by chlorophyll decomposition; covering the weak absorption bands such as 970nm and 1200nm, which are weak absorption valleys of liquid water molecules, specifically for detecting changes in absorption depth caused by water stress; and accurately locating the 550nm green peak and the 670nm chlorophyll absorption valley of photosynthetic feature points to quantify the fluctuation of photosynthetic pigment concentration.

[0043] Meanwhile, the vegetation stress sensitive band mapping table described in this embodiment supports dynamic adjustment of band boundaries according to different vegetation types. For example, crops focus on the red edge zone while forest areas strengthen the water absorption zone.

[0044] The reflectance data of the aforementioned sensitive bands are extracted from the pure vegetation cover pixels with reference to the vegetation stress sensitive band mapping table. Spatial coordinates and phenological parameters are recorded simultaneously to construct structured stress sensitive band data, which includes red edge region data, weak absorption band data and photosynthetic feature point data.

[0045] Leveraging the advantages of parallel computing strategies and GPU acceleration technology to achieve millisecond-level response and meet real-time requirements, this embodiment employs a distributed computing architecture to synchronously process three types of sensitive bands. Specifically, based on the red-edge region data, the wavelength standard deviation of the maximum first derivative of all vegetation pixels is calculated, and the red-edge stability index, which characterizes the degree of red-edge position drift dispersion, is output. For the weak absorption band data, the ratio of the absorption valley depth to the noise variance of adjacent bands is calculated using a sliding window, generating the weak absorption signal-to-noise ratio index. For the photosynthetic feature point data, the reflectance envelope is fitted using the continuum removal method, and the coefficient of variation of the feature point reflectance value in the same type of vegetation is evaluated, outputting the continuum stability index.

[0046] The red-edge stability index, weak absorption signal-to-noise ratio index, and continuum stability index together form a real-time static stability index, which respectively reflects the resolvability of red-edge features, sensitivity to water stress, and radiation consistency.

[0047] This embodiment generates a real-time anti-interference capability evaluation index by extracting the stability characteristics of the stress-sensitive bands in single-phase hyperspectral data, providing a key quantitative benchmark for subsequent stress spectral identification evaluation and comprehensive stress sensitivity quality index generation.

[0048] P200: The stress identification of the single-phase hyperspectral data is evaluated by forging virtual stress end-members, and the real-time stress detectability index is output.

[0049] In one implementation, the stress identifiability evaluation of the single-temporal hyperspectral data is performed by forging virtual stress end-members, and a real-time stress detectability index is output. Step P200 may further include:

[0050] P210: The preset NDVI baseline is used to extract continuous pixel areas of healthy vegetation from the single-phase hyperspectral data.

[0051] P220: Calculate the average reflectance spectrum of the continuous pixel area of ​​the healthy vegetation and output the baseline healthy endmember spectrum curve.

[0052] P230: Perform stress simulation on the baseline healthy endmember spectral curve based on a preset perturbation scale, and output the red-edge stress type endmember spectral curve and the moisture stress type endmember spectral curve.

[0053] P240: Using the baseline healthy endmember spectral curve as a reference, perform spectral distinguishability quantification on the red-edge stress type endmember spectral curve and the water stress type endmember spectral curve, and output the red-edge stress type normalized difference index, the water stress type normalized difference index, the red-edge stress type spectral angular distance, and the water stress type spectral angular distance.

[0054] P250: Based on the stress type mapping, the normalized difference index of red-edge stress type, the normalized difference index of water stress type, the spectral angular distance of red-edge stress type, and the spectral angular distance of water stress type are fused to output the detectability index of red-edge stress type and the detectability index of water stress type.

[0055] P260: Compare the red-edge stress detectability index and the moisture stress detectability index, and filter and output the real-time stress detectability index.

[0056] It should be understood that the NDVI value is the result of calculating the ratio of red light to near-infrared reflectance. In this embodiment, the NDVI baseline refers to the normalized difference vegetation index threshold range dynamically calibrated based on the historical phenological characteristics and healthy physiological state of the vegetation type planted in the target area. The NDVI baseline serves as a core screening criterion for extracting continuous pixel regions of healthy vegetation from single-temporal hyperspectral data.

[0057] Specifically, the NDVI value of each pixel in the image of the single-phase hyperspectral data is compared with the NDVI baseline range, and only pixels within the NDVI baseline range are retained to obtain the initial screened pixels.

[0058] Spatial clustering is performed on the initial screened pixels to remove broken areas with an area smaller than the minimum patch size threshold, resulting in continuous pixel areas with distortion. Then, spectral angle matching technology is used to filter these continuous pixel areas a second time to eliminate spectral distortion pixels caused by edge mixing or shadow interference. The final output of the healthy vegetation continuous pixel area meets the following requirements: NDVI value is within the NDVI baseline range, vegetation coverage is >95%, and spatial aggregation meets the standard.

[0059] A baseline healthy endmember spectral curve is generated by performing average reflectance spectral calculations on continuous pixel areas of healthy vegetation. This calculation process adopts a spatial weighted averaging strategy, assigning differentiated weights based on the spatial clustering of pixels. Pixels with high clustering are given higher weights to suppress discrete noise interference. Atmospheric correction residual error compensation is performed simultaneously, linear correction is applied to aerosol scattering residues, and nonlinear compensation is applied to the water vapor absorption band to restore the true surface reflectance. The final generated baseline healthy endmember spectral curve covers the entire 400-2500nm band, fully including the 550nm green peak reflectance peak, the 670nm chlorophyll absorption valley characteristics, and the typical morphology of the near-infrared 700-1300nm high plateau region.

[0060] The baseline healthy endmember spectral curve serves as the original reference for virtual stress forgery, and its reflectance value and spectral morphology characteristics directly affect the accuracy of subsequent quantification of the distinguishability between healthy and stressed spectra.

[0061] The preset perturbation scale refers to the intensity level of spectral variation defined according to the physiological mechanism of vegetation stress. Preferably, in this embodiment, for red-edge stress type, a gradual blue-shift perturbation is applied in the 680-750nm range, and the wavelength shift is calculated based on the chlorophyll degradation model. For example, mild stress results in a 3nm blue shift, and severe stress results in an 8nm blue shift. The red-edge slope is compressed simultaneously to simulate the characteristics of chlorophyll decomposition. For water stress type, reflectance enhancement is performed in the liquid water absorption valleys at 970nm and 1200nm. The absorption depth attenuation ratio is quantified according to the water stress index model. For example, a reflectance enhancement of 0.5% to 2% simulates stomatal closure.

[0062] Finally, stress simulation was performed. The output red-edge stress-type endmember spectral curves showed a red-edge blue shift and a decrease in slope, while the water stress-type endmember spectral curves showed a weak absorption valley filling phenomenon. Both follow the physical laws of plant spectral response.

[0063] Extract the reflectance R700 at 700nm wavelength from the baseline healthy endmember spectral curve and the reflectance R730 at 730nm wavelength from the red-edge stress type endmember spectral curve. Calculate the red-edge position offset using the formula (R730-R700) / (R730+R700) and output the normalized difference index for the red-edge stress type.

[0064] The reflectance at 970 nm wavelength of the baseline healthy endmember spectral curve (R970) and the reflectance at 980 nm wavelength of the water stress type endmember spectral curve (R980) are extracted. The variation intensity of the water absorption valley is quantified according to the formula (R980-R970) / (R980+R970), and the normalized difference index of the water stress type is output.

[0065] The baseline healthy endmember spectral curve and the red-edge stressed endmember spectral curve are regarded as multidimensional vectors. The cosine value of the vector space angle between the two in the 680-750nm range is calculated, and the red-edge stressed spectral angular distance is output.

[0066] The baseline healthy endmember spectral curve and the water stress type endmember spectral curve are regarded as multidimensional vectors. The cosine value of the vector space angle between the two in the 900-1300nm range is calculated, and the water stress type spectral angular distance is output.

[0067] The weights are assigned based on the sensitivity differences of different stress mechanisms. Specifically, for red edge stress scenarios, the red edge position difference index accounts for 70% and dominates the fusion process because it is directly related to changes in chlorophyll concentration; the spectral angle distance accounts for 30% to supplement the assessment of overall spectral variation; for water stress scenarios, the absorption variation index accounts for 60% and the spectral angle distance accounts for 40%.

[0068] Preferably, the fusion process introduces noise floor correction, using the local band standard deviation of healthy pixels as a noise reference. When the distinguishability parameter is close to the noise level, it is automatically downweighted. The resulting red-edge stress detectability index is used to quantify the data's ability to detect chlorophyll-related stresses, while the water stress detectability index specializes in water stress response assessment. Both are in the 0-1 range, with higher values ​​indicating that stress features are more easily identified.

[0069] The comparison and screening mechanism makes dynamic decisions based on the dominant stress type in the target area. When the red edge stress type index is more than 20% higher than the water stress type index, it is determined that the current data is more suitable for chlorophyll stress detection and the red edge index is directly output; otherwise, the water index is output.

[0070] When the difference between the red-edge stress detectability index and the water stress detectability index is less than a threshold, a weighted average fusion is used. The weights are predefined by the vegetation type library. For example, the red-edge weight for crop scenes is 0.7, and the water weight for forest areas is 0.6.

[0071] The final output of the real-time stress detectability index, as a strong correlation indicator between data quality and stress detection rate, directly determines the rendering logic of the subsequent stress detection confidence heatmap. This implementation analysis effectively avoids the deficiency of general evaluation methods in terms of insufficient sensitivity to specific stress types.

[0072] P300: After weighted fusion of the real-time static stability index and the real-time stress detectability index, spatial mapping restoration is performed to output a stress detection confidence heatmap.

[0073] In one implementation, after weighted fusion of the real-time static stability index and the real-time stress detectability index, spatial mapping reconstruction is performed to output a stress detection confidence heatmap. Step P300 may further include:

[0074] P310: The red-edge stability index, weak absorption signal-to-noise ratio index, and continuum stability index in the real-time static stability index are weighted and fused using preset weights to output a static stability value.

[0075] P320: Adaptive weighted fusion of the static stable value and the real-time stress detectability index to output the vegetation stress detection quality index.

[0076] P330: Map the vegetation stress detection quality index to the vegetation pixel spatial location of the single-phase hyperspectral data, and perform spatial interpolation to fill in the non-vegetation pixel spatial location to obtain a vegetation stress quality distribution map.

[0077] P340: After normalizing the vegetation stress quality distribution map, perform pseudo-color mapping and output the stress detection confidence heatmap.

[0078] Specifically, in this embodiment, the preset weights are predefined by the target vegetation type and application scenario. For example, in the crop scenario, the red edge stability weight can be configured as 0.5, the weak absorption signal-to-noise ratio weight as 0.3, and the continuum stability weight as 0.2.

[0079] Weighted fusion refers to the linear superposition of three indicators according to weight coefficients: static stability value = w1 × red edge stability index + w2 × weak absorption signal-to-noise ratio index + w3 × continuum stability index. The static stability value obtained by weighted fusion comprehensively reflects the static anti-interference capability of single-phase data.

[0080] Adaptive weighting refers to dynamically adjusting the weights based on the dominant stress type. Specifically, when the detectability index of red-edge stress is high, the weight of the static stable value is increased to 0.7; when water stress is dominant, the weight of the real-time stress detectability index is increased to 0.6.

[0081] The vegetation stress detection quality index is calculated using the formula VSD=α×static stability value+β×real-time stress detectability index, where α+β=1, and α and β are weights. The vegetation stress detection quality index is directly related to the reliability of data in detecting the current dominant stress.

[0082] The vegetation stress detection quality index is spatially mapped to the specific vegetation pixel location coordinates in the single-temporal hyperspectral data. Here, vegetation pixels refer to pure vegetation cover pixels extracted through previous vegetation index segmentation. A geographic coordinate matching algorithm is used to establish a one-to-one correspondence between the quality index value and the geographic coordinates of each vegetation pixel. For invalid pixel areas outside the vegetation mask in the image, including areas covered by land cover types such as bare soil, water bodies, and buildings, an inverse distance weighted interpolation algorithm is used to fill them in based on spatial distance attenuation weight.

[0083] Finally, a raster-based vegetation stress quality distribution map covering the entire target area is generated. The number of rows and columns of this raster dataset is strictly aligned with the original hyperspectral image. Each cell stores the vegetation stress detection quality index value after spatial mapping or interpolation, realizing the transformation from abstract quality indicators to geospatial visualization.

[0084] Normalization linearly transforms the vegetation stress detection quality index values ​​to the [0,1] interval, with higher values ​​indicating better data quality. Pseudo-color mapping is applied based on the normalized values, with the following gradations: 0.8-1.0 maps to dark green (very high confidence), 0.6-0.8 to light green (high confidence), 0.4-0.6 to yellow (medium confidence), 0.2-0.4 to orange (low confidence), and 0-0.2 to red (very low confidence).

[0085] In this embodiment, after normalizing the vegetation stress quality distribution map, pseudo-color mapping is performed to output the stress detection confidence heatmap. The stress detection confidence heatmap is finally superimposed on the original image as a semi-transparent layer to intuitively display the reliability level of the data for stress detection in each region.

[0086] This embodiment spatially visualizes the quality assessment results of vegetation stress data. Compared with the shortcomings of existing technologies where the overall signal-to-noise ratio assessment is disconnected from the stress-sensitive area, it achieves the technical effect of visual early warning and detection probability mapping of stress development stages. At the same time, it provides adaptive weighted fusion quantitative input for subsequent continuous tracking of stress evolution.

[0087] P400: Perform incremental quality diagnosis on the stress detection confidence heatmap and output the time series deviation.

[0088] In one implementation, incremental quality diagnosis is performed on the stress detection confidence heatmap to output the time series deviation. Step P400 may further include:

[0089] P410: Based on the timestamp of the single-phase hyperspectral data acquisition, retrieve historical hyperspectral data of the same phenological period in the target area and construct a multi-phase confidence heatmap.

[0090] P420: Based on the multi-phase confidence heatmap, perform incremental quality diagnosis of the stress detection confidence heatmap and output the time series deviation.

[0091] In one implementation, incremental quality diagnosis of the stress detection confidence heatmap is performed based on the multi-temporal confidence heatmap, and the temporal deviation is output. Step P420 may further include:

[0092] P421: Construct a dynamic threshold range for historical sensitive indicators based on the multi-temporal confidence heatmap.

[0093] P422: Calculate the red-edge deviation of the red-edge stability index relative to the dynamic threshold range of the historical sensitivity index.

[0094] P423: Calculate the weak absorption deviation of the weak absorption signal-to-noise ratio index relative to the dynamic threshold range of the historical sensitivity index.

[0095] P424: After fusing the red edge deviation and weak absorption deviation, the spatial autocorrelation correction is performed using the continuum stability index, and the temporal deviation is output.

[0096] This embodiment is used to perform time-series noise disturbance assessment on the current single-phase hyperspectral data. The noise disturbance assessment here adopts incremental quality diagnosis. Specifically, the incremental diagnosis refers to constructing a dynamic baseline based on historical phenological data and quantifying the degree of abnormal deviation of the current data's key indicators from the historical baseline.

[0097] The time-series deviation, as an output parameter of incremental diagnostic assessment, comprehensively reflects the cumulative impact of time-series noise such as atmospheric disturbance and sensor drift on the reliability of stress detection.

[0098] Specifically, in this embodiment, the historical hyperspectral data of the target area at the same phenological period is retrieved based on the timestamp of the single-phase hyperspectral data acquisition. The same phenological period refers to a historical period that strictly matches the vegetation growth stage, such as the heading stage of winter wheat, which is used to limit the time to ensure that the physiological state of historical data and real-time data is comparable.

[0099] The historical hyperspectral data is preferably retrieved from a local or cloud database, covering the same period data of the past 3 to 5 years. Using the same method as the stress detection confidence heatmap, multiple historical confidence heatmaps of multiple synchronous hyperspectral data in the historical hyperspectral data are constructed to form the multi-temporal confidence heatmap.

[0100] It should be understood that the core of incremental quality diagnosis in this embodiment lies in comparing the differences in sensitive indicators between the multi-temporal confidence heatmap and the stress detection confidence heatmap.

[0101] The red-edge stability index and weak absorption signal-to-noise ratio index are extracted map by map from the multi-temporal confidence heatmap to obtain the multi-temporal red-edge stability index and the multi-temporal weak absorption signal-to-noise ratio index. For each pixel location, the 25% to 75% percentile value of its historical red-edge stability index for the same period in the past 3 to 5 years is calculated as a reasonable fluctuation range. The same applies to the weak absorption signal-to-noise ratio index. The dynamic threshold range of historical sensitive indicators is constructed.

[0102] The red-edge deviation measure quantifies the degree to which the current red-edge stability index exceeds the historical threshold. The calculation method is as follows: if the current red-edge stability index is within the historical range, the deviation is 0; if it is below the lower limit, the deviation = (lower limit value - current value) / lower limit value; if it is above the upper limit, the deviation = (current value - upper limit value) / upper limit value.

[0103] For example, if the current value of the red edge stability index is 0.75 and the historical sensitivity index dynamic threshold range is [0.82, 0.89], then the deviation = (0.82 - 0.75) / 0.82 = 0.085.

[0104] The weak absorption deviation is calculated using the same logic, taking the percentile interval of the historical weak absorption signal-to-noise ratio index as a benchmark, and generating the weak absorption deviation according to the deviation ratio of the current weak absorption signal-to-noise ratio index from the interval boundary. The weak absorption deviation directly reflects the degree of temporal anomaly in the quality of moisture absorption characteristic data.

[0105] The initial deviation is preferably generated by merging the red edge deviation and the weak absorption deviation using a weighted average, and the spatial autocorrelation correction is performed using the spatial distribution characteristics of the continuum stability index to output the time series deviation.

[0106] For example, if a pixel has a high initial deviation but its neighboring pixels have small differences in continuum stability index, it is determined to be a true deviation caused by system noise; otherwise, it is weighted down and the final output temporal deviation is the comprehensive deviation value after spatial correlation calibration.

[0107] This embodiment achieves the technical effect of providing adaptive weighted fusion quantization input for continuous tracking of stress evolution.

[0108] P500: Starting from the timestamp of the single-phase hyperspectral data acquisition, the stress confidence history of the target region is reconstructed and traced back to obtain a historical time series confidence heatmap.

[0109] Using the current single-phase hyperspectral data acquisition timestamp as a benchmark, the hyperspectral data of all historical growth stages in the target area earlier than this time in the current growing season are traced back along the vegetation growth time axis. For example, if the current stage is the heading stage, the data of earlier stages such as the tillering stage and the jointing stage are traced back.

[0110] By recursively calling the entire quality evaluation chain from P100 to P400, a stress detection confidence heatmap is generated for each period of historical data within the retrospective period. Finally, the data is integrated in chronological order into a spatiotemporal continuous raster dataset covering the entire life cycle of the current growing season, namely the historical time series confidence heatmap.

[0111] The historical time-series confidence heatmap preserves the quality evolution trajectory of each pixel at different growth stages in the current growing season through geospatial coding. Its data structure includes three-dimensional attributes: time, space, and stress confidence. The time dimension is indexed by the acquisition time from the earliest growth stage to the current stage, the spatial dimension is strictly aligned with the original image coordinate system, and the stress confidence dimension stores normalized quality evaluation values, forming a spatiotemporal cube that can quantify and analyze the dynamic changes in stress detection reliability with vegetation growth.

[0112] P600: After splicing the historical time-series confidence heatmap and the stress detection confidence heatmap, perform stress stage evolution situation tracking and output the stage evolution heatmap.

[0113] In one implementation, see Figure 2 As shown, after splicing the historical time-series confidence heatmap and the stress detection confidence heatmap, stress stage evolution situation tracking is performed, and a stage evolution heatmap is output. Step P600 may further include:

[0114] P610: After spatiotemporally aligning the historical time series confidence heatmap and the stress detection confidence heatmap, resolution normalization is performed to obtain the time series normalized confidence heatmap.

[0115] P620: Based on the preset stress stage classification threshold, the pixel values ​​of the time-series normalized confidence heatmap are mapped to multi-level stress state codes.

[0116] P630: Using the multi-level stress state code as the dynamic state carrier, the transition direction of the time-series normalized confidence heatmap is compared pixel by pixel to identify the stress evolution type and output the evolution state matrix.

[0117] P640: Perform spatiotemporal aggregation intensity mapping of the same type of evolutionary pixels on the evolutionary state matrix, and output the stage evolution heatmap.

[0118] Specifically, in this embodiment, the preferred method for splicing the historical time-series confidence heatmap and the stress detection confidence heatmap is to perform resolution normalization processing after spatiotemporally aligning the historical time-series confidence heatmap and the stress detection confidence heatmap to obtain a time-normalized confidence heatmap.

[0119] Spatiotemporal alignment refers to projecting the geographic coordinate system and cell size of the historical time-series confidence heatmap and the stress detection confidence heatmap to the same size. Resolution normalization processing eliminates the spatial scale difference between historical and current data through bilinear interpolation or aggregation algorithms, ensuring that each cell is strictly comparable in both time and space dimensions, and finally forming a spatiotemporally continuous unified raster dataset, which serves as the time-series normalized confidence heatmap.

[0120] Based on preset stress stage grading thresholds, the pixel values ​​of the time-series normalized confidence heatmap are mapped to multi-level stress state codes. The preset stress stage grading thresholds are set as discretized levels according to the vegetation physiological model, such as 0-0.2 for healthy, 0.2-0.4 for latent period, 0.4-0.6 for early stress, 0.6-0.8 for mid-term stress, and 0.8-1.0 for severe stress. The normalized confidence value of each pixel is converted into an integer state code according to its threshold range, such as healthy = 1, latent period = 2, etc., realizing a symbolic mapping from continuous quality values ​​to discrete stress stages.

[0121] The comparison logic focuses on adjacent time points, such as the change in state encoding between the current time and the previous week. If the encoding value increases, it is marked as an upgrade evolution, such as healthy → latent period; if the encoding value remains unchanged, it is marked as a stable state; and if the encoding value decreases, it is marked as a reversal state. The final output evolution state matrix is ​​a grid of the same size as the time-series normalized confidence heatmap, with each cell storing the evolution type encoding, such as upgrade = 1, stable = 2, and reversal = 3.

[0122] The spatiotemporal aggregation intensity mapping of the same type of evolutionary pixels is performed on the evolutionary state matrix. Specifically, in the spatial dimension, an 8-neighborhood analysis window is established with each pixel as the center. The density ratio of pixels of the same type of evolutionary state within the window is statistically analyzed to generate a spatial aggregation density grid. At the same time, in the temporal dimension, the same type of evolutionary density grid between the current analysis period and the previous period is extracted, the relative change rate of the density value of each pixel position is calculated, and a temporal change rate grid is generated. Then, the spatial density ratio and the temporal change rate are multiplied and fused to obtain the spatiotemporal aggregation intensity value. Finally, the aggregation intensity value is pseudo-color mapped according to the preset intensity grading threshold.

[0123] The preferred pseudo-color mapping strategy is to render the high-intensity areas of the upgrade evolution type as a gradient from deep red to orange, the medium-intensity areas as yellow, the low-intensity areas as light yellow, the high-intensity areas of the reversal type as a gradient from deep blue to light blue, and the stable areas as gray. The output result is the stage evolution heatmap.

[0124] In the stage evolution heatmap, the red gradient area intuitively indicates the core area where stress spreads significantly, the blue gradient area indicates the stress relief area, and the gray area represents the stress stability area. This embodiment achieves the technical effect of spatial visualization of the evolution path and intensity of vegetation stress.

[0125] P700: Based on the stress evolution type identified in the stage evolution heatmap, generate the stage separability of progressive stress intensity.

[0126] Specifically, this embodiment generates virtual stress end-member spectral clusters based on preset progressive stress intensity spectral shifts, such as mild stress: red-edge blue shift of 2nm; moderate stress: blue shift of 5nm; severe stress: blue shift of 8nm. For the pixel regions marked as upgraded evolution in the stage evolution heatmap, their spatial aggregation intensity values ​​are extracted as weighting factors. The spectral angular distance and normalized difference index between the virtual end-members and healthy end-members under each stress intensity level are calculated in a weighted manner to generate weighted separability parameters.

[0127] Finally, the three sets of separability output according to the evolution type (upgrade / stable / reversal) constitute the stage separability. Among them, the output value of the upgrade evolution type is the weighted average separability of each intensity level, the reversal type calculates the separability by constructing the stress relief spectral gradient, and the stable state directly uses the self-verified value of the healthy endmember.

[0128] P800: The three-dimensional fusion of static stability indices, temporal deviation, and stage separability outputs a spatiotemporal fusion stress confidence heatmap.

[0129] In one implementation, the static stability index, temporal deviation, and stage separability are fused in three dimensions to output a spatiotemporal fused stress confidence heatmap. Step P800 may further include:

[0130] P810: Perform global normalization on the static stability index to generate a static stability normalized value.

[0131] P820: Calculate the timing quality score based on the timing deviation.

[0132] P830: Map the stage separability to a separability score.

[0133] P840: Configure real-time adaptive weights according to the stress-dominant type of the stage evolution heatmap, and perform weighted fusion of the static stability normalized value, temporal quality score and separability score to obtain the spatiotemporal fusion quality index.

[0134] P850: After mapping the real-time confidence level of the spatiotemporal fusion quality index according to the multi-level confidence threshold, pseudo-color rendering is performed according to the real-time confidence level to output the spatiotemporal fusion stress confidence heatmap.

[0135] In this embodiment, the static stability index is normalized globally to generate a static stability normalized value that characterizes the static anti-interference ability level of vegetation pixels in the current time phase.

[0136] The timing deviation output by P400 is converted into a positive quality evaluation parameter. Specifically, the baseline quality score is defined as 1.0. When the timing deviation is >0, the quality decay coefficient is calculated according to the exponential decay function. The final generated timing quality score range is (0,1], and the higher the value, the smaller the timing noise disturbance. The timing quality score directly quantifies the cumulative impact of historical data fluctuations on the reliability of current stress detection.

[0137] Based on the three types of separability sets output by P700, including the weighted mean of upgrade evolution, the separability value of reversal type, and the basic value of steady state, a nonlinear mapping is performed using a sigmoid function to compress the original separability values ​​to the interval (0.2, 0.98) and transform them into a monotonically increasing relationship. The resulting separability score uniformly represents the spectral discrimination ability of the data for the current stress evolution stage. A score >0.8 indicates high separability, and <0.3 indicates a high theoretical detection risk.

[0138] In this embodiment, the weight configuration rules are bound to the dominant color regions of the evolution heatmap. Specifically, the red evolution zone is given the highest weight for separability score, such as α:β:γ=0.3:0.3:0.4; the blue reversal zone increases the weight of temporal quality score, such as 0.4:0.4:0.2; and the gray stable state focuses on static stability, such as 0.5:0.3:0.2. The fusion formula is a three-dimensional linear weighting, where the quality index = w1 × static value + w2 × temporal score + w3 × separability score. The final spatiotemporal fusion quality index comprehensively reflects the overall reliability of data in stress detection at a specific spatiotemporal location.

[0139] Based on the preset dynamic quantile threshold, the spatiotemporal fusion quality index is divided into five real-time confidence levels. Specifically, a spatiotemporal fusion quality index value greater than 0.85 is mapped to an extremely high confidence level, 0.7 to 0.85 is mapped to a high confidence level, 0.5 to 0.7 is mapped to a medium confidence level, 0.3 to 0.5 is mapped to a low confidence level, and less than 0.3 is mapped to an extremely low confidence level.

[0140] Then, pseudo-color rendering is performed according to the confidence level: areas with very high confidence level are filled with dark green, areas with high confidence level are filled with light green, areas with medium confidence level are filled with yellow, areas with low confidence level are filled with orange, and areas with very low confidence level are filled with red.

[0141] Finally, the rendered pseudo-color layer is overlaid onto the original hyperspectral image with strict geographic coordinate alignment, using a 50% opacity blending mode to generate a spatiotemporal fusion stress confidence heatmap.

[0142] The spatiotemporal fusion stress confidence heatmap shows that the dark green area represents the optimal reliability of vegetation stress detection, while the red area warns that the data quality cannot be guaranteed and stress features cannot be effectively extracted. This achieves spatial visualization decision support for stress detection reliability based on three-dimensional fusion parameters.

[0143] This embodiment achieves the following technical effects:

[0144] 1. By using a dual engine of time-series baseline construction and progressive virtual stress verification, the single-phase static assessment is upgraded to continuous monitoring of the cumulative effects of stress, enabling visualized early warning of vegetation stress development stages at the data quality level.

[0145] 2. The spatiotemporal fusion stress confidence heatmap directly maps the detectability probability of different stress stages, providing a priori criteria for early intervention and avoiding the risk of misjudgment or missed detection due to blind spots in data quality.

[0146] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

[0147] The foregoing description of specific exemplary embodiments of the invention is for illustrative and explanatory purposes. These descriptions are not intended to limit the invention to the precise forms disclosed, and it will be apparent that many changes and variations can be made in accordance with the foregoing teachings. The exemplary embodiments were chosen and described in order to explain the specific principles of the invention and its practical application, thereby enabling those skilled in the art to implement and utilize various different exemplary embodiments of the invention, as well as various different choices and variations. The scope of the invention is intended to be defined by the claims and their equivalents.

Claims

1. An adaptive evaluation method for the quality of hyperspectral data for continuous monitoring of vegetation stress, characterized in that, include: After acquiring single-temporal hyperspectral data of the target area, perform stress-sensitive band quality analysis and output real-time static stability index; The stress identification of the single-phase hyperspectral data is evaluated by forging virtual stress end-members, and a real-time stress detectability index is output. After weighted fusion of the real-time static stability index and the real-time stress detectability index, spatial mapping restoration is performed to output a stress detection confidence heatmap. Incremental quality diagnosis is performed on the stress detection confidence heatmap, and the time series deviation is output. Starting from the timestamp of the single-phase hyperspectral data acquisition, the stress confidence history of the target region is reconstructed and traced back to obtain a historical time series confidence heatmap; After splicing the historical time-series confidence heatmap and the stress detection confidence heatmap, stress stage evolution situation tracking is performed, and a stage evolution heatmap is output. Based on the stress evolution type identified in the stage evolution heatmap, the stage separability of progressive stress intensity is generated; The three-dimensional fusion of static stability indices, temporal deviation, and stage separability outputs a spatiotemporal fusion stress confidence heatmap.

2. The adaptive evaluation method for hyperspectral data quality for continuous monitoring of vegetation stress as described in claim 1, characterized in that, After concatenating the historical time-series confidence heatmap and the stress detection confidence heatmap, stress stage evolution situation tracking is performed, and a stage evolution heatmap is output, including: After spatiotemporally aligning the historical time-series confidence heatmap and the stress detection confidence heatmap, resolution normalization is performed to obtain a time-normalized confidence heatmap. Based on a preset stress stage classification threshold, the pixel values ​​of the time-series normalized confidence heatmap are mapped to multi-level stress state codes. Using the multi-level stress state code as a dynamic state carrier, the transition direction of the time-series normalized confidence heatmap is compared pixel by pixel to identify the stress evolution type and output the evolution state matrix. Perform spatiotemporal aggregation intensity mapping on the evolution state matrix for the same type of evolution pixels, and output the stage evolution heatmap.

3. The adaptive evaluation method for hyperspectral data quality for continuous monitoring of vegetation stress as described in claim 1, characterized in that, Incremental quality diagnosis is performed on the stress detection confidence heatmap to output time series deviation, including: Based on the single-phase hyperspectral data acquisition timestamp, historical hyperspectral data of the same phenological period in the target area are retrieved to construct a multi-phase confidence heatmap; Incremental quality diagnosis of the stress detection confidence heatmap is performed based on the multi-temporal confidence heatmap, and the temporal deviation is output.

4. The adaptive evaluation method for hyperspectral data quality for continuous monitoring of vegetation stress as described in claim 3, characterized in that, After acquiring single-temporal hyperspectral data of the target region, stress-sensitive band quality analysis is performed, and real-time static stability indices are output, including: Pure vegetation cover pixels are extracted from the single-temporal hyperspectral data based on the vegetation index; Referring to a preset vegetation stress-sensitive band mapping table, stress-sensitive band data are extracted from the pure vegetation cover pixels, wherein the stress-sensitive band data includes red edge region data, weak absorption band data, and photosynthetic feature point data; Multi-threaded parallel stability quantization is performed on the red-edge region data, weak absorption band data, and photosynthetic feature point data to output the red-edge stability index, weak absorption signal-to-noise ratio index, and continuum stability index, which constitute the real-time static stability index.

5. The adaptive evaluation method for hyperspectral data quality for continuous monitoring of vegetation stress as described in claim 4, characterized in that, Incremental quality diagnosis of the stress detection confidence heatmap is performed based on the multi-temporal confidence heatmap, and the temporal deviation is output, including: Based on the multi-temporal confidence heatmap, a dynamic threshold range for historical sensitive indicators is constructed. Calculate the red-edge deviation of the red-edge stability index relative to the dynamic threshold range of the historical sensitivity index; Calculate the weak absorption signal-to-noise ratio index relative to the weak absorption deviation of the historical sensitivity index dynamic threshold range; After fusing the red-edge deviation and the weak absorption deviation, the spatial autocorrelation correction is performed using the continuum stability index, and the temporal deviation is output.

6. The adaptive evaluation method for hyperspectral data quality for continuous monitoring of vegetation stress as described in claim 4, characterized in that, The stress identifiability of the single-temporal hyperspectral data is evaluated by forging virtual stress endmembers, and a real-time stress detectability index is output, including: The preset NDVI baseline is used to extract continuous pixel regions of healthy vegetation from the single-phase hyperspectral data. The average reflectance spectrum of the continuous pixel area of ​​the healthy vegetation is calculated, and the baseline healthy endmember spectrum curve is output. The stress simulation of the baseline healthy endmember spectral curve is performed based on a preset perturbation scale, and the red-edge stress type endmember spectral curve and the water stress type endmember spectral curve are output. Using the baseline healthy endmember spectral curve as a reference, the spectral distinguishability of the red-edge stress type endmember spectral curve and the water stress type endmember spectral curve is quantified, and the red-edge stress type normalized difference index, the water stress type normalized difference index, the red-edge stress type spectral angular distance, and the water stress type spectral angular distance are output. Based on the stress type mapping and fusion of the red-edge stress type normalized difference index, the water stress type normalized difference index, the red-edge stress type spectral angular distance, and the water stress type spectral angular distance, the red-edge stress type detectability index and the water stress type detectability index are output. By comparing the red-edge stress detectability index and the moisture stress detectability index, the real-time stress detectability index is selected and output.

7. The adaptive evaluation method for hyperspectral data quality for continuous monitoring of vegetation stress as described in claim 4, characterized in that, After weighted fusion of the real-time static stability index and the real-time stress detectability index, spatial mapping reconstruction is performed to output a stress detection confidence heatmap, including: The red-edge stability index, weak absorption signal-to-noise ratio index, and continuum stability index in the real-time static stability index are weighted and fused using preset weights to output a static stability value. The static stable value and the real-time stress detectability index are adaptively weighted and fused to output the vegetation stress detection quality index; The vegetation stress detection quality index is mapped to the vegetation pixel spatial location of the single-temporal hyperspectral data, and spatial interpolation is performed to fill the non-vegetation pixel spatial location to obtain a vegetation stress quality distribution map. After normalizing the vegetation stress quality distribution map, pseudo-color mapping is performed to output the stress detection confidence heatmap.

8. The adaptive evaluation method for hyperspectral data quality for continuous monitoring of vegetation stress as described in claim 1, characterized in that, The three-dimensional fusion of the static stability index, temporal deviation, and stage separability outputs a spatiotemporal fusion stress confidence heatmap, including: The static stability index is subjected to global normalization to generate a static stability normalized value. Calculate the timing quality score based on the timing deviation; The stage-specific separability is mapped to a separability score; Based on the stress-dominant type of the stage evolution heatmap, real-time adaptive weights are configured, and the static stability normalized value, temporal quality score, and separability score are weighted and fused to obtain the spatiotemporal fusion quality index. After mapping the real-time confidence level of the spatiotemporal fusion quality index according to the multi-level confidence threshold, pseudo-color rendering is performed based on the real-time confidence level to output the spatiotemporal fusion stress confidence heatmap.

Images