Method for analyzing phenological characteristics of wheat by using hyperspectral image

Abstract

The invention discloses a method for analyzing wheat phenological characteristics by using a hyperspectral image, and particularly relates to the technical field of remote sensing data processing, which comprises the following steps of: decomposing a pixel into a background component and a canopy component based on end member unmixing, and executing curvature projection correction in a moisture absorption band aiming at a wet soil end member in the background so as to obtain a corrected canopy reflection sequence; extracting red edge, near-infrared platform and short-wave infrared absorption characteristics through continuum normalization, and synchronously outputting chlorophyll-related characters, structure-related characters, water-content-related characters and derivatives thereof through a character inversion operator; based on a stage template library and sequence limited transfer graph search, transfer candidates are determined according to joint transition indication, red edge phase synergy indexes and water-containing structure unwrapping residual indexes are calculated, a confirmation set is formed through a phenological consistency judgment model and a phase consistency threshold value, a stage space distribution graph layer and a time sequence table are generated, and a stage space distribution graph layer is obtained. And inflection point positioning, robustness and mobility under cross-variety and cross-management are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of remote sensing data processing, and more particularly, relates to a method for analyzing wheat phenology features by using hyperspectral images. BACKGROUND

[0002] In real production fields, multiple varieties, different planting methods and multiple physiological stresses often coexist in wheat. Hyperspectral images continuously obtain subtle changes in red edge position, near-infrared platform and short-wave infrared absorption throughout the growing season. However, these changes are not only driven by phenological processes, but also influenced by leaf structure, pigment ratio, vertical relationship between spike layer and flag leaf, and water and nutrient status. As a result, there are obvious apparent differences in the same phenological stage under different varieties or different management, and similar spectral appearances may appear in different phenological stages under stress superposition. Therefore, the time series curve is covered by the disturbance of the wheat environment, and the phenology can be clearly reflected in the time series curve. However, because the environmental disturbance is too strong, the phenological signal is shielded by noise or interference, and it is difficult to stably identify the key stages such as green return, jointing, heading and grain filling by using the traditional method relying on fixed threshold or single index, thereby affecting the grasp of the growth evaluation and the decision window of agricultural management.

[0003] The technical problem to be solved by the present application is that, for the above-mentioned field situation of coexistence of multiple varieties and multiple management and easy disturbance by stress, a method for extracting wheat phenology features by using hyperspectral images is proposed, so that the phenology information can be distinguished from the physiological state influence, thereby obtaining a phenological stage determination result which has consistency and migration among different varieties, different plots and different years. The technical problem emphasizes that, without relying on too many external priors, the time series relationship of key narrowband features such as red edge, near-infrared platform and short-wave infrared absorption is modeled, the apparent heterogeneity and appearance similarity caused by variety difference and stress are inhibited, the interference of background and geometric changes is reduced, and the positioning of the phenological inflection point is more stable, continuous and can guide agricultural management and yield evaluation. SUMMARY

[0004] In order to overcome the above-mentioned defects of the prior art, the present application provides a method for analyzing wheat phenology features by using hyperspectral images. The pixels are decomposed into background and canopy components by end member unmixing, and then the curvature projection correction is performed on the wet soil end member in the water absorption band in the background, so as to obtain a corrected canopy reflection sequence. The continuous uniformity normalization and trait inversion operator output chlorophyll-related traits, structure-related traits and water-containing-related traits. The stage template library and monotonic path search are used to generate candidates, and the red edge phase coordination index and water-containing structure disentanglement residual error index are used to confirm the key stages by using a phenology consistency discrimination model, so as to solve the problems proposed in the background.

[0005] To achieve the above object, the application provides the following technical scheme: a method for analyzing wheat phenology characteristics by using hyperspectral images, comprising: Step S1: obtaining multi-period hyperspectral images, performing band resampling and geometric alignment on a common wavelength grid after completing radiation correction and atmospheric correction, outputting a standardized sequence spectrum stack sorted according to consistent time markers, a land parcel pixel index and an observation off-axis angle; Step S2: based on the standardized sequence spectrum stack, calling an endmember library consistent with the common wavelength grid to perform group sparse variable endmember unmixing, decomposing the pixel into a background proportion sequence and a canopy reflection sequence, and implementing curvature projection correction on the wet soil endmember in the background proportion sequence (background group) in the water absorption band, while generating an endmember selection quality mark and a fitting residual; Step S3: performing continuum normalization and band consistency on the canopy reflection sequence in the red edge, near-infrared platform and short-wave infrared absorption band, calling a trait inversion operator to simultaneously output a chlorophyll-related trait sequence, a structure-related trait sequence and a water-related trait sequence and their first and second derivatives, generating a trait inversion quality mark, and performing local equidistant reconstruction in a candidate window used only for phase calculation without rewriting the consistent time markers and trait values; Step S4: taking the chlorophyll-related trait sequence, the structure-related trait sequence, the water-related trait sequence and their derivatives as observations, under the suppression of the quality mark, calculating the stage compatibility according to a stage template library composed of symbol and relative strength rules, constructing a sequential restricted transition graph that only allows stages to remain or progress adjacently, obtaining a phenology state sequence and generating a candidate time point set of stage transition by dynamic programming in combination with a joint transition indicator; Step S5: calculating the red edge phase coordination index and the water-structure disentanglement residual index on the candidate time point set of stage transition, inputting the two indexes into a phenology consistency discrimination model to obtain a phenology consistency coefficient, and comparing it with a phase consistency threshold to form a confirmed set of stage transition, wherein a historical stable segment is used as a quantile scale baseline, and the observation off-axis angle is used for geometric consistency correction; Step S6: mapping the confirmed set of stage transition to a key stage spatial distribution layer and a time sequence table based on the land parcel pixel index and the consistent time markers, implementing spatial consistency correction by using a four-connected majority consistency rule without rewriting the stage timestamp, and outputting a phenology characteristic set composed of the key stage spatial distribution layer and the time sequence table for production management and evaluation.

[0006] Preferably, the common wavelength grid is obtained by uniformly resampling spectral response functions of different platforms; the consistent time markers cover the key periods of green-up, jointing, heading and grain-filling phenology; and the observation off-axis angle is recorded by positioning and inertial navigation and archived with time markers for subsequent geometric consistency correction.

[0007] Preferably, the endmember library is set up with background group and canopy group, and is sparsely selected with group sparsity, and is applied with time consistency constraint in adjacent time markers; the wet soil endmember in the background group is executed with curvature projection correction in the short wave infrared water absorption band to weaken the influence of nonlinear curvature on the unmixing result, and the endmember selection quality mark and fitting residual are output.

[0008] Preferably, the continuum normalization is respectively implemented in the red edge transition band, the near infrared platform band and the short wave infrared water absorption band, and is uniformly connected between bands to ensure that the three types of band domain features can be jointly used as the input of the trait inversion operator under the same public wavelength grid and consistent time marker; the trait inversion output is accompanied by a trait inversion quality mark for subsequent quality mark suppression.

[0009] Preferably, the local equidistant reconstruction is only temporarily triggered in the candidate window for calculating the red edge phase coordination index, and the reconstruction result does not rewrite the consistent time marker and trait value and is not permanently stored, serving as the phase extraction input of step S5.

[0010] Preferably, the stage template library is used to define the sign and relative strength relationship of the first and second derivatives of the chlorophyll-related traits, the first and second derivatives of the structure-related traits, and the first and second derivatives of the water content-related traits; when calculating the stage compatibility, the observations determined to be suspicious by the trait inversion quality mark are implemented with quality mark suppression without changing the time marker.

[0011] Preferably, the order-constrained transition graph only allows the stage to remain or move forward by one, and the all-time sequence maximum compatibility path is obtained by dynamic programming; when there are multiple paths with the same score, the local optimal path is selected according to the neighborhood maximum position of the joint transition indicator, and a candidate time point set for stage transition is generated at the position.

[0012] Preferably, the scale baseline of the water content-structure disentanglement residual index is derived from the historical stable segment of the same plot in the previous growing season; if the historical data is missing, a replacement set is formed by the water content-related traits in the non-candidate window of the current season; the geometric consistency correction determines the correction coefficient according to the observation off-axis angle.

[0013] Preferably, the phenology consistency discrimination model is a monotonic bounded two-input one-output mapping, and when either the red edge phase coordination index or the water content-structure disentanglement residual index is close to zero, the output tends to zero, and when both are simultaneously increased, the output shows a synergistic enhancement trend, and the confirmation set for stage transition is generated by comparing the phenology consistency coefficient with the phase consistency threshold.

[0014] Preferably, the spatial consistency correction adopts a four-connected majority consistency rule, and only adjusts the spatial coherence of the stage label of the isolated image element in the rendering stage without rewriting the stage timestamp and time sequence table; the output phenology feature set is composed of the key stage spatial distribution layer and the time sequence table and takes the plot image element index and the consistent time label as the unique organization key.

[0015] Technical effects and advantages of the present application: The present application establishes a public wavelength grid and a consistent time label, combines variable end member unmixing and semi-linear curvature correction of the wet soil end member, generates stable canopy reflection sequences, extracts red edge, near-infrared platform and short-wave infrared features through continuous uniform normalization, and synchronously outputs chlorophyll-related trait sequences, structure-related trait sequences and water-related trait sequences and their derivatives through trait inversion operators, to jointly derive the phenology state sequence and the candidate time point set of stage transition with the stage template library, the joint transition indicator and the sequentially restricted monotonic path search, to realize the homologous and same-scale conversion from spectrum to trait and from trait to stage, and the globally optimal path solving, to effectively solve the problems of narrowband indication distortion caused by background confusion, instability of single exponential threshold method, and poor migration of phenology stage recognition in cross-species and cross-management scenarios.

[0016] The present application calculates the red edge phase coordination index and the water-structure disentanglement residual index on the candidate time point set of stage transition in step five, inputs the phenology consistency discrimination model to obtain the phenology consistency coefficient, forms the confirmed set of stage transition with the phase consistency threshold, realizes the generation of the key stage spatial distribution layer and the time sequence table with the plot image element index and the consistent time label, and constitutes the phenology feature set, to realize spatial coherent expression through the topological consistency correction of the four-connected majority consistency, significantly reduce the misjudgment caused by pseudo-inflexion points and geometric disturbances, strengthen the complementary coordination of "phase synchronization" and "disentanglement anomaly", and effectively solve the problems of inaccurate positioning of key stages, broken spatial expression, and difficulty in directly using the results for fine management and cross-year comparison under multi-source interference. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 A flowchart of the method for analyzing wheat phenology features using hyperspectral images according to the present application. DETAILED DESCRIPTION

[0018] Exemplary embodiments of the present disclosure will be described in greater detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided so that the present disclosure can be more thoroughly understood and the scope of the present disclosure can be accurately conveyed to those skilled in the art.

[0019] It should be understood that the dimensions of the various elements shown in the figures are shown for purposes of illustration only and are not intended to be limiting.

[0020] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the application or its application or uses.

[0021] Techniques, methods, and apparatus known to those of ordinary skill in the relevant art can not be discussed in detail herein, but should be considered within the scope of the present description.

[0022] Referring to Figure 1 The present application provides a method for analyzing wheat phenology characteristics using hyperspectral images as shown in Figure 1 The method for analyzing wheat phenology characteristics using hyperspectral images comprises the following steps: Step S1: Data acquisition and standardization After acquiring multi-period hyperspectral images, performing radiation correction and atmospheric correction, and executing band resampling and geometric alignment on the common wavelength grid, the output is a standardized sequence spectrum stack sorted by consistent time markers, a field pixel index, and an observation off-axis angle; Further, the specific implementation of step S1 is as follows: Multi-temporal hyperspectral data acquisition and standardization preprocessing are organized around the wheat phenology process, and satellite and unmanned aerial vehicle push-broom hyperspectral data are jointly acquired, and clear and stable wind speed periods are selected before and after the green-up, jointing, heading, and grain-filling stages to perform coordinated observation; Ground high-reflectance and medium-reflectance reference plates are set up, and dark field completion is recorded to complete the radiation reference, and a positioning and inertial navigation system with differential positioning capability is used to record the flight attitude and trajectory and lay out ground control markers to support geometric correction. In other words, inertial navigation system records all data collected, processed, and output during operation; Dark current and stray light correction is performed on the original radiation count, absolute radiation calibration is completed based on the reference plate reflectivity, and atmospheric correction is performed with synchronous atmospheric parameters to obtain the ground surface reflectivity; According to the spectral response function of each platform, resampling is performed to a common wavelength grid, strip stitching and sub-pixel registration of multi-temporal identical pixels are completed under unified projection and resolution, cloud, fog, and strong shadows are masked according to the water vapor absorption band and continuum shape joint discrimination, and residual brightness and inter-band flat field correction is performed according to the star-ground calibration information; The pixel-level observation off-axis angle is calculated from the flight attitude and sensor geometry, and is archived with time information one by one to construct a consistent time marker covering the key phenology stage as the only time axis throughout the whole process; The final output is a standardized sequence spectrum stack sorted by consistent time markers, a land parcel pixel index, an observation off-axis angle and a consistent time marker, which are directly used for subsequent unmixing, trait inversion and phase determination.

[0023] Step S2: End-member unmixing and background correction: Based on the standardized sequence spectrum stack, the end-member library consistent with the common wavelength grid is called to perform group sparse variable end-member unmixing, and the pixel is decomposed into a background proportion sequence and a canopy reflection sequence. The wet soil end-member in the background proportion sequence (background group) is corrected by curvature projection in the water absorption band, and the end-member selection quality mark and fitting residual are generated simultaneously. Further, the specific implementation of step S2 is: The standardized sequence spectrum stack sorted by consistent time markers, the land parcel pixel index and the consistent time markers output in step S1 are used as the only data input. In the same common wavelength grid and projection framework as step S1, the cloud, fog and strong shadow pixels are masked, skipped and kept unchanged in the consistent time marker placeholder. An end-member library composed of soil, wet soil, straw weathering layer, green leaf, aging leaf and ear layer is called and aligned with the common wavelength grid. Group sparse variable end-member unmixing is performed on the sequence spectrum stack pixel by pixel and phase by phase. Non-negative and sum-to-one constraints are set, and the end-members are group sparse selected into background groups and canopy groups. Time consistency constraints are imposed on adjacent consistent time markers to smooth the end-member selection and proportion change. The wet soil end-member in the background group is corrected by curvature projection in the short-wave infrared water absorption band to weaken the influence of soil water content on the nonlinear curvature of the red edge and short-wave infrared shape. After unmixing, two types of results are output according to the land parcel pixel index and the consistent time marker. One is the background proportion sequence, which records the total proportion of soil and straw in each phase and the end-member selection identifier. The other is the canopy reflection sequence, which subtracts the background contribution from the pixel reflectivity and normalizes it with the canopy proportion to obtain the background-removed canopy reflection. Fitting residuals and end-member selection quality marks are also generated simultaneously for subsequent quality mark suppression. The above output is fully connected with the canopy reflection sequence and consistent time marker required in step S3.

[0024] Step S3: Trait inversion and derivative generation: The canopy reflection sequence is executed in the red edge, near-infrared platform and short-wave infrared absorption band. The continuum is normalized and the consistency between bands is connected. The trait inversion operator is called to output the chlorophyll-related trait sequence, the structure-related trait sequence and the water-related trait sequence and their first and second derivatives simultaneously. Trait inversion quality marks are generated, and local equidistant reconstruction is performed in the candidate window used only for phase calculation without rewriting the consistent time marker and trait value. Prospectus: The goal of step S3 is to build trait inversion operators around the three types of narrowband features closely related to wheat phenology, namely red edge, near-infrared plateau, and shortwave infrared absorption, based on the canopy reflectance sequence output in step S2 and consistent time markers, to construct leaf chlorophyll-related trait sequences, structure-related trait sequences, and water-related trait sequences, and to generate first and second derivatives corresponding to them, providing homogeneous, homogeneous, and time-marked observations and changes for the phenological state inference in step S4 and the consistency discrimination in step S5; Further, the specific implementation of step S3 is: Sub-step 301: Call the canopy reflectance sequence, background proportion sequence, endmember selection quality mark, and fitting residual, and use the standardized sequence spectrum stack established in step S1 and the plot pixel index as a reference to verify consistent time markers and spectral band integrity on a per-pixel and per-time basis. Keep time markers in place for moments masked by clouds and strong shadows, but do not participate in numerical calculations. Mark abnormal spectral shapes indicated by endmember selection quality marks and fitting residuals to form an effective observation set for feature extraction and register with the pixel index one by one. Sub-step 302: For each canopy reflectance curve in the effective observation set, establish a continuum in the red edge transition band, near-infrared plateau band, and water absorption band and implement in-band normalization to extract red edge displacement curves and slope-related curves, platform intensity curves and local concave-convex degree curves, absorption depth curves and absorption area curves, and retain continuum parameters and band-to-band connection information for subsequent trait inversion input consistency. The entire process maintains consistent sampling positions and sequences with the public wavelength grid; Sub-step 303: Take the band domain features output in sub-step 302 as input, call the trait inversion operator in the offline stage, map the red edge-related input to the leaf chlorophyll-related trait sequence, the near-infrared platform-related input to the structure-related trait sequence, and the water absorption-related input to the water-related trait sequence; Use continuum parameters and band-to-band connection information to complete input consistency and scale unification during the mapping process, so that the three types of traits are all dimensionless relative quantities; Sub-step 304: Calculate the time first and second derivatives of the leaf chlorophyll-related trait sequence, the structure-related trait sequence, and the water-related trait sequence under consistent time markers; Sub-step 305: Select quality marks, fitting residuals, continuum fitting deviations, and band-to-band connection consistency to generate trait inversion quality marks for each time's three types of traits and their derivatives; In subsequent calculations, implement quality mark suppression for suspicious observations marked; Sub-step 306: Organize the chlorophyll-related trait sequence, the structure-related trait sequence, the water content-related trait sequence, and their first and second derivatives, together with the consistent time marker and the trait inversion quality mark, into output trait sequences indexed by plot pixels for direct calling by step S4; the fixed object name, the belonging step, and the meaning in the interface declaration, the three types of traits are dimensionless relative quantities, the derivatives are dimensionless change rates, and the trait inversion quality mark is a Boolean or rating type mark that does not participate in numerical calculation.

[0025] Summary and inventive point of step S3: This step couples the continuous uniformity, the trait inversion, and the derivative generation under the consistent time marker into a unified data object system without increasing the adjustable factors, and adds a local equidistant reconstruction declaration only for phase extraction in step S5, while the quality mark is used to ensure abnormal isolation and time coherence, realizing stable mapping from canopy reflection sequences to trait sequences and their change quantities, providing homologous, homoscale, and homochronous inputs for the regularized stage inference of step S4 and the dual-evidence consistency confirmation of step S5, and ensuring complete consistency with the independent claims and dependent claims in terms and technical features.

[0026] Step S4: Sequential phenological state inference With the chlorophyll-related trait sequence, the structure-related trait sequence, the water content-related trait sequence, and their derivatives as observations, and under the quality mark suppression, the phase compatibility is calculated according to the phase template library composed of symbols and relative strength rules, the sequential restricted transition graph that only allows the phase to remain or advance adjacent is constructed, and the phenological state sequence is obtained by dynamic programming combined with the joint transition indicator to generate the candidate time point set of phase transition; Prospectus: Step S4 takes the chlorophyll-related trait sequence, the structure-related trait sequence, the water content-related trait sequence, and their trait derivative sequences output by step S3 as the only observation source, combines the consistent time marker and the trait inversion quality mark, and constructs a sequential phenological state model without introducing new adjustable factors. First, the observations are mapped to the phase compatibility field using the phase template library, then the monotonic path search is performed on the sequential restricted transition graph to obtain the phenological state sequence, and the candidate time point set of phase transition is extracted according to the joint transition indicator.

[0027] Further, the specific implementation of step S4 is: Sub-step 401: Read the chlorophyll-related trait sequence, the structure-related trait sequence, the water content-related trait sequence, and their first and second time derivatives from step S3, and align them by consistent time markers; use the trait inversion quality flag to suppress the numerical contribution of suspicious observations, specifically mark them as not participating in matching and accumulation when calculating, but do not change the consistent time markers and plot pixel indexes; accordingly, form an observation vector composed of the three types of traits and their two-level derivatives at each time marker, which is the only input of this step; Sub-step 402: Define the stage template library in step S4, which is used to describe the expected change pattern of each phenological stage on the observation vector, including the expected sign and relative strength of the first and second derivatives of the chlorophyll-related traits, the expected inflection point relationship of the first and second derivatives of the structure-related traits, and the expected cooperative direction of the first and second derivatives of the water content-related traits; for each time marker, match the observation vector with the stage template library piece by piece, form a stage compatibility degree according to the proportion of the number of satisfied items to the total number of items that can participate, and implement quality flag suppression on the observation items marked by the trait inversion quality flag without counting in the denominator and numerator; the stage compatibility degree is a dimensionless relative quantity, ensuring comparability between different plots and years; Sub-step 403: In a small window of consistent time markers, calculate the sign consistency degree of the first derivative of the chlorophyll-related trait and the first derivative of the structure-related trait, and express the curvature transition indication by the zero-crossing density of the second derivative of the water content-related trait; after implementing quality flag suppression on suspicious observations, generate a joint transition indication quantity by the conjunctive relationship of the rule table according to the sign consistency degree meeting the standard and the curvature transition indication meeting the standard; the joint transition indication quantity is a dimensionless relative quantity, only serving as a geometric constraint for subsequent transition positioning, and not participating in the numerical accumulation of the stage compatibility degree; Sub-step 404: Define a sequential constraint transition graph that only allows the phenological stage to advance or remain unchanged between adjacent stages, and prohibits reverse order and jumping; use the stage compatibility degree as the score of each stage at each time marker, and use dynamic programming to find the maximum score path in the allowed edge set; when there are multiple paths with the same score, preferentially select the time marker with the maximum neighborhood of the joint transition indication quantity in the adjacent stage switching position; output the phenological state sequence from the search result, and keep it consistent with the plot pixel index; Sub-step 405: Detect the time markers where the stage marker changes in the phenological state sequence, find the neighborhood maximum value of the joint transition indication quantity in a small window around each change and register it as a candidate time point for stage transition, while preserving the corresponding stage pair information and the proportion of entries suppressed by the quality flag as metadata for subsequent traceable analysis; if the interval between adjacent changes is too close and the stage pairs are consistent, merge them into a single candidate time point represented by the one with higher joint transition indication quantity; Sub-step 406: Output the phenological state sequence and the candidate time point set of phase transition with the plot pixel index as the organization key; declare the fixed object name, source, and acceptable range, wherein the phenological state sequence is the time sequence of phase markers, and only the predefined phase set from the phase template library is taken; the candidate time point set of phase transition is a subset of consistent time markers, and is accompanied by phase pair information, joint transition indicator, and entry proportion inhibited by quality markers; the above objects are dimensionless and one-to-one corresponding to consistent time markers.

[0028] Summary and creative point of step S4: Step S4 cooperates the monotonous path search on the geometric constraint order limited transition graph of the joint transition indicator with the rule-based characterization of the phase template library and the single path search, transcribes the multi-channel observation quantity and its derivative of step S3 into the phase compatibility field which can be optimized on the same time axis, and uses the joint transition indicator as a position constraint in the path search, thereby ensuring the global optimality of phase recognition and the local accuracy of transition positioning at the same time; the combination avoids the fragility of threshold stacking and empirical threshold, realizes the decision-making process driven by the three elements of observation quantity, geometric consistency, and phase order, and outputs the phenological state sequence and the candidate time point set of phase transition which naturally connects with the phenological consistency discrimination of step S5, forming the overall technical effect beyond the conventional single-index threshold method.

[0029] Step S5: Consistency discrimination and phase confirmation: Calculate the red edge phase synergy index and the water-structure disentanglement residual index on the candidate time point set of phase transition, input the two indexes into the phenological consistency discrimination model to obtain the phenological consistency coefficient, and compare it with the phase consistency threshold to form the confirmation set of phase transition, wherein the historical stable section is used for quantile scale baseline, and the observation off-axis angle is used for geometric consistency correction; Prospect: Step S5 takes the phenological state sequence and the candidate time point set of phase transition output by step S4 as the only trigger source, and only calls the chlorophyll-related trait sequence, the structure-related trait sequence, the water-related trait sequence, and their first and second derivatives output by step S3 as calculation inputs, calculates the red edge phase synergy index and the water-structure disentanglement residual index in the local time window of each candidate time point without adding adjustable factors, inputs the two indexes into the phenological consistency discrimination model to generate the phenological consistency coefficient, and compares it with the phase consistency threshold to form the confirmation set of phase transition.

[0030] Further, the specific implementation of step S5 is: Sub-step 501 : For each candidate time point in the candidate time point set given in step S4 for the phase transition, construct a closed interval time window on the consistent time scale, and lock the local slices of the following input objects as the only data source, i.e., the chlorophyll-related trait sequence, the structure-related trait sequence, the water-related trait sequence, the first-order derivative of the chlorophyll-related trait, the first-order derivative of the structure-related trait, the second-order derivative of the chlorophyll-related trait, and the second-order derivative of the structure-related trait; Sub-step 502: Red edge phase synergy index calculation: construct the analytic signal with Hilbert transform and extract the derivative phase within the time window, and define the red edge phase synergy index as the product of the phase concentration and the curvature homodirection, specifically: wherein the analytic time window represents a local equidistant window around the time point τ, is the sampling number thereof; the first-order derivative of the chlorophyll-related trait , the first-order derivative of the structure-related trait corresponds to the instantaneous phase , is the imaginary unit; the second-order derivative of the chlorophyll-related trait , and the second-order derivative of the structure-related trait ; the phase concentration is used to measure the consistency of the first-order derivative of the chlorophyll-related trait and the first-order derivative of the structure-related trait , is the sign function; is used to calculate the vector norm; the curvature homodirection is used to measure the proportion of the second-order derivative of the chlorophyll-related trait and the second-order derivative of the structure-related trait that are of the same sign; is used to represent the comprehensive synchronous transition strength; Sub-step 503: Water-structure disentanglement residual index calculation: construct a local linear mapping with the robust median slope method within the time window, and implement quantile scaling and geometric consistency correction on the water-related residual, and define the water-structure disentanglement residual index as: where the local slope and the intercept define the water-related property a local robust linear mapping of the structure-related property , both real numbers; the water residual is the water-related property after removing the structure-explained part, real value; the quantile distance >0 is the interquartile distance of the water-related property W(u) of the same plot in the historical stable period, u is the set of time markers of the historical stable period; e>0 is a fixed tiny constant to avoid zero division, not as an adjustable factor; the scaled residual is obtained by suppressing the influence of extreme values; the geometric consistency coefficient is defined by the observation deviation angle , where the observation off-axis angle is recorded in step S1, the value interval is [0, π / 2]; the water-structure disentanglement residual index is a dimensionless quantity, used to represent the water anomaly strength that is difficult to be explained by structure under the condition of geometric consistency; Substep 504: Phenomenon consistency coefficient calculation: input the red edge phase synchronization index and the water-structure disentanglement residual index into a predefined monotonic bounded discriminant mapping to generate the phenomenon consistency coefficient , specifically: where: the phenomenon consistency coefficient is a dimensionless quantity, which adopts square root geometric coupling and compression to realize the synergistic enhancement and in-bound mapping of the two indexes; this mapping significantly suppresses when either of the two indexes approaches zero, and produces super-linear promotion when both indexes increase simultaneously, embodying the complementary synergy of phase synchronization and disentanglement anomaly double evidence; Substep 505: Compare the phenomenon consistency coefficient K with the phase consistency threshold value, and only when the phenomenon consistency coefficient K is not less than the phase consistency threshold value, the time point and its corresponding stage pair information are merged into the confirmation set of stage transition, accompanied by recording (red edge phase synchronization index , water-structure disentanglement residual index , phenomenon consistency coefficient ; Sub-step 506: output the phase transition confirmation set with the plot pixel index as the organization key, and fix the following objects and mapping relationship in the interface declaration, i.e. the phase transition confirmation set, the phenological consistency coefficient sequence, the red edge phase coordination index sequence, and the water-structure unwrapping residual index sequence, all of which are one-to-one corresponding to the consistent time marker and are dimensionless quantities.

[0031] Summary of step S5: This step constructs the red edge phase coordination index by analyzing the phase concentration x curvature co-directionality, avoids dependence on amplitude, and directly corresponds to the synchronization characteristics of phenological transition; at the same time, the water-structure unwrapping residual index is constructed by robust slope median mapping + quantile scaling + geometric consistency correction, which highlights the water anomaly related to phenology after stripping the structure interpretable component; the two indices are coupled by square root and bounded by arctangent to form the phenological consistency coefficient, so that the double evidence is superlinearly enhanced when they are lifted together, and is automatically inhibited when any evidence is weak, which reduces the probability of false inflection points from the mechanism, embodies the complementary coordination of phenological transition synchronization and unwrapping, and does not rely on the conventional path of variance, standard deviation or weighted average, which is novel and implementable.

[0032] Step S6: spatial mapping and result generation: Map the phase transition confirmation set into the key stage spatial distribution layer and time sequence table based on the plot pixel index and the consistent time marker, implement spatial consistency correction without rewriting the stage timestamp using the four-connected majority consistency rule, and output the phenological feature set composed of the key stage spatial distribution layer and the time sequence table for production management and evaluation; Further, the specific implementation of step S6 is: Input the phase transition confirmation set output by step S5 as the core data, and simultaneously input the phenological consistency coefficient sequence, the red edge phase coordination index sequence, and the water-structure unwrapping residual index sequence output by the same step, only using the plot pixel index and the consistent time marker established in step S1 as index information; Through spatiotemporal mapping and index matching, each time point in the phase transition confirmation set is bound to the corresponding pixel one by one and attached with stage pair information and the values of the above three types of indices, to construct a stage timestamp table with pixel as the primary key; According to the consistent time marker, perform time topology sorting to generate a stage order chain at the pixel level and calculate the duration and interval of adjacent stages and other derived items, and write the same coefficient of the phenological consistency coefficient sequence into the corresponding stage node as the confidence field; On the spatial side, use geographic coding and raster mapping engine to map the timestamp of each key stage to generate the corresponding key stage spatial distribution layer, and only in the rendering stage, apply the four-connected majority consistency rule to implement spatial consistency correction without rewriting the stage timestamp and time sequence table of the isolated pixel. The output includes key stage spatial distribution layers (each layer corresponds to a stage, and the pixel attribute contains the stage time point, the phenology consistency coefficient, the red edge phase coordination index, and the water-structure disentanglement residual error index) and a time sequence table (taking the pixel as the primary key, the stage as the field, recording the stage time point, the duration, the interval, and the corresponding phenology consistency coefficient), and the two are combined into a phenology feature set as the standard external result.

[0033] Finally: the above only for the preferred embodiments of the present application, and not for limiting the present application, any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for analyzing wheat phenological characteristics using hyperspectral imagery, characterized in that, include: Step S1: Acquire multiple hyperspectral images, perform radiometric and atmospheric corrections, then perform spectral band resampling and geometric alignment on a common wavelength grid, and output a standardized sequence spectral stack, plot pixel index, and observation off-axis angle sorted by consistent time stamp; Step S2: Based on the standardized sequence spectrum stack, call the endmember library consistent with the common wavelength raster to perform sparse variable endmember demixing, decompose the pixel into background ratio sequence and canopy reflection sequence, and perform curvature projection correction on the wet soil endmember in the background ratio sequence within the moisture absorption zone, while generating endmember selection quality label and fitting residual. Step S3: Perform continuous unification and inter-band connection consistency on the canopy reflectance sequence in the red edge, near-infrared plateau and short-wave infrared absorption band, call the trait inversion operator to synchronously output the chlorophyll-related trait sequence, the structure-related trait sequence and the water content-related trait sequence and their first and second derivatives, and generate trait inversion quality indicators. Step S4: Using chlorophyll-related trait sequences, structure-related trait sequences, water content-related trait sequences and their derivatives as observations, under the suppression of quality markers, and combined with joint transition indicators, dynamic programming is used to obtain phenological state sequences and generate a candidate time point set for stage transitions. Step S5: Calculate the red-edge phase coherence index and the water-bearing-structure unwrapping residual index on the candidate time point set of the stage transition, input them into the phenological consistency discrimination model to obtain the phenological consistency coefficient, and compare them with the phase consistency threshold to form the confirmation set of the stage transition. The historical stable segment is used as the quantile scale baseline, and the observed off-axis angle is used for geometric consistency correction. Step S6: Based on the plot cell index and consistent time stamp, map the confirmed set of stage transitions to a key stage spatial distribution layer and time sequence table. Use the four-connected majority consistency rule to implement spatial consistency correction without rewriting the stage timestamp. Output a phenological feature set composed of the key stage spatial distribution layer and time sequence table for production management and evaluation.

2. The method for analyzing wheat phenological characteristics using hyperspectral imagery according to claim 1, characterized in that, The common wavelength grid is obtained by uniform resampling of the spectral response functions of different platforms; the consistent time stamp covers the key phenological periods of greening, jointing, heading and grain filling; the observed off-axis angle is recorded by positioning and inertial navigation and archived with time stamp for subsequent geometric consistency correction.

3. The method for analyzing wheat phenological characteristics using hyperspectral imagery according to claim 1, characterized in that, The endmember library is set to sparse selection according to the background group and the canopy group, and time consistency constraints are applied to adjacent time markers; the wet soil endmember in the background group is subjected to curvature projection correction in the short-wave infrared moisture absorption band to reduce the influence of nonlinear curvature on the unmixing results, and the endmember selection quality label and fitting residual are output.

4. The method for analyzing wheat phenological characteristics using hyperspectral imagery according to claim 1, characterized in that, The continuous unification is implemented separately in the red-edge transition band, the near-infrared plateau band, and the short-wave infrared moisture absorption band. The inter-band connection consistency ensures that the three types of band features can be jointly used as inputs to the trait inversion operator under the same common wavelength grid and consistent time mark. The trait inversion output is accompanied by a trait inversion quality flag for subsequent quality flag suppression.

5. The method for analyzing wheat phenological characteristics using hyperspectral imagery according to claim 1, characterized in that, The local equidistant reconstruction is only temporarily triggered within the candidate window used to calculate the red-edge phase co-existence index. The reconstruction result does not rewrite the consistency time stamp and trait value and is not persistently stored, serving as the phase extraction input for step S5.

6. The method for analyzing wheat phenological characteristics using hyperspectral imagery according to any one of claims 1-5, characterized in that, The stage template library is used to define the signs and relative strengths of the first and second derivatives of chlorophyll-related traits, structure-related traits, and water-related traits.

7. The method for analyzing wheat phenological characteristics using hyperspectral imagery according to claim 1, characterized in that, The sequence-restricted transition graph only allows stage retention or adjacent advancement, and dynamic programming is used to obtain the path with the maximum compatibility across the entire time series. When there are multiple paths with the same score, local optimization is performed at the neighborhood maximum position of the joint transition indicator, and a set of candidate time points for stage transition is generated at that position.

8. The method for analyzing wheat phenological characteristics using hyperspectral imagery according to claim 1, characterized in that, The scale baseline of the water content-structure unwrapping residual index is derived from the historical stable period of the previous growing season of the same plot; if historical data is missing, a replacement set is formed by water content-related traits within the non-candidate window of the current season; the geometric consistency correction is based on the deterministic correction coefficient calculated according to the observed off-axis angle.

9. The method for analyzing wheat phenological characteristics using hyperspectral imagery according to claim 1, characterized in that, The phenological consistency discrimination model is a monotonically bounded two-input one-output mapping. When either the red-edge phase coordination index or the water-containing-structure untangling residual index approaches zero, the output approaches zero. When both increase simultaneously, the output shows a coordinated enhancement trend. The phenological consistency coefficient is compared with the phase consistency threshold to generate a confirmation set of stage transitions.

10. The method for analyzing wheat phenological characteristics using hyperspectral imagery according to claim 1, characterized in that, The spatial consistency correction adopts the four-connected majority consistency rule, which adjusts the spatial coherence of the stage markers of isolated pixels only during the rendering stage without rewriting the stage timestamps and time sequence tables. The output phenological feature set consists of a key stage spatial distribution layer and a time sequence table, with the plot cell index and consistent time marker as the unique organization keys.

Images