Spectroscopic data separation method and system

Abstract

The present application provides a kind of spectral data separation method and system, the method comprises: according to initial vegetation coverage estimation and initial vegetation spectral data, the initial non-vegetation spectral data corresponding to the mixed spectral original data to be processed is calculated;Based on initial vegetation spectral data, initial non-vegetation spectral data and initial vegetation abundance value, initial mixed spectral fitting data is fitted;Based on double-target optimization function, vegetation spectral feature constraint condition and initial mixed spectral fitting data, iterative optimization solving process is repeatedly executed, and spectral data separation result is obtained;Double-target optimization function is constructed to minimize the difference between mixed spectral original data and mixed spectral fitting data;Vegetation spectral feature constraint condition is constructed based on the absorption characteristics and reflection characteristics of vegetation spectrum in the preset wave band.The present application simplifies spectral data separation operation, improves the degree of automation and adaptability of spectral data separation.

Description

Technical Field

[0001] This invention relates to the fields of remote sensing technology and signal processing technology, and in particular to a method and system for spectral data separation. Background Technology

[0002] In fields such as agricultural remote sensing and ecological environment monitoring, the spectra of ground objects acquired by spectrometers or remote sensing satellites are often mixed spectra. For example, a single pixel may simultaneously contain crop canopy, bare soil, and waterlogging, or forest canopy may contain understory soil and litter. Such spectral signals composed of multiple ground objects are called mixed spectra.

[0003] In the field of remote sensing spectral analysis, extracting spectral information of target ground features (such as vegetation) from mixed pixels has long been a fundamental challenge. Existing solutions can be mainly categorized into methods based on linear spectral mixture models and their variants, machine learning / deep learning-based methods, and vegetation index thresholding methods. However, each of these existing spectral separation methods has significant limitations. For example, methods based on linear spectral mixture models heavily rely on prior knowledge and are complex to operate; in practical applications, obtaining pure endmember spectra is both difficult and time-consuming. Deep learning-based methods lack a physical foundation and require large amounts of data. Methods based on vegetation index thresholding suffer from low accuracy and strong subjectivity in the threshold.

[0004] Therefore, there is an urgent need for a method and system for spectral data separation to solve the above problems. Summary of the Invention

[0005] To address the problems existing in the prior art, the present invention provides a method and system for spectral data separation.

[0006] This invention provides a method for separating spectral data, comprising:

[0007] Based on the initial vegetation cover estimate and the initial vegetation spectral data, calculate the initial non-vegetation spectral data corresponding to the raw mixed spectral data to be processed;

[0008] Based on the initial vegetation spectral data, the initial non-vegetation spectral data, and the initial vegetation abundance value, initial mixed spectral fitting data are obtained by fitting.

[0009] Based on the dual-objective optimization function, the vegetation spectral characteristic constraints, and the initial mixed spectral fitting data, the iterative optimization solution process is repeatedly executed. When the iterative optimization result meets the preset conditions, the spectral data separation result corresponding to the original mixed spectral data is obtained.

[0010] The dual-objective optimization function is constructed with the objective of minimizing the difference between the original mixed spectral data and the fitted mixed spectral data; the vegetation spectral feature constraint is constructed based on the absorption and reflection characteristics of the vegetation spectrum in a preset band.

[0011] According to a spectral data separation method provided by the present invention, the iterative optimization solution process includes:

[0012] Based on the vegetation spectral data, non-vegetation spectral data, and vegetation abundance values ​​of the current round, the mixed spectral fitting data of the current round is obtained.

[0013] Calculate the squared error between the original mixed spectrum data and the fitted mixed spectrum data of the current round;

[0014] If the squared error is less than or equal to the preset minimum spectral fitting squared error, and the vegetation spectral data in the mixed spectral fitting data of the current round satisfies the vegetation spectral feature constraint, the spectral data separation result is generated based on the vegetation spectral data of the current round, the non-vegetation spectral data of the current round, and the vegetation abundance value of the current round.

[0015] If the squared error is determined to be less than or equal to the preset minimum spectral fitting squared error, and the vegetation spectral data in the current round of mixed spectral fitting data does not meet the vegetation spectral feature constraint, the vegetation spectral data of the current round is adjusted based on the vegetation spectral feature constraint, and the adjusted vegetation spectral data is used as the vegetation spectral data for the next round of the iterative optimization solution process.

[0016] If the squared error is determined to be greater than the preset minimum spectral fitting squared error, and the vegetation spectral data in the current round of mixed spectral fitting data does not meet the vegetation spectral feature constraint conditions, the vegetation abundance value of the current round is adjusted, and the adjusted vegetation abundance value is used as the vegetation abundance value of the next round of the iterative optimization solution process.

[0017] According to the spectral data separation method provided by the present invention, the vegetation spectral characteristic constraints include vegetation spectral absorption depth constraints, vegetation spectral curve steepness constraints, and spectral reflectance threshold constraints, wherein:

[0018] If the absorption rate of the vegetation spectral data in the current round is greater than or equal to the lower limit of the preset absorption rate and less than or equal to the upper limit of the preset absorption rate in the preset band, it is determined that the vegetation spectral data in the current round satisfies the vegetation spectral absorption depth constraint condition.

[0019] If the first derivative of the vegetation spectral data of the current round is greater than or equal to the lower limit of the preset derivative and less than or equal to the upper limit of the preset derivative, then the vegetation spectral data of the current round is determined to satisfy the vegetation spectral curve steepness constraint condition.

[0020] If the vegetation reflectance of the current round's vegetation spectral data in the second preset spectral band range is greater than or equal to the preset lower limit of reflectance and less than or equal to the preset upper limit of reflectance, then the vegetation spectral data of the current round is determined to satisfy the spectral reflectance threshold constraint condition.

[0021] According to a spectral data separation method provided by the present invention, the method further includes:

[0022] When the absorptivity of the vegetation spectral data of the current round in the preset band is less than the preset absorptivity lower limit, the reflectivity of the vegetation spectral data of the current round in the preset band is adjusted upward by a preset adjustment range to obtain the adjusted vegetation spectral data.

[0023] When the absorption rate of the vegetation spectral data of the current round in the preset band is greater than the upper limit of the preset absorption rate, the reflectance of the vegetation spectral data of the current round in the preset band is adjusted downward by a preset adjustment range to obtain the adjusted vegetation spectral data.

[0024] According to a spectral data separation method provided by the present invention, the method further includes:

[0025] When the first derivative of the vegetation spectral data of the current round is less than the lower limit of the preset derivative in the first preset spectral band range, the spectral value of the vegetation spectral data of the current round corresponding to the upper limit of the spectral band in the first preset spectral band range is adjusted upward by a preset adjustment range to obtain the adjusted vegetation spectral data.

[0026] When the first derivative of the vegetation spectral data of the current round is greater than the upper limit of the preset derivative in the first preset spectral band range, the spectral value of the vegetation spectral data of the current round corresponding to the upper limit of the spectral band in the first preset spectral band range is adjusted downward according to the preset adjustment range to obtain the adjusted vegetation spectral data.

[0027] According to a spectral data separation method provided by the present invention, the method further includes:

[0028] When the vegetation reflectance of the current round's vegetation spectral data in the second preset spectral band is less than the preset lower limit value, the reflectance of the current round's vegetation spectral data in each spectral band within the range of the second preset spectral band is increased to obtain the adjusted vegetation spectral data.

[0029] Alternatively, if the vegetation reflectance of the current round's vegetation spectral data in the second preset spectral band is greater than the preset upper limit of reflectance, the reflectance of the current round's vegetation spectral data in each spectral band within the second preset spectral band range is reduced to obtain the adjusted vegetation spectral data.

[0030] The present invention also provides a spectral data separation system, comprising:

[0031] The first processing module is used to calculate the initial non-vegetation spectral data corresponding to the mixed spectral raw data to be processed based on the initial vegetation coverage estimate and the initial vegetation spectral data.

[0032] The second processing module is used to fit initial mixed spectral fitting data based on the initial vegetation spectral data, the initial non-vegetation spectral data, and the initial vegetation abundance value.

[0033] The spectral data separation module is used to repeatedly execute the iterative optimization solution process based on the bi-objective optimization function, vegetation spectral characteristic constraints and the initial mixed spectral fitting data, and obtain the spectral data separation result corresponding to the original mixed spectral data when the iterative optimization result meets the preset conditions.

[0034] The dual-objective optimization function is constructed with the objective of minimizing the difference between the original mixed spectral data and the fitted mixed spectral data; the vegetation spectral feature constraint is constructed based on the absorption and reflection characteristics of the vegetation spectrum in a preset band.

[0035] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the spectral data separation method as described above.

[0036] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the spectral data separation method as described above.

[0037] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the spectral data separation method as described above.

[0038] The spectral data separation method and system provided by this invention calculates initial non-vegetation spectral data using initial vegetation coverage estimates and vegetation spectral data; then, by combining the initial vegetation spectral data, non-vegetation spectral data, and vegetation abundance values, initial mixed spectral fitting data is obtained; finally, based on a bi-objective optimization function constructed with the goal of minimizing the difference between the original and fitted mixed spectral data and vegetation spectral feature constraints, the method iteratively optimizes and solves the problem to obtain the spectral data separation result. This eliminates the need to rely on a large endmember spectral library, simplifies the spectral data separation operation, and improves the automation and adaptability of spectral data separation. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in this 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 some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0040] Figure 1 A schematic flowchart of the spectral data separation method provided by the present invention;

[0041] Figure 2 This is a schematic diagram showing the effect comparison before and after separation of mixed spectral data provided by the present invention;

[0042] Figure 3 A schematic diagram of the spectral data separation system provided by the present invention;

[0043] Figure 4 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0045] In the field of remote sensing spectral analysis, existing solutions for extracting spectral information of target features from mixed pixels can be mainly divided into the following categories, but each of these methods has significant limitations:

[0046] I. Methods based on Linear Spectral Mixture Model (LSMM) and its variants assume that the spectral signal within a pixel is a linear combination of the spectra of all pure land features (called "endmembers") within it. However, this method heavily relies on prior knowledge and is complex to operate, specifically manifested in the following ways: (1) Strong endmember dependence: The decomposition accuracy is highly dependent on the accuracy of the pre-selected endmember spectra. In practical applications, obtaining pure endmember spectra is both difficult and time-consuming. (2) "Different spectra for the same object": The spectra of the same land feature (such as soil) vary greatly under different locations and humidity conditions, making it difficult to represent with a single endmember. (3) Nonlinear mixing problem: In densely vegetated areas, light will be scattered multiple times, and simple linear models are no longer applicable. (4) Complex operation: Manual intervention is required to select endmembers, making it difficult to automate the process.

[0047] II. Machine Learning / Deep Learning-Based Methods. Currently, deep learning methods are being attempted for endmember extraction and abundance inversion, such as using models like autoencoders, convolutional neural networks (CNNs), or generative adversarial networks (GANs). These models learn the inherent laws of spectral mixing through a large amount of training data, and endmember extraction and abundance inversion are implicitly embedded in the network weights. By inputting the mixed spectrum, the model can directly output the estimated abundance map or even reconstruct the component spectra. However, this method lacks a physical basis and has a large data requirement, specifically: (1) Extremely strong data dependence: The model performance is heavily dependent on large-scale, high-quality training datasets. These datasets require accurate annotation (i.e., real endmembers and abundances), and obtaining the real values ​​is extremely costly. (2) Lack of interpretability and physical meaning: Deep learning models are often regarded as "black boxes," and their decomposition process lacks clear physical and optical principles to support it. The results are difficult to interpret and verify physically, which presents an application obstacle in fields like remote sensing that require high reliability. (3) Poor generalization ability: A model trained on one region or sensor is difficult to be directly applied to data obtained from another environment with large differences or different sensors, and its universality is insufficient.

[0048] Third, the vegetation index threshold method is a simple and intuitive non-decomposition method. It calculates the vegetation index (e.g., NDVI, EVI) for each pixel and sets an empirical threshold. Pixels with a vegetation index above the threshold are considered "pure vegetation" and their spectra can be used directly; pixels with an index below the threshold are considered mixed pixels or non-vegetation, and their spectra are discarded or require further processing. However, this method suffers from low separation accuracy and high subjectivity. For example, it cannot extract vegetation component spectra from mixed pixels, only performing simple binary classification, wasting a large number of pixels with low to medium vegetation cover containing useful information. Furthermore, the selection of the threshold is highly dependent on experience and specific scenarios, lacking objective standards and exhibiting poor adaptability.

[0049] The aforementioned existing methods fail to effectively utilize the most stable and universal spectral morphological characteristics of vegetation as an inherent constraint to guide the decomposition process. To address the problems in these existing technologies, this invention provides a spectral data separation method that leverages the strong absorption characteristics of vegetation in the red-edge band (approximately 680nm to 750nm) and the high reflectance plateau characteristics in the near-infrared band (approximately 780nm to 900nm)—two of the most stable and universal spectral characteristics—as physical constraints to guide the vegetation spectral separation process. Simultaneously, by comparing the smooth and monotonously changing spectral morphology of non-vegetation components (such as soil and water) in this range, an iterative optimization algorithm is used to find an optimal vegetation spectrum that, when mixed with an optimal non-vegetation background spectrum in a certain proportion, perfectly fits the input mixed spectrum while simultaneously satisfying the aforementioned vegetation characteristic constraints.

[0050] Figure 1 This is a flowchart illustrating the spectral data separation method provided by the present invention, as shown below. Figure 1 As shown, the present invention provides a method for spectral data separation, comprising:

[0051] Step 101: Based on the initial vegetation cover estimate and the initial vegetation spectral data, calculate the initial non-vegetation spectral data corresponding to the raw mixed spectral data to be processed.

[0052] In this invention, it is assumed that the mixed spectrum is a linear mixture of vegetation and non-vegetation components, meaning that the mixed spectrum (the original mixed spectrum data to be processed) is the result of combining the vegetation spectrum and the non-vegetation spectrum in a certain proportion (vegetation coverage).

[0053] Specifically, in the initial stage, for the mixed spectral data S_mixed(λ) to be processed, the initial vegetation cover estimate (e.g., set to 0.5, denoted as Fv_initial) and the initial vegetation spectral data (e.g., a standard green plant spectral curve, denoted as V_initial) can be known based on prior knowledge.

[0054] Then, based on the linear mixing relationship, the non-vegetation background spectrum, i.e., the initial non-vegetation spectral data B(λ), is estimated using the following formula for calculating the non-vegetation background spectrum:

[0055] B(λ) = (S_mixed(λ)-Fv_initial×V_initial(λ)) / (1-Fv_initial);

[0056] When estimating non-vegetation spectral data, non-physical spectra (such as negative values) may be generated, which require subsequent constraint processing. For example, B(λ) must be spectrally smoothed (this can be achieved by constraining its second derivative), i.e., |B''(λ)|≦4% / (nm).2 All spectral values ​​B(λ) are non-negative. It should be noted that for each subsequent iteration of the optimization process, the non-vegetation spectral data of the current round can also be calculated based on the above calculation formula for the non-vegetation background spectrum, the vegetation spectral data of the current round, and the vegetation abundance value of the current round.

[0057] Step 102: Based on the initial vegetation spectral data, the initial non-vegetation spectral data, and the initial vegetation abundance value, initial mixed spectral fitting data are obtained by fitting.

[0058] In this invention, the mixed spectral fitting data is based on a linear mixing model, which combines vegetation and non-vegetation spectra according to vegetation abundance (coverage) to simulate the theoretically obtainable mixed spectral data.

[0059] Specifically, using the initial non-vegetation spectral data B(λ), the initial vegetation spectral data V_initial(λ), and the initial vegetation abundance value Fv (which can be set empirically in the initial stage and adjusted according to the results of each iteration optimization) obtained in step 101, the initial mixed spectral fitting data are calculated according to the following linear mixing formula:

[0060] Fv_initial×V_initial(λ)+(1-Fv_initial)×B(λ).

[0061] This mixed spectrum fitting data is a theoretical mixed spectrum used to compare with the actual mixed spectrum raw data S_mixed(λ).

[0062] Step 103: Based on the bi-objective optimization function, vegetation spectral characteristic constraints, and the initial mixed spectral fitting data, repeat the iterative optimization solution process. If the iterative optimization result meets the preset conditions, obtain the spectral data separation result corresponding to the original mixed spectral data.

[0063] The dual-objective optimization function is constructed with the objective of minimizing the difference between the original mixed spectral data and the fitted mixed spectral data; the vegetation spectral feature constraint is constructed based on the absorption and reflection characteristics of the vegetation spectrum in a preset band.

[0064] In this invention, the bi-objective optimization function is constructed with the goal of minimizing the difference between the original mixed spectral data and the fitted mixed spectral data. The aim is to make the decomposed vegetation spectrum and non-vegetation spectrum as close as possible to the original mixed spectrum after mixing them according to a certain coverage rate.

[0065] Specifically, the goodness-of-fit constraint, which minimizes the difference between the original mixed spectral data and the fitted mixed spectral data, is formulated as follows:

[0066] Min||S_mixed(λ)-[Fv×V(λ)+(1-Fv) ×B(λ)]|| 2 .

[0067] By continuously iterating and optimizing, the vegetation spectral data V(λ) (the vegetation spectral data corresponding to each iteration after the initial stage) and the vegetation abundance value Fv (the vegetation abundance value corresponding to each iteration after the initial stage) are adjusted to minimize this difference and achieve the best fitting effect.

[0068] Meanwhile, the vegetation spectral features in each round of iterative optimization need to meet constraints, namely, vegetation spectral feature constraints. These constraints are constructed based on the absorption and reflection characteristics of vegetation spectra in preset bands. Vegetation spectra have unique characteristics, which play a key role in accurately separating vegetation spectra from non-vegetation spectra.

[0069] In this invention, the vegetation spectral data separated during each round of iterative optimization meets the following conditions:

[0070] Depth constraint: In the red light absorption valley (such as around 680nm), the absorption depth must be greater than a threshold (generally the absorption rate is between 90% and 95%), that is, the red light absorption valley must be deep enough.

[0071] Steepness constraint: In the “red edge” region (680nm to 750nm), its first derivative must be greater than a threshold (the spectral rise must be steep enough, with a first derivative value of 4% to 7% / nm).

[0072] Reflectance threshold constraints: In the near-infrared band (780nm to 900nm), vegetation reflectance ≥40%, non-vegetation spectral reflectance ≤40%, and mixed spectral reflectance lies between the two.

[0073] Furthermore, during the iterative optimization process, optimization algorithms (such as gradient descent, genetic algorithm, least squares method, etc.) are employed. Simultaneously, in each iteration, the vegetation spectrum V(λ) and vegetation cover Fv are adjusted. V(λ) must satisfy the aforementioned vegetation spectrum characteristic constraints; B(λ) must have spectral smoothness, which can be achieved by constraining its second derivative, i.e., |B''(λ)|≦4% / (nm). 2 All spectral values ​​B(λ) are non-negative.

[0074] Before the iterative optimization converges, a corresponding adjustment strategy is executed for the results of each round, as follows:

[0075] If the obtained vegetation spectrum V(λ) is not steep enough in the "red edge" region (i.e., wavelength 680nm to 750nm), that is, the value of V'(λ) is not in the range of 4% to 7% / nm, then the corresponding adjustment is performed based on the range where the value of V'(λ) is located. For example, the value of V(λ) at the near-infrared end (λ≧750nm) is increased by 1%, or the value of V(λ) at the red end (≦680nm) is decreased by 1%, and then the iterative calculation is performed again.

[0076] When the reflectance spectrum of the obtained vegetation spectral data is not within the 90% to 95% range in the red absorption valley (i.e., around 680nm), the reflectance V (λ=680nm) in this band is adjusted by 1% using the gradient descent method, and then the calculation is iterated again.

[0077] If the reflectance of the obtained vegetation reflectance spectrum in the near-infrared band (780nm to 900nm) is less than 40% or greater than 85%, the values ​​of V(λ) in each band should be increased or decreased as a whole. However, the adjustment range still needs to comply with the above reflectance threshold constraints, and then the calculation should be iterated again.

[0078] Even after adjusting the V(λ) value as described above, the resulting vegetation reflectance spectrum and non-vegetation reflectance spectrum still fail to meet the constraints and Min||S_mixed(λ)-[Fv×V(λ)+(1-Fv) ×B(λ)]|| required by the vegetation spectral characteristics. 2 If the condition is ≤0.5%, then continue to adjust the value of Fv.

[0079] If Min||S_mixed(λ)-[Fv×V(λ)+(1-Fv) ×B(λ)]|| 2 If the condition of ≤0.5% cannot be met, the iterative optimization algorithm can stop after 100 iterations. When the iterative optimization result meets the preset conditions (such as the difference condition and constraint conditions mentioned above) or the number of iterations meets the preset number of iterations, the currently obtained V(λ), Fv, and B(λ) are the final results separated, which are the pure vegetation spectrum V_final(λ), the estimated non-vegetation background spectrum B_final(λ), and the vegetation coverage Fv_final corresponding to the original mixed spectral data.

[0080] The spectral data separation method and system provided by this invention calculates initial non-vegetation spectral data using initial vegetation coverage estimates and vegetation spectral data; then, by combining the initial vegetation spectral data, non-vegetation spectral data, and vegetation abundance values, initial mixed spectral fitting data is obtained; finally, based on a bi-objective optimization function constructed with the goal of minimizing the difference between the original and fitted mixed spectral data and vegetation spectral feature constraints, the method iteratively optimizes and solves the problem to obtain the spectral data separation result. This eliminates the need to rely on a large endmember spectral library, simplifies the spectral data separation operation, and improves the automation and adaptability of spectral data separation.

[0081] Based on the above embodiments, the iterative optimization solution process includes:

[0082] Based on the vegetation spectral data, non-vegetation spectral data, and vegetation abundance values ​​of the current round, the mixed spectral fitting data of the current round is obtained.

[0083] Calculate the squared error between the original mixed spectrum data and the fitted mixed spectrum data of the current round;

[0084] If the squared error is less than or equal to the preset minimum spectral fitting squared error, and the vegetation spectral data in the mixed spectral fitting data of the current round satisfies the vegetation spectral feature constraint, the spectral data separation result is generated based on the vegetation spectral data of the current round, the non-vegetation spectral data of the current round, and the vegetation abundance value of the current round.

[0085] If the squared error is determined to be less than or equal to the preset minimum spectral fitting squared error, and the vegetation spectral data in the current round of mixed spectral fitting data does not meet the vegetation spectral feature constraint, the vegetation spectral data of the current round is adjusted based on the vegetation spectral feature constraint, and the adjusted vegetation spectral data is used as the vegetation spectral data for the next round of the iterative optimization solution process.

[0086] If the squared error is determined to be greater than the preset minimum spectral fitting squared error, and the vegetation spectral data in the current round of mixed spectral fitting data does not meet the vegetation spectral feature constraint conditions, the vegetation abundance value of the current round is adjusted, and the adjusted vegetation abundance value is used as the vegetation abundance value of the next round of the iterative optimization solution process.

[0087] The iterative optimization solution process aims to continuously adjust the vegetation spectral data, non-vegetation spectral data, and vegetation abundance value so that the fitted mixed spectrum is as close as possible to the original mixed spectrum, while ensuring that the decomposed vegetation spectrum meets specific vegetation spectral feature constraints. Finally, it obtains accurate spectral data separation results, namely, pure vegetation spectrum, non-vegetation background spectrum, and vegetation coverage.

[0088] In this invention, based on a linear mixing model, it is assumed that the mixed spectrum is a linear combination of vegetation and non-vegetation spectra according to a certain vegetation abundance (coverage). Using the vegetation spectral data, non-vegetation spectral data, and vegetation abundance values ​​obtained in the current round, the mixed spectrum fitting data for the current round is calculated according to the linear mixing formula, for example, S_fit(λ) = Fv × V(λ) + (1 - Fv) × B(λ), where S_fit(λ) is the fitted mixed spectrum data, Fv is the vegetation abundance value, V(λ) is the vegetation spectral data, and B(λ) is the non-vegetation spectral data. This step is to obtain a theoretically valid mixed spectrum under the current parameters for comparison with the original mixed spectrum.

[0089] Furthermore, the squared error between the original mixed spectrum data S_mixed(λ) and the fitted mixed spectrum data S_fit(λ) of the current round is calculated using the formula ||S_mixed(λ)-[Fv×V (λ)+(1-Fv)×B(λ)|| 2 The smaller this error value, the better the fitting effect and the closer it is to the original mixed spectrum.

[0090] When the fitted mixed spectrum is sufficiently close to the original mixed spectrum (the squared error is less than or equal to the preset minimum spectral fitting squared error), and the decomposed vegetation spectrum satisfies its unique characteristic constraints, it indicates that the current parameter combination has achieved a good separation effect, and the results can be output.

[0091] Specifically, if the squared error is less than or equal to the preset minimum spectral fitting squared error threshold (e.g., 0.5%), and the vegetation spectral data in the current round of mixed spectral fitting data meets the vegetation spectral characteristic constraints (e.g., absorption depth of the red absorption valley, steepness of the red edge region, reflectance threshold of the near-infrared band, etc.), the vegetation spectral data, non-vegetation spectral data, and vegetation abundance value of the current round are used as the final spectral data separation results, thus obtaining the pure vegetation spectrum, the estimated non-vegetation background spectrum, and the vegetation coverage.

[0092] In this invention, although the fitted mixed spectrum is close to the original mixed spectrum, the vegetation spectrum does not meet its characteristic constraints, indicating that there is a deviation in the shape or characteristics of the vegetation spectrum. Therefore, the vegetation spectral data needs to be adjusted to conform to the true characteristics of the vegetation. For example, if the squared error is determined to be less than or equal to a preset threshold for minimizing the squared error of the spectral fitting, but the vegetation spectral data in the current round of mixed spectrum fitting data does not meet the vegetation spectral characteristic constraints, the vegetation spectral data for the current round is adjusted based on these characteristic constraints (e.g., if the red absorption valley is not deep enough, the values ​​of the vegetation spectrum near the red absorption valley are adjusted according to certain rules; if the red edge region is not steep enough, the spectral values ​​of the red edge region are adjusted, etc.). The adjusted vegetation spectral data will be used as the vegetation spectral data for the next round of the iterative optimization solution process, and subsequent fitting and judgment will be performed again.

[0093] In this invention, when the fitted mixed spectrum differs significantly from the original mixed spectrum, and the vegetation spectrum does not meet the feature constraints, it indicates that the current vegetation abundance value may be inappropriate, and adjustment of the vegetation abundance value is needed to improve the fitting effect and vegetation spectral characteristics. For example, if the squared error is determined to be greater than the preset threshold for minimizing the squared error of spectral fitting, and the vegetation spectral data in the current round of mixed spectrum fitting data does not meet the vegetation spectral feature constraints, the vegetation abundance value of the current round is adjusted. The adjustment method can be determined according to specific circumstances, such as increasing or decreasing the vegetation abundance value by a certain step size. The adjusted vegetation abundance value will be used as the vegetation abundance value for the next round of the iterative optimization solution process, continuing subsequent fitting and judgment until the stopping condition is met (such as reaching the maximum number of iterations or meeting dual conditions).

[0094] Based on the above embodiments, the vegetation spectral characteristic constraints include vegetation spectral absorption depth constraints, vegetation spectral curve steepness constraints, and spectral reflectance threshold constraints, wherein:

[0095] If the absorption rate of the vegetation spectral data in the current round is greater than or equal to the lower limit of the preset absorption rate and less than or equal to the upper limit of the preset absorption rate in the preset band, it is determined that the vegetation spectral data in the current round satisfies the vegetation spectral absorption depth constraint condition.

[0096] If the first derivative of the vegetation spectral data of the current round is greater than or equal to the lower limit of the preset derivative and less than or equal to the upper limit of the preset derivative, then the vegetation spectral data of the current round is determined to satisfy the vegetation spectral curve steepness constraint condition.

[0097] If the vegetation reflectance of the current round's vegetation spectral data in the second preset spectral band range is greater than or equal to the preset lower limit of reflectance and less than or equal to the preset upper limit of reflectance, then the vegetation spectral data of the current round is determined to satisfy the spectral reflectance threshold constraint condition.

[0098] In this invention, the vegetation spectral characteristic constraint condition is to ensure that the vegetation spectral data decomposed from the mixed spectrum conforms to the real physical characteristics of the vegetation. By setting specific constraints on three key aspects of the vegetation spectrum, namely absorption depth, curve steepness and spectral reflectance, the decomposed vegetation spectrum is screened and corrected to make it closer to the real vegetation spectrum.

[0099] Vegetation exhibits significant absorption characteristics in the red light absorption valley (e.g., around 680 nm), due to the strong absorption of red light by chlorophyll in plant leaves. Absorption depth reflects the vegetation's ability to absorb red light; deeper absorption indicates potentially higher chlorophyll content and photosynthetic activity, consistent with the vegetation's physiological characteristics. In this invention, if the absorptivity of the current round of vegetation spectral data in a preset band (i.e., the red light absorption valley band, such as around 680 nm) is greater than or equal to a preset lower limit (e.g., 90%) and less than or equal to a preset upper limit (e.g., 95%), then the current round of vegetation spectral data is determined to meet the vegetation spectral absorption depth constraint condition. This indicates that the degree of absorption of the separated vegetation spectrum in this band must be within a reasonable range—neither too weak (below the lower limit) nor abnormally strong (above the upper limit). Only within this range does it conform to the true absorption characteristics of vegetation in the red light absorption valley. For example, if the preset lower limit of absorption rate is 90% and the upper limit is 95%, the absorption depth constraint condition is met when the absorption rate of the separated vegetation spectrum at 680nm is 92%. If the absorption rate is 85%, it is not met, indicating that the absorption of the spectrum in this band is too weak, which may not be the true vegetation spectrum or there may be a problem with the separation process.

[0100] Vegetation spectra exhibit unique spectral characteristics in the "red edge" region (680nm to 750nm), where spectral reflectance rises rapidly, forming a steep curve. This characteristic arises from the shift in vegetation leaves from strong absorption of red light to strong reflection of near-infrared light, reflecting the health and physiological activity of the vegetation. A greater steepness suggests potentially better photosynthetic capacity and growth status. In this invention, the vegetation spectral data of the current round is deemed to satisfy the vegetation spectral curve steepness constraint if the first derivative of the current round's data within the first preset spectral band range (i.e., the "red edge" region 680nm to 750nm) is greater than or equal to the preset lower limit of the derivative (e.g., 4% / nm) and less than or equal to the preset upper limit of the derivative (e.g., 7% / nm). The first derivative reflects the rate of ascent of the spectral curve in this band; by setting a reasonable derivative range, it can be ensured that the separated vegetation spectrum in the "red edge" region exhibits a steep ascent characteristic consistent with real vegetation. For example, if the preset lower limit of the derivative is 4% / nm and the upper limit is 7% / nm, the steepness constraint is satisfied when the first derivative of the separated vegetation spectrum in the "red edge" region is 5% / nm; if the first derivative is 3% / nm, it is not satisfied, indicating that the rise of the spectrum in the "red edge" region is not steep enough and may not be the true vegetation spectrum.

[0101] Different ground features exhibit different reflectance characteristics in different wavelength bands. Vegetation has high reflectance in the near-infrared band (780nm to 900nm), due to the multiple scattering of near-infrared light by the internal structure of vegetation leaves. Non-vegetation features (such as soil and water bodies) have relatively low reflectance in this band. By setting reflectance thresholds for vegetation, non-vegetation, and mixed spectra in the near-infrared band, vegetation and non-vegetation spectra can be further distinguished, ensuring that the decomposed vegetation spectrum has a reasonable reflectance value in the near-infrared band. In this invention, if the vegetation spectral data of the current round has a reflectance in the second preset spectral band range (i.e., the near-infrared band 780nm to 900nm) that is greater than or equal to a preset lower limit (e.g., 40%) and less than or equal to a preset upper limit (e.g., 85%, which can be adjusted according to the situation in practical applications), then the vegetation spectral data of the current round is determined to meet the spectral reflectance threshold constraint condition. Simultaneously, the reflectance of the non-vegetation spectrum in this band should be less than or equal to 40%, and the reflectance of the mixed spectrum should be between the two. This constraint ensures that the decomposed vegetation spectrum has a reflectance value in the near-infrared band that matches the true characteristics of vegetation. For example, if the preset lower limit of reflectance is 40%, the reflectance threshold constraint is satisfied when the reflectance of the decomposed vegetation spectrum at 800nm ​​is 50%. If the reflectance is 30%, it is not satisfied, indicating that the reflectance of the spectrum in the near-infrared band is too low and may not be the true vegetation spectrum.

[0102] In this invention, in addition to the three main constraints mentioned above, the non-vegetation spectrum B(λ) can also be constrained and adjusted, requiring B(λ) to be spectrally smooth. This can be achieved by constraining its second derivative, i.e., |B''(λ)|≦4% / (nm). 2 This avoids abnormal fluctuations and sharp changes in the spectra of non-vegetated areas, making them more consistent with the smooth spectral characteristics of non-vegetated features (such as soil and water bodies). Simultaneously, all spectral values ​​B(λ) are non-negative, based on the physical meaning of spectral reflectance, which cannot be negative. These supplementary constraints further improve the accuracy and rationality of spectral decomposition.

[0103] Based on the above embodiments, the method further includes:

[0104] When the absorptivity of the vegetation spectral data of the current round in the preset band is less than the preset absorptivity lower limit, the reflectivity of the vegetation spectral data of the current round in the preset band is adjusted upward by a preset adjustment range to obtain the adjusted vegetation spectral data.

[0105] When the absorption rate of the vegetation spectral data of the current round in the preset band is greater than the upper limit of the preset absorption rate, the reflectance of the vegetation spectral data of the current round in the preset band is adjusted downward by a preset adjustment range to obtain the adjusted vegetation spectral data.

[0106] In this invention, when it is detected that the absorption rate of the vegetation spectrum obtained from the current round of decomposition in a specific preset band (red light absorption valley, such as around 680nm) does not meet the preset reasonable range (90% to 95%), the vegetation spectral data is corrected by adjusting the reflectance of that band, so that it is more consistent with the actual absorption characteristics of vegetation in the red light absorption valley, thereby improving the accuracy of spectral data separation.

[0107] In this invention, the preset wavelength band is the red light absorption valley, generally around 680 nm. Vegetation exhibits a significant absorption characteristic in this band due to the strong absorption of red light by chlorophyll, making this band one of the important criteria for judging the rationality of the vegetation spectrum. The preset absorption rate range can be set to 90% to 95%. The absorption rate reflects the degree of red light absorption by the vegetation, and this range is determined based on the actual physiological characteristics and spectral features of the vegetation. If the absorption rate is within this range, it indicates that the absorption of the vegetation spectrum in this wavelength band is consistent with the behavior of normal vegetation; if it is not within this range, it indicates that there may be a problem with the spectral data, and adjustments are needed.

[0108] Specifically, if the absorbance of the vegetation spectrum at 680 nm in the current round is lower than 90% or higher than 95%, subsequent adjustment operations will be triggered. For example, if the calculated absorbance of the current round's vegetation spectrum at 680 nm is 85%, which is less than the lower limit of 90%; or if the absorbance is 96%, which is greater than the upper limit of 95%, the adjustment conditions are met. In this invention, the preset adjustment range can be set to 1%. That is, each time an adjustment is made, the reflectance of the current round's vegetation spectrum in the preset band (around 680 nm) is decreased or increased by 1%.

[0109] In this invention, gradient descent can be used to adjust the reflectance of this wavelength band. Gradient descent is an optimization algorithm that finds the optimal solution by progressively adjusting the parameter (reflectance) along the negative gradient direction of the objective function (i.e., making the spectrum more consistent with the true characteristics). In practice, the reflectance is adjusted in increments of 1% based on the current reflectance value. For example, if the reflectance of the current vegetation spectrum at 680nm is 20%, after one adjustment, the reflectance becomes 21%.

[0110] After adjusting the reflectance as described above, the adjusted vegetation spectral data is obtained. This adjusted data will serve as the input for the next round of iterative calculations, where spectral fitting, error calculation, and checks for compliance with constraints will be performed again. Through continuous iteration, the vegetation spectral data will gradually become closer to the actual situation.

[0111] Based on the above embodiments, the method further includes:

[0112] When the first derivative of the vegetation spectral data of the current round is less than the lower limit of the preset derivative in the first preset spectral band range, the spectral value of the vegetation spectral data of the current round corresponding to the upper limit of the spectral band in the first preset spectral band range is adjusted upward by a preset adjustment range to obtain the adjusted vegetation spectral data.

[0113] When the first derivative of the vegetation spectral data of the current round is greater than the upper limit of the preset derivative in the first preset spectral band range, the spectral value of the vegetation spectral data of the current round corresponding to the upper limit of the spectral band in the first preset spectral band range is adjusted downward according to the preset adjustment range to obtain the adjusted vegetation spectral data.

[0114] In this invention, when the steepness of vegetation spectral data in a specific spectral region (“red edge” region) does not meet the requirements, a method of adjusting the spectral values ​​and re-iteratio calculation is adopted to optimize the vegetation spectral data so that it better matches the true spectral characteristics of the vegetation.

[0115] The first preset spectral band range represents the "red edge" region of the vegetation spectrum, with a wavelength range of 680nm to 750nm. This region is a key characteristic area of ​​the vegetation spectrum, where vegetation changes from strong absorption of red light to strong reflection of near-infrared light, and the spectral reflectance rises rapidly, forming a steep curve. This characteristic is closely related to the health status and physiological activity of the vegetation.

[0116] In this invention, the preset lower limit of the derivative is 4% / nm, and the upper limit is 7% / nm. The first derivative reflects the slope of the spectral curve at a certain point. In the "red edge" region, the magnitude of the first derivative reflects the steepness of the increase in spectral reflectance. When the first derivative is within this range, it indicates that the steepness of the increase in vegetation spectrum in the "red edge" region is consistent with the characteristics of normal vegetation.

[0117] When the first derivative of the calculated vegetation spectrum in the "red edge" region (680nm to 750nm) is less than 4% / nm or greater than 7% / nm, the steepness of the vegetation spectrum in the "red edge" region is deemed unacceptable and requires adjustment. In this invention, adjustments are made with a preset adjustment range, such as 1%, which is a specific quantitative standard for adjusting the spectral values. This standard is used to precisely control the degree of adjustment and avoid over- or under-adjustment.

[0118] Specifically, when the first derivative of the current round of vegetation spectral data is less than the lower limit of the preset derivative within the first preset spectral band range, the spectral value of the current round of vegetation spectral data at the near-infrared end (i.e., to the upper limit of the spectral band, with a wavelength greater than or equal to 750 nm) is adjusted upwards by a preset margin of 1%. This is because the near-infrared end is a region with high vegetation spectral reflectance; increasing the spectral value in this region allows the spectral curve to rise more rapidly after the "red edge" region, thus increasing the steepness of the "red edge" region. For example, if the reflectance of the current round of vegetation spectrum at 760 nm is 60%, after an upward adjustment of 1%, it becomes 61%.

[0119] When the first derivative of the current round of vegetation spectral data in the first preset spectral band range is greater than the upper limit of the preset derivative, the spectral values ​​of the current round of vegetation spectral data on the side of the upper limit of the spectral band are adjusted downward by a preset margin of 1%. For example, if the reflectance of the current round of vegetation spectrum at 750nm is 10%, after adjusting downward by 1%, it becomes 9%. After the above adjustment of the near-infrared spectral values, the adjusted vegetation spectral data is obtained. This adjusted data will be used as the input for the next round of iteration calculation, and operations such as mixed spectrum fitting, error calculation, and determination of whether the vegetation spectral characteristic constraints are met will be performed again.

[0120] Based on the above embodiments, the method further includes:

[0121] When the vegetation reflectance of the current round's vegetation spectral data in the second preset spectral band is less than the preset lower limit value, the reflectance of the current round's vegetation spectral data in each spectral band within the range of the second preset spectral band is increased to obtain the adjusted vegetation spectral data.

[0122] Alternatively, if the vegetation reflectance of the current round's vegetation spectral data in the second preset spectral band is greater than the preset upper limit of reflectance, the reflectance of the current round's vegetation spectral data in each spectral band within the second preset spectral band range is reduced to obtain the adjusted vegetation spectral data.

[0123] In this invention, an adjustment strategy is implemented for vegetation spectral data when reflectance is abnormal in a specific near-infrared band (a second preset spectral band). When the reflectance is below a preset lower limit or above a preset upper limit, the reflectance of each spectral band within the band is increased or decreased to make the vegetation spectral data more within a reasonable range. Then, the data is recalculated iteratively based on the adjusted data, and the adjustment process must follow specific constraints.

[0124] In this invention, the second preset spectral band refers to the near-infrared band, with a wavelength range of 780nm to 900nm. Within this band, vegetation exhibits strong reflectivity to near-infrared light due to factors such as the internal structure of its leaves, and its reflectivity can reflect some physiological and structural information about the vegetation. The preset lower limit for reflectivity is 40%, which is the threshold for judging whether the reflectivity of the vegetation spectrum in the near-infrared band is too low. When the reflectivity is below 40%, it indicates that the reflectivity of the vegetation in this band may deviate from the normal range, possibly due to poor vegetation growth, measurement errors, or other factors. The preset upper limit for reflectivity is 85%, which is the threshold for judging whether the reflectivity of the vegetation spectrum in the near-infrared band is too high. When the reflectivity is above 85%, it also indicates that the vegetation spectrum may be abnormal and needs adjustment.

[0125] When the reflectance of the current vegetation spectrum in the 780nm to 900nm band is less than 40% as determined by calculation or measurement, a reflectance increase operation is triggered. When the reflectance of the vegetation spectrum in this band is greater than 85%, a reflectance decrease operation is triggered. For example, assuming the reflectance of the current vegetation spectrum at 800nm ​​is 35%, which is less than the lower limit of 40%, an increase will be performed; if the reflectance at 850nm is 90%, which is greater than the upper limit of 85%, a decrease will be performed.

[0126] In this invention, when the reflectivity is less than the lower limit, the reflectivity of each spectral band in the 780nm to 900nm band is increased. The increase can be set according to the actual situation, with the aim of raising the reflectivity to a reasonable range.

[0127] When the reflectivity exceeds the upper limit, the reflectivity of each spectral band within the 780nm to 900nm band is reduced to a reasonable range.

[0128] Simultaneously, when adjusting reflectance, it is essential to ensure that the adjusted data still meets these pre-defined constraints. After obtaining the adjusted vegetation spectral data, it is substituted into the original calculation model or algorithm for iterative recalculation. Through continuous optimization and adjustment, the final results, such as vegetation spectra, non-vegetation spectra, and coverage, become more accurate and reliable.

[0129] Figure 2 This is a schematic diagram comparing the effects of separating mixed spectral data before and after the present invention, which can be used as a reference. Figure 2 As shown, during the separation process, the vegetation spectrum must meet constraints such as absorption depth, red edge steepness, and near-infrared reflectance; the non-vegetation spectrum must remain smooth and non-negative. Through the fitting algorithm and physical constraints provided by this invention, vegetation and non-vegetation spectra and coverage rates that conform to the true spectral characteristics can be reconstructed.

[0130] This invention deeply integrates "physical constraints" with "optimization solution." By incorporating hard constraints driven by "physics" (i.e., vegetation spectral feature constraints), the solution found by the optimization algorithm not only fits mathematically well but also represents the "true" vegetation spectrum (with deep valleys and steep edges), effectively avoiding the "non-physical solutions" that may arise from pure mathematical decomposition. Furthermore, this invention does not require prior knowledge of any endmembers; it starts from a guessed initial value and simultaneously optimizes the endmember spectra and their proportions, making its application scenarios more flexible. Simultaneously, this invention explicitly lists "goodness of fit" and "vegetation features" as dual objectives and proposes quantitative constraints on specific features of the vegetation spectrum (depth, steepness, reflectance threshold) and the smoothness of the background spectrum, making this invention highly specific and operable.

[0131] The spectral data separation system provided by the present invention is described below. The spectral data separation system described below can be referred to in correspondence with the spectral data separation method described above.

[0132] Figure 3 A schematic diagram of the spectral data separation system provided by the present invention is shown below. Figure 3As shown, this invention provides a spectral data separation system, including a first processing module 301, a second processing module 302, and a spectral data separation module 303. The first processing module 301 calculates the initial non-vegetation spectral data corresponding to the mixed spectral raw data to be processed based on an initial vegetation cover estimate and initial vegetation spectral data. The second processing module 302 fits the initial mixed spectral data based on the initial vegetation spectral data, the initial non-vegetation spectral data, and the initial vegetation abundance value. The spectral data separation module 303 repeatedly executes an iterative optimization process based on a bi-objective optimization function, vegetation spectral feature constraints, and the initial mixed spectral fitting data. When the iterative optimization result satisfies preset conditions, it obtains the spectral data separation result corresponding to the mixed spectral raw data. The bi-objective optimization function is constructed with the objective of minimizing the difference between the mixed spectral raw data and the mixed spectral fitting data. The vegetation spectral feature constraints are constructed based on the absorption and reflection characteristics of the vegetation spectrum in a preset band.

[0133] The spectral data separation system provided by this invention calculates initial non-vegetation spectral data from initial vegetation coverage estimates and vegetation spectral data; then, by combining the initial vegetation spectral data, non-vegetation spectral data, and vegetation abundance values, it obtains initial mixed spectral fitting data; finally, based on a bi-objective optimization function constructed with the goal of minimizing the difference between the original and fitted mixed spectral data and vegetation spectral feature constraints, it iterative optimization is performed to obtain the spectral data separation result. This eliminates the need to rely on a large endmember spectral library, simplifies the spectral data separation operation, and improves the automation and adaptability of spectral data separation.

[0134] The system provided in this embodiment of the invention is used to execute the above-described method embodiments. For specific processes and details, please refer to the above embodiments, which will not be repeated here.

[0135] Figure 4 This is a schematic diagram of the structure of the electronic device provided by the present invention, such as... Figure 4As shown, the electronic device may include: a processor 401, a communications interface 402, a memory 403, and a communication bus 404, wherein the processor 401, the communications interface 402, and the memory 403 communicate with each other through the communication bus 404. The processor 401 can call logic instructions in the memory 403 to execute a spectral data separation method. This method includes: calculating initial non-vegetation spectral data corresponding to the original mixed spectral data to be processed based on an initial vegetation cover estimate and initial vegetation spectral data; fitting initial mixed spectral fitting data based on the initial vegetation spectral data, the initial non-vegetation spectral data, and the initial vegetation abundance value; repeatedly executing an iterative optimization solution process based on a bi-objective optimization function, vegetation spectral feature constraints, and the initial mixed spectral fitting data; and obtaining the spectral data separation result corresponding to the original mixed spectral data when the iterative optimization result meets preset conditions. The bi-objective optimization function is constructed with the objective of minimizing the difference between the original mixed spectral data and the fitted mixed spectral data; the vegetation spectral feature constraints are constructed based on the absorption and reflection characteristics of the vegetation spectrum in a preset band.

[0136] Furthermore, the logical instructions in the aforementioned memory 403 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0137] On the other hand, the present invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium. The computer program includes program instructions, which, when executed by a computer, enable the computer to execute the spectral data separation method provided by the above methods. The method includes: calculating initial non-vegetation spectral data corresponding to the raw mixed spectral data to be processed based on an initial vegetation cover estimate and initial vegetation spectral data; fitting initial mixed spectral fitting data based on the initial vegetation spectral data, the initial non-vegetation spectral data, and the initial vegetation abundance value; repeatedly executing an iterative optimization solution process based on a bi-objective optimization function, vegetation spectral feature constraints, and the initial mixed spectral fitting data; and obtaining the spectral data separation result corresponding to the raw mixed spectral data when the iterative optimization result meets preset conditions. The bi-objective optimization function is constructed with the objective of minimizing the difference between the raw mixed spectral data and the fitted mixed spectral data; the vegetation spectral feature constraints are constructed based on the absorption and reflection characteristics of the vegetation spectrum in a preset band.

[0138] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements the spectral data separation method provided in the above embodiments. The method includes: calculating initial non-vegetation spectral data corresponding to the raw mixed spectral data to be processed based on an initial vegetation cover estimate and initial vegetation spectral data; fitting initial mixed spectral fitting data based on the initial vegetation spectral data, the initial non-vegetation spectral data, and the initial vegetation abundance value; repeatedly executing an iterative optimization solution process based on a bi-objective optimization function, vegetation spectral feature constraints, and the initial mixed spectral fitting data; and obtaining the spectral data separation result corresponding to the raw mixed spectral data when the iterative optimization result satisfies a preset condition. The bi-objective optimization function is constructed with the objective of minimizing the difference between the raw mixed spectral data and the fitted mixed spectral data; the vegetation spectral feature constraints are constructed based on the absorption and reflection characteristics of the vegetation spectrum in a preset band.

[0139] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0140] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0141] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for separating spectral data, characterized in that, include: Based on the initial vegetation cover estimate and the initial vegetation spectral data, calculate the initial non-vegetation spectral data corresponding to the raw mixed spectral data to be processed; Based on the initial vegetation spectral data, the initial non-vegetation spectral data, and the initial vegetation abundance value, initial mixed spectral fitting data are obtained by fitting. Based on the dual-objective optimization function, the vegetation spectral characteristic constraints, and the initial mixed spectral fitting data, the iterative optimization solution process is repeatedly executed. When the iterative optimization result meets the preset conditions, the spectral data separation result corresponding to the original mixed spectral data is obtained. The dual-objective optimization function is constructed with the objective of minimizing the difference between the original mixed spectral data and the fitted mixed spectral data; the vegetation spectral characteristic constraint is constructed based on the absorption and reflection characteristics of the vegetation spectrum in a preset band. The vegetation spectral characteristic constraints include vegetation spectral absorption depth constraints, vegetation spectral curve steepness constraints, and spectral reflectance threshold constraints, wherein: If the absorption rate of the vegetation spectral data in the current round is greater than or equal to the preset lower limit of absorption rate and less than or equal to the preset upper limit of absorption rate in the preset band, then the vegetation spectral data in the current round is determined to meet the vegetation spectral absorption depth constraint condition. If the first derivative of the vegetation spectral data of the current round is greater than or equal to the lower limit of the preset derivative and less than or equal to the upper limit of the preset derivative, then the vegetation spectral data of the current round is determined to satisfy the vegetation spectral curve steepness constraint condition. If the vegetation spectral data of the current round has a vegetation reflectance greater than or equal to a preset lower limit value and less than or equal to a preset upper limit value in the second preset spectral band range, then the vegetation spectral data of the current round is determined to satisfy the spectral reflectance threshold constraint condition.

2. The spectral data separation method according to claim 1, characterized in that, The iterative optimization solution process includes: Based on the vegetation spectral data, non-vegetation spectral data, and vegetation abundance values ​​of the current round, the mixed spectral fitting data of the current round is obtained. Calculate the squared error between the original mixed spectrum data and the fitted mixed spectrum data of the current round; If the squared error is less than or equal to the preset minimum spectral fitting squared error, and the vegetation spectral data in the mixed spectral fitting data of the current round satisfies the vegetation spectral feature constraint, the spectral data separation result is generated based on the vegetation spectral data of the current round, the non-vegetation spectral data of the current round, and the vegetation abundance value of the current round. If the squared error is determined to be less than or equal to the preset minimum spectral fitting squared error, and the vegetation spectral data in the current round of mixed spectral fitting data does not meet the vegetation spectral feature constraint, the vegetation spectral data of the current round is adjusted based on the vegetation spectral feature constraint, and the adjusted vegetation spectral data is used as the vegetation spectral data for the next round of the iterative optimization solution process. If the squared error is determined to be greater than the preset minimum spectral fitting squared error, and the vegetation spectral data in the current round of mixed spectral fitting data does not meet the vegetation spectral feature constraint conditions, the vegetation abundance value of the current round is adjusted, and the adjusted vegetation abundance value is used as the vegetation abundance value of the next round of the iterative optimization solution process.

3. The spectral data separation method according to claim 2, characterized in that, The method further includes: When the absorptivity of the vegetation spectral data of the current round in the preset band is less than the preset absorptivity lower limit, the reflectivity of the vegetation spectral data of the current round in the preset band is adjusted upward by a preset adjustment range to obtain the adjusted vegetation spectral data. When the absorption rate of the vegetation spectral data of the current round in the preset band is greater than the upper limit of the preset absorption rate, the reflectance of the vegetation spectral data of the current round in the preset band is adjusted downward by a preset adjustment range to obtain the adjusted vegetation spectral data.

4. The spectral data separation method according to claim 2, characterized in that, The method further includes: When the first derivative of the vegetation spectral data of the current round is less than the lower limit of the preset derivative in the first preset spectral band range, the spectral value of the vegetation spectral data of the current round corresponding to the upper limit of the spectral band in the first preset spectral band range is adjusted upward by a preset adjustment range to obtain the adjusted vegetation spectral data. When the first derivative of the vegetation spectral data of the current round is greater than the upper limit of the preset derivative in the first preset spectral band range, the spectral value of the vegetation spectral data of the current round corresponding to the upper limit of the spectral band in the first preset spectral band range is adjusted downward according to the preset adjustment range to obtain the adjusted vegetation spectral data.

5. The spectral data separation method according to claim 2, characterized in that, The method further includes: When the vegetation reflectance of the current round's vegetation spectral data in the second preset spectral band is less than the preset lower limit value, the reflectance of the current round's vegetation spectral data in each spectral band within the range of the second preset spectral band is increased to obtain the adjusted vegetation spectral data. Alternatively, if the vegetation reflectance of the current round's vegetation spectral data in the second preset spectral band is greater than the preset upper limit of reflectance, the reflectance of the current round's vegetation spectral data in each spectral band within the second preset spectral band range is reduced to obtain the adjusted vegetation spectral data.

6. A spectral data separation system, characterized in that, include: The first processing module is used to calculate the initial non-vegetation spectral data corresponding to the mixed spectral raw data to be processed based on the initial vegetation coverage estimate and the initial vegetation spectral data. The second processing module is used to fit initial mixed spectral fitting data based on the initial vegetation spectral data, the initial non-vegetation spectral data, and the initial vegetation abundance value. The spectral data separation module is used to repeatedly execute the iterative optimization solution process based on the bi-objective optimization function, vegetation spectral characteristic constraints and the initial mixed spectral fitting data, and obtain the spectral data separation result corresponding to the original mixed spectral data when the iterative optimization result meets the preset conditions. The dual-objective optimization function is constructed with the objective of minimizing the difference between the original mixed spectral data and the fitted mixed spectral data; the vegetation spectral characteristic constraint is constructed based on the absorption and reflection characteristics of the vegetation spectrum in a preset band. The vegetation spectral characteristic constraints include vegetation spectral absorption depth constraints, vegetation spectral curve steepness constraints, and spectral reflectance threshold constraints, wherein: If the absorption rate of the vegetation spectral data in the current round is greater than or equal to the preset lower limit of absorption rate and less than or equal to the preset upper limit of absorption rate in the preset band, then the vegetation spectral data in the current round is determined to meet the vegetation spectral absorption depth constraint condition. If the first derivative of the vegetation spectral data of the current round is greater than or equal to the lower limit of the preset derivative and less than or equal to the upper limit of the preset derivative, then the vegetation spectral data of the current round is determined to satisfy the vegetation spectral curve steepness constraint condition. If the vegetation spectral data of the current round has a vegetation reflectance greater than or equal to a preset lower limit value and less than or equal to a preset upper limit value in the second preset spectral band range, then the vegetation spectral data of the current round is determined to satisfy the spectral reflectance threshold constraint condition.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the spectral data separation method as described in any one of claims 1 to 5.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the spectral data separation method as described in any one of claims 1 to 5.

9. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the spectral data separation method as described in any one of claims 1 to 5.

Images