Crop FAPAR inversion method and system based on adaptive spectral unmixing

Abstract

The invention provides a crop FAPAR inversion method and system based on adaptive spectral unmixing, and the method comprises the steps: obtaining a close-range multispectral image and a full-coverage multispectral image of a crop sample field, and obtaining the per-pixel reflectivity of each multispectral image; according to the pixel-by-pixel reflectivity of the close-range multispectral image, constructing a simplified spectrum library containing foreground and background end member spectrums based on an adaptive end member extraction method; constructing a crop-oriented pixel-by-pixel unmixing model based on the multiple linear model; based on an exhaustion cross pairing strategy, solving the full-coverage multispectral image pixel by pixel by using a simplified spectrum library and a crop-oriented pixel-by-pixel unmixing model, and obtaining a pixel-by-pixel foreground abundance unmixing result; fitting a linear regression model of the foreground abundance and the FAPAR, and inverting the FAPAR of the crops in the sample field based on the linear regression model.

Description

Technical Field

[0001] This invention belongs to the field of agricultural remote sensing, specifically relating to a crop FAPAR inversion method and system based on adaptive spectral unmixing. Background Technology

[0002] The fraction of absorbed photosynthetically active radiation (FAPAR) is a parameter describing the proportion of sunlight absorbed by vegetation during canopy radiative transmission. It reflects the canopy's ability to capture and absorb light energy, directly impacting vegetation growth, yield, and product quality. Therefore, accurate estimation and monitoring of FAPAR are crucial for optimizing crop production management, selecting high-photometric crop varieties, monitoring carbon cycles, and addressing climate change. With the increasing demand for economical, efficient, and convenient monitoring methods, remote sensing technology has become an important tool for estimating FAPAR.

[0003] Traditional FAPAR remote sensing estimation methods primarily target canopy FAPAR. Empirical inversion methods typically establish correlations between various vegetation indices and measured or simulated surface FAPAR through simple linear regression. These methods have few model parameters and are computationally simple and efficient. However, the signals received by sensors are usually the result of multiple scattering and reflections of light within the scene. Differences in canopy morphology at different growth stages and the presence of irrelevant ground features such as soil can affect the establishment of empirical relationships, resulting in weak universality and generalization ability of VI (visual index). Spectral mixing analysis (SMA) offers a new approach to overcome this limitation.

[0004] Spectral mixture analysis (SMA) separates different ground cover signals from pixel spectra by inputting different endmembers into a spectral mixture model and solving it, thereby obtaining information about the target ground cover at the sub-pixel scale, i.e., abundance. SMA is beneficial for improving the classification accuracy of ground covers, enhancing the ability to distinguish small targets such as crop organs, and improving the qualitative and quantitative application accuracy of remote sensing. It is currently widely used in large-scale mapping such as land classification and fire severity assessment, as well as specific application scenarios such as shadow removal and leaf lesion detection. Abundance information can characterize the proportion of vegetation information, which is closely related to the proportion of incident solar radiation intercepted by the plant canopy. However, SMA is relatively less used in crop growth scenarios, and the application value of model accuracy and abundance needs further exploration. Sorghum and rice, as representatives of C4 and C3 plants respectively, have different photosynthetic pathways, canopy structures, and growth environments. After emergence, the stems and leaves always intertwine in three-dimensional space in various postures. As they grow and develop, the canopy volume continuously increases until the panicle emerges from the top. Sorghum has a relatively sparse canopy, with larger ears located at the top of the plant; rice has a sparse canopy in the early stages, but shows obvious canopy closure as it grows through jointing. Soil or water, together with the crop canopy, constitutes a dynamically changing, multi-layered mixed environment. However, the complexity and dynamic nature of this environment can affect the establishment of empirical relationships between VI and FAPAR. Summary of the Invention

[0005] To overcome the limitations of existing FAPAR remote sensing estimation techniques, which suffer from weak universality and generalization ability of empirical methods, spectral mixture analysis offers a breakthrough approach. However, spectral mixture analysis is relatively rarely applied in crop growth scenarios. To address these shortcomings, this invention provides a crop FAPAR inversion method and system based on adaptive spectral unmixing. By proposing an adaptive spectral unmixing strategy and solving for abundance based on this strategy, the invention better explains endmember variability and scenario complexity, and further performs inversion, thus realizing a crop FAPAR inversion method based on adaptive spectral unmixing. The strategy is effective and the inversion accuracy is high.

[0006] According to one aspect of the present invention, a crop FAPAR inversion method based on adaptive spectral unmixing is provided, comprising:

[0007] Acquire close-up multispectral images and full-coverage multispectral images of crop sample fields, and obtain the reflectance of each pixel in each multispectral image;

[0008] Based on the pixel-wise reflectance of near-field multispectral images, a simplified spectral library containing foreground and background endmember spectra is constructed using an adaptive endmember extraction method.

[0009] A pixel-by-pixel unmixing model for crops is constructed based on a multilinear model;

[0010] Based on the exhaustive cross-pairing strategy, a simplified spectral library and a crop-oriented pixel-by-pixel unmixing model are used to solve the full-coverage multispectral image pixel by pixel to obtain the foreground abundance unmixing results pixel by pixel.

[0011] Based on the measured FAPAR of crop sample fields and the unmixing results of pixel-by-pixel foreground abundance from full-coverage multispectral images, a linear regression model of foreground abundance and FAPAR is fitted, and the FAPAR of crops in the sample fields is inverted based on this linear regression model.

[0012] As a further technical solution, the step of obtaining the reflectance of each pixel in each multispectral image includes:

[0013] The average pixel DN value of the center band of the multispectral image under the reference standard reflectance is obtained, and the linear regression relationship between the pixel DN value and reflectance under different center bands is fitted piecewise based on the least squares method.

[0014] Based on the linear regression relationship between pixel DN values ​​and reflectance under different center bands, the reflectance of each pixel in the near-field multispectral image and the far-field multispectral image is obtained by solving the pixel-by-pixel method.

[0015] As a further technical solution, the steps for constructing a simplified spectral library include:

[0016] Define the crop canopy as the foreground, select the two central bands with the greatest difference between the foreground and background in the near-field multispectral image, and use them as the horizontal and vertical coordinates to construct a two-dimensional triangular feature space, and extract the pure pixel spectrum in the near-field multispectral image by category;

[0017] Based on K-means clustering and iterative endmember selection, foreground and background endmember spectra are adaptively extracted from the pure pixel spectra of different categories in near-field multispectral images to construct a simplified spectral library.

[0018] As a further technical solution, the mathematical representation of the crop-oriented pixel-by-pixel unmixing model is as follows:

[0019]

[0020]

[0021] in, Indicates pixel reflectivity; , They represent the first The abundance and albedo of land-like features, where p is the number of land-like endmembers and P is the re-collision probability of photons.

[0022] As a further technical solution, based on an exhaustive cross-pairing strategy, the steps of solving the full-coverage multispectral image pixel by pixel using a simplified spectral library and a crop-oriented pixel-by-pixel unmixing model include:

[0023] Based on a simplified spectral library, an exhaustive cross-pairing strategy is used to obtain several endmember spectral combinations. Each endmember spectral combination consists of a foreground endmember spectrum and a background endmember spectrum.

[0024] The pixel reflectance of the full-coverage multispectral image is combined with several endmember spectra and input into the crop-oriented pixel-by-pixel unmixing model to solve for the corresponding foreground abundance and background abundance. The endmember spectra combination that results in the lowest unmixing residual is selected as the best-fit endmember spectra combination for the pixel. The best-fit endmember spectra combination and the corresponding foreground abundance are output pixel by pixel as the foreground abundance unmixing result.

[0025] As a further technical solution, the fitting process of the linear regression model between foreground abundance and FAPAR includes:

[0026] FAPAR was obtained from field measurements of each crop sample plot;

[0027] The foreground abundance of the most suitable endmember spectral combination corresponding to the pixel-by-pixel correspondence of the full-coverage multispectral image was matched with the FAPAR of each crop plot, and a linear regression model of foreground abundance and FAPAR was constructed using a simple linear fitting method.

[0028] As a further technical solution, the process of retrieving FAPAR of crops in the sample field includes:

[0029] For the crop to be predicted in the crop sample field, the pixel position of the crop in the full-coverage multispectral image is obtained, and the foreground abundance of the corresponding pixel is used to solve the linear regression model of foreground abundance and FAPAR to obtain the FAPAR of the pixel as the FAPAR of the crop to be predicted.

[0030] According to another aspect of this specification, a crop FAPAR inversion system based on adaptive spectral unmixing is provided, comprising:

[0031] The reflectance calculation module is used to acquire close-up multispectral images and full-coverage multispectral images of crop sample fields, and to obtain the reflectance of each pixel in each multispectral image.

[0032] The simplified spectral library construction module is used to construct a simplified spectral library containing foreground and background endmember spectra based on the pixel-by-pixel reflectance of near-field multispectral images using an adaptive endmember extraction method.

[0033] A module for building a crop-oriented pixel-by-pixel unmixing model is used to build a crop-oriented pixel-by-pixel unmixing model based on a multilinear model.

[0034] The global unmixing module is used to solve the full-coverage multispectral image pixel by pixel based on an exhaustive cross-pairing strategy, using a simplified spectral library and a crop-oriented pixel-by-pixel unmixing model, and obtain the foreground abundance unmixing results pixel by pixel.

[0035] The FAPAR inversion module is used to unmix the measured FAPAR of crop sample fields with the pixel-by-pixel foreground abundance of full-coverage multispectral images, fit a linear regression model between foreground abundance and FAPAR, and invert the FAPAR of crops in the sample fields based on this linear regression model.

[0036] According to another aspect of this specification, an electronic device is provided, including a memory and a processor, the memory storing program instructions executed by the processor, the processor invoking the program instructions to perform a crop FAPAR inversion method based on adaptive spectral unmixing.

[0037] According to another aspect of this specification, a non-transitory computer-readable storage medium is provided, the non-transitory computer-readable storage medium storing computer instructions that cause the computer to execute a crop FAPAR inversion method based on adaptive spectral unmixing.

[0038] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention proposes an adaptive spectral unmixing strategy for multi-level mixed scenarios of crop growth, such as sorghum and rice, and solves for abundance based on this strategy. Compared with traditional FAPAR prediction using only a single spectral feature, this invention can better explain endmember variability and scenario complexity, and more accurately simulate the complex light propagation process in crop growth scenarios, quantifying the effective contribution of crop canopy to the spectrum, and enhancing the accuracy and interpretability of canopy abundance estimation in complex scenarios. Based on the abundance solution, FAPAR inversion is performed, realizing crop FAPAR inversion based on adaptive spectral unmixing, with an effective strategy and high inversion accuracy. The data used in this invention can be easily obtained through UAV remote sensing, avoiding the inefficiency and high labor costs of field measurements, and has great potential for applications in precision agriculture. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 A schematic diagram of the process for a crop FAPAR inversion method based on adaptive spectral unmixing provided in an embodiment of the present invention;

[0041] Figure 2 This is a schematic diagram illustrating the difference in reflectivity between the foreground and background in an embodiment of the present invention;

[0042] Figure 3 This is a schematic diagram illustrating the principle of the crop-oriented pixel-by-pixel demixing model in an embodiment of the present invention.

[0043] Figure 4 This is a multispectral UAV image covering all fields in an embodiment of the present invention;

[0044] Figure 5 This is a close-up image of the UAV in an embodiment of the present invention;

[0045] Figure 6 This is a schematic diagram of UAV image radiometric calibration in an embodiment of the present invention;

[0046] Figure 7 This is a schematic diagram of the R-NIR two-dimensional triangular feature space in an embodiment of the present invention;

[0047] Figure 8 This is an example diagram illustrating the extraction of a set of simple endmembers in an embodiment of the present invention;

[0048] Figure 9 This is a schematic diagram of FAPAR inversion in an embodiment of the present invention;

[0049] Figure 10 A schematic diagram of the structure of a crop FAPAR inversion system based on adaptive spectral unmixing provided in an embodiment of the present invention;

[0050] Figure 11 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0051] It should be noted that:

[0052] The terms “comprising” and “having”, and any variations thereof, in the specification, claims, and accompanying drawings of this invention are intended to cover a non-exclusive inclusion, such as a process, method, system, product, or apparatus that includes a series of steps or units, not necessarily limited to those explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0053] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices. The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily have to be performed in the described order. For example, some operations / steps can be decomposed, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.

[0054] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. In addition, the technical features of the various embodiments or individual embodiments provided by the present invention can be arbitrarily combined to form new technical solutions. Such combinations are not bound by the order of steps and / or structural composition patterns, but must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0055] like Figure 1 As shown, a crop FAPAR (photosynthetically active radiation absorption coefficient) inversion method based on adaptive spectral unmixing includes:

[0056] Step 1: Obtain close-up multispectral images and full-coverage multispectral images of the crop sample field, and obtain the reflectance of each pixel in each image;

[0057] Step 2: Based on the pixel-by-pixel reflectance of the near-field multispectral image, construct a simplified spectral library containing foreground and background endmember spectra using an adaptive endmember extraction method.

[0058] Step 3: Construct a pixel-by-pixel unmixing model for crops based on the multilinear model;

[0059] Step 4: Based on the exhaustive cross-pairing strategy, use a simplified spectral library and a crop-oriented pixel-by-pixel unmixing model to solve the full-coverage multispectral image pixel by pixel and obtain the foreground abundance unmixing results pixel by pixel.

[0060] Step 5: Based on the unmixing results of the measured FAPAR of the crop sample field and the foreground abundance of the full-coverage multispectral image pixel by pixel, fit a linear regression model of foreground abundance and FAPAR, and invert the FAPAR of the crop in the sample field based on the linear regression model.

[0061] Step 1, the step of obtaining the reflectance pixel by pixel in each image, includes:

[0062] The average pixel DN value of the center band of the multispectral image under the reference standard reflectance is obtained, and the linear regression relationship between the pixel DN value and reflectance under different center bands is fitted piecewise based on the least squares method.

[0063] Based on the linear regression relationship between pixel DN values ​​and reflectance under different center bands, the reflectance of each pixel in the near-field and far-field multispectral images of crop sample fields is obtained by solving the problem pixel by pixel.

[0064] Specifically, in step 1, the multispectral images of the crop sample field are acquired using a drone, with image acquisition starting at noon on a sunny day. Optionally, during the implementation process, the center bands of the multispectral images are 490, 520, 550, 570, 670, 680, 700, 720, 800, 850, 900, and 950 nm.

[0065] The reference standard reflectance was provided by eight calibration blankets with standard reflectances of 0.03, 0.06, 0.12, 0.24, 0.36, 0.48, 0.56 and 0.80 laid on the ground.

[0066] During data acquisition, firstly, the UAV is flown to a height of A meters to acquire near-field multispectral images with a spatial resolution of not less than 1 cm. At this point, the relationship between spatial resolution and flight altitude is shown in formula (1). The captured images need to cover part of the crop canopy and the entire calibration blanket to obtain organ-level pure pixels. Secondly, after acquiring the A-meter image, the UAV is immediately flown to a height of B meters (B > A) to acquire large-scale crop field images (distant multispectral images). The captured images need to cover the entire canopy and calibration blanket.

[0067] (1)

[0068] Where GSD is the ground resolution, which is 0.01m in this case; f is the focal length of the multispectral sensor in millimeters; and s is the physical size of a single pixel of the sensor in millimeters.

[0069] Specifically, the process of fitting a linear regression relationship between pixel DN values ​​and reflectance under different center bands using the least squares method is as follows: Radiometric calibration of the acquired UAV imagery is performed using a piecewise linear correction method with reference to a standard reflectance, including:

[0070] For each center band λ, the region of interest (ROI) is drawn on the calibration blanket to obtain its average pixel DN value. The standard reflectance of the calibration blanket and its average DN value in the λ band are substituted into formula (2), and the gain coefficient and offset coefficient in the λ band are obtained by least squares method.

[0071] (2)

[0072] In the formula, This represents the reflectivity of the center band λ. The pixel DN value represents the center band λ; Gain λ and Offset λ These represent the gain coefficient and offset coefficient of the center band λ, respectively.

[0073] By applying formula (2) globally and inputting the pixel DN value of the multispectral image, the reflectance can be calculated.

[0074] Step 2, the construction steps of the simplified spectral library include:

[0075] Step 2-1: Define the crop canopy as the foreground, select the two central bands with the greatest difference between the foreground and background in the near-field multispectral image, use them as the horizontal and vertical coordinates to construct a two-dimensional triangular feature space, and extract the pure pixel spectrum in the near-field multispectral image by category.

[0076] Step 2-2: Based on K-means clustering and iterative endmember selection, foreground and background endmember spectra are adaptively extracted from the pure pixel spectra of different categories in near-field multispectral images to construct a simplified spectral library.

[0077] Optionally, in step 2-1, considering the differences in reflectivity between foreground features and the background, such as... Figure 2 As shown, taking the red band reflectance (wavelength 670nm) as the X-axis and the near-infrared band reflectance (wavelength 850nm) as the Y-axis as an example, a two-dimensional R-NIR triangular feature space is constructed. Since the foreground mainly consists of the green organs of crops (leaves and ears), the spectral characteristics of vegetation are obvious, with low red band reflectance and high near-infrared reflectance, and it is located in the upper left of the R-NIR two-dimensional triangular feature space; while the background mainly consists of soil or turbid water, which has high red band reflectance and low near-infrared reflectance, and is therefore located in the lower right corner of the feature space; mixed pixels are located in other positions in the feature space.

[0078] Step 2-1 involves extracting the pure pixel spectra from the near-field multispectral images by category to obtain a spectral library. In step 2-2, a simplified spectral library containing foreground and background endmember spectra is adaptively constructed using a method based on K-means clustering and iterative endmember selection (IES). Specifically:

[0079] Pure pixel spectra exhibit both intra-class variability and repetitiveness. To improve the efficiency of subsequent model operation, after outputting a large number of foreground and background pure pixel spectra extracted in step 2-1, the K-means algorithm is used to cluster the pure pixels into categories, initially reducing redundancy. Then, the IES algorithm is used, taking the spectrum and its corresponding category (foreground / background) as input. The IES algorithm can capture intra-class spectral variability while quickly selecting the endmember spectra that best represent the land cover (endmembers refer to pure pixels containing information about a single land cover), obtaining the best subset of the original spectral library as a simplified spectral library.

[0080] In step 3, a pixel-wise unmixing model for crops is constructed based on the multilinear mixing (MLM) model. The model takes the foreground endmember spectrum, the background endmember spectrum combination, and the reflectance of the pixel to be unmixed as inputs and outputs the foreground abundance and background abundance for each pixel.

[0081] This model, based on a discrete Markov chain, describes the multiple effects of light within a scene, as follows:

[0082] First, assume that after entering the scene, light interacts with at least one type of ground feature, and the probability of interaction with a ground feature is proportional to its abundance. After each interaction with a ground feature, the photon continues to reflect within the scene with probability P, and escapes the scene with probability (1-P) and enters the multispectral sensor. When light is scattered on a ground feature, its intensity changes with the albedo w. i ∈[0,1] d The subscript i represents the end-member number of the ground feature. Taking the number of end-members p=2 as an example, for example... Figure 3 As shown:

[0083] The first reflected signal received by the sensor, i.e., the first term of the model, such as Figure 3 (a) can be represented as,

[0084] (3)

[0085] in, denoted as a first-order term of the pixel-wise unmixing model; a represents abundance, w represents albedo.

[0086] The signal obtained by the sensor after the light undergoes a secondary interaction within the scene, i.e., the quadratic term of the model, such as... Figure 3 (b) can be represented as,

[0087] (4)

[0088] in, represents the quadratic term of the pixel-by-pixel demixing model, indicating that after a photon interacts with ground objects twice, it escapes the scene and enters the multispectral sensor; ⊙ is the Hadamard product.

[0089] Similarly, reflectivity is obtained by weighted summation of the contributions from all possible paths, thus the pixel reflectivity is... It can be represented as:

[0090] (5)

[0091] in, .

[0092] For crop growth scenarios, assuming that there are only two types of land cover in the growth scenarios of crops such as sorghum and rice, i=[1,2], where i is a positive integer. Plant organs such as ears, leaves, and stems constitute the foreground, and soil or turbid water constitutes the background. In the crop growth scenario, the foreground abundance is defined as a1, and the albedo as w1; the background abundance is defined as a2, and the albedo as w2. Then, the mathematical representation of the pixel-by-pixel unmixing model for crops is:

[0093] (6)

[0094] Here, it is assumed that only pixels with an albedo of w exist. i When there is only one type of land feature, then:

[0095] (7)

[0096] In the formula, That is, the pure pixel reflectance, which is the endmember spectrum of this type of land cover, then:

[0097] (8)

[0098] In the formula, e i P represents the endmember spectrum of the i-th type of land cover; i Let P represent the photon re-collision probability of the i-th type of land cover. Therefore, theoretically, according to equation (8), as long as the re-collision probability P of a certain type of land cover is obtained... i This allows us to determine its endmember spectrum e. i With the corresponding albedo w i The relationship.

[0099] In practical applications of the model, the foreground endmember spectrum e1, the background endmember e2, and the reflectance x of the pixel to be unmixed are used as inputs. The model performs optimization using the fmincon function in Matlab according to equations (6) and (8), and outputs the foreground abundance a1 and the background abundance a2 pixel by pixel. Among them, the abundance satisfies the constraints of "summing to one" and "non-negative", and P < 1.

[0100] Step 4, based on an exhaustive cross-pairing strategy, involves solving the full-coverage multispectral image pixel-by-pixel using a simplified spectral library and a crop-oriented pixel-by-pixel unmixing model. The steps include:

[0101] Step 4-1: Based on the simplified spectral library, an exhaustive cross-pairing strategy is used to obtain several endmember spectral combinations. Each endmember spectral combination consists of a foreground endmember spectrum and a background endmember spectrum.

[0102] Step 4-2: Pixel by pixel, combine the pixel reflectance of the full-coverage multispectral image with several endmember spectra, input the pixel-by-pixel unmixing model for crops, solve for the corresponding foreground abundance and background abundance, select the endmember spectra combination that results in the lowest unmixing residual as the best-fit endmember spectra combination for the pixel, and output the best-fit endmember spectra combination and the corresponding foreground abundance pixel by pixel as the foreground abundance unmixing result.

[0103] Each endmember spectral combination consists of a foreground endmember spectrum and a background endmember spectrum.

[0104] Specifically, in step 4-1, assuming that the spectral library established in step 2 contains m foreground endmembers and n background endmembers, an exhaustive cross-pairing strategy is used to input one spectrum from each of the foreground and background endmembers into the model, then there are a total of m×n endmember spectral combinations.

[0105] In step 4-2, each endmember spectral combination is traversed, and along with the reflectance of the pixel, it is input into the crop-oriented pixel-by-pixel unmixing model constructed in step 3 for solving. Then, for a certain pixel x, m×n foreground abundances and m×n background abundances can be obtained.

[0106] Then, based on the model solution results, the unmixing residual corresponding to each endmember spectral combination is calculated one by one according to Equation (9). Optionally, the root mean square error (RMSE) is used for the unmixing residual to evaluate the best endmember spectral combination for each pixel.

[0107] (9)

[0108] in, The reconstructed spectrum is obtained by substituting the abundance a and endmember spectrum e obtained from the solution into the model; N is the number of spectral bands.

[0109] Ultimately, each pixel can obtain m×n RMSEs. The lowest RMSE is used to determine the best-fit endmember spectral combination used in the final pixel result, as well as the foreground abundance of the corresponding best-fit endmember spectral combination.

[0110] Step 5, the process of retrieving the FAPAR of crops in the sample field includes:

[0111] Step 5-1: Measure and obtain the FAPAR of each crop sample plot.

[0112] Step 5-2: Match the foreground abundance and FAPAR of the full-coverage multispectral image after pixel-by-pixel screening, and construct a linear regression model of foreground abundance and FAPAR using a simple linear fitting method.

[0113] Optionally, in step 5-1, the FAPAR of the field is the average FAPAR of the field. The steps for obtaining the measured FAPAR are as follows:

[0114] On the day of the drone flight, between 9:30 and 10:30 local time, FAPAR was collected plot by plot. The LI-190R and LI-191R rod-shaped FAPAR sensors were used to measure FAPAR data plot by plot in the field. The LI-190R was placed horizontally 50 cm above the top of the canopy to receive incident PAR from the top of the canopy. in and the top of the canopy reflect PAR out The LI-191R was placed horizontally 10cm from the bottom of the canopy, sensing light within a 1m range and collecting the transmitted canopy PAR received at the bottom of the canopy. ct PAR reflected from background elements such as soil or water rs Three sets of data were collected for each field, with LI-191R data taken parallel to the row direction (0°) and diagonally towards the row direction (45° and -45°). The three sets of data were substituted into the following formula and the average was calculated to obtain the average FAPAR for the field. At least five sets of data were collected to establish the relationship between abundance and FAPAR.

[0115] (10)

[0116] In the formula, This represents the field's FAPAR, which here refers to the average FAPAR of the field.

[0117] Optionally, step 5-2 essentially involves constructing the relationship between abundance and FAPAR, including:

[0118] First, collect "Foreground Abundance-FAPAR" matching data pairs. Each matching data pair must correspond to the same pixel / the same region, in the format of ( , (k=1,2,…,K, representing the number of sample regions), and then construct a linear regression model for the sample. The simple linear fitting method is used to construct a linear regression model of foreground abundance (independent variable) and FAPAR (dependent variable), as shown in formula (11):

[0119] (11)

[0120] Where A and B are the parameters of the linear regression model.

[0121] Preferably, in solving for the parameters of the linear regression model, A and B are calculated by minimizing the sum of squared residuals between the measured FAPAR and the model-predicted FAPAR, and the coefficient of determination (R²) is used. 2 To determine the fitting ability of a linear regression model, if R0 2 If the value is greater than 0.6, it proves that the model can be used to invert FAPAR, and the output linear regression model is the final linear regression model of foreground abundance and FAPAR.

[0122] (12)

[0123] Step 5, the process of retrieving the FAPAR of crops in the sample field includes:

[0124] For the crop to be predicted in the crop sample field, obtain the pixel position of the crop in the full-coverage multispectral image, and use the foreground abundance unmixing result of the corresponding pixel (obtained in steps 2 to 4) to solve the linear regression model of foreground abundance and FAPAR (obtained in step 5) to obtain the FAPAR of the pixel as the FAPAR of the crop to be predicted.

[0125] Preferably, the predicted FAPAR of the inverted crop can also be extended to unknown fields: for a field with an unknown FAPAR, the abundance of the corresponding image pixels of the field is obtained through steps two to four, and then substituted into the formula in step 5-2 to calculate the FAPAR at that location, thus realizing the prediction of the field's FAPAR.

[0126] As a specific implementation method, the present invention provides an example of crop FAPAR inversion based on adaptive spectral unmixing. The example consists of six steps: close-range and full-coverage image acquisition and preprocessing based on multispectral UAV, spectral library construction based on adaptive endmember extraction, pixel-by-pixel unmixing based on multilinear model, pixel-by-pixel screening of optimal combination, construction of abundance and FAPAR relationship based on measured FAPAR, and prediction of field FAPAR.

[0127] Step 1: Acquisition and preprocessing of close-range and full-coverage images based on multispectral UAVs.

[0128] This invention uses sorghum as an example for illustration. First, imagery from a drone is acquired. Figure 4 To obtain a multispectral drone image that covers the entire field from a distance of 200m (corresponding to a pixel resolution of 6cm, which is a full-coverage multispectral image). Figure 5 This is a close-up image of a 20m drone taken simultaneously (corresponding to a pixel resolution of 1cm, a close-up multispectral image). Next, radiometric calibration is performed on the image. Figure 6 A schematic diagram of the reflectance of a typical vegetation canopy before and after radiometric calibration of UAV images.

[0129] Step 2: Construct a spectral library based on adaptive image endmember extraction.

[0130] First, regarding close-up drone images of sorghum, see... Figure 7 (a) Construct the R-NIR two-dimensional triangular feature space. Figure 7 (b) shows the constructed R-NIR two-dimensional triangular feature space, where the horizontal axis represents the red band R and the vertical axis represents the near-infrared band NIR. A schematic diagram of clean pixel extraction is shown below. Figure 7 (c) Output 71,908 foreground clean pixel spectra and 79,281 background clean pixel spectra. Then, based on K-means clustering, set the number of clusters to 1,000 foreground endmember spectra and 1,000 background endmember spectra. Finally, input these 2,000 endmember spectra and their categories into an iterative endmember selection algorithm to filter endmembers and construct a simplified spectral library containing foreground and background endmember spectra. Figure 8 Here is an example of a set of simple endmembers extracted: Figure 8 (a) Corresponding to the foreground endmember, Figure 8 (b) Corresponding background endmember.

[0131] Step 3: Pixel-by-pixel demixing based on a multilinear model.

[0132] Pixel-by-pixel unmixing is performed on full-coverage UAV imagery. The spectral library extracted in step 2 contains 16 foreground endmembers and 7 background endmembers. One foreground endmember spectrum e1, one background endmember e2, and the reflectance x of the pixel to be unmixed are used as inputs. A crop-oriented pixel-by-pixel unmixing model is constructed using a multilinear model (MLM). This model is then used to solve the imagery, outputting the foreground abundance a1 and background abundance a2 pixel-by-pixel. The abundances satisfy the constraints of "summing to one" and "non-negativity," with P < 1.

[0133] Step 4: Filter the optimal combination pixel by pixel.

[0134] At this point, in step 3, each endmember spectral combination in the spectral library obtained in step 2 is traversed, with a total of 16×7 endmember combinations traversed for each pixel. Therefore, for a given pixel, 16×7 foreground abundances and 16×7 background abundances can be obtained. Then, the unmixing residual corresponding to each endmember spectral combination is calculated, i.e., the root mean square error (RMSE), to evaluate the most suitable endmember spectral combination for each pixel. Finally, 16×7 RMSEs are obtained for each pixel, and the endmember combination and corresponding foreground abundance used in the final result are determined based on the lowest RMSE.

[0135] Step 5: Construct the relationship between foreground abundance and FAPAR based on measured FAPAR.

[0136] A total of 233 sets of FAPAR were measured in the field at different time periods, and the relationship between foreground abundance and FAPAR was constructed using a simple linear fitting method.

[0137]

[0138] The abundance range is [0,1], and the FAPAR range is [0,1].

[0139] Step 6: Predict field FAPAR.

[0140] A multispectral image of a field with an unknown FAPAR is acquired. After calculations in steps 2-4, its abundance is found to be 0.335. Substituting this into the formula in step 5, the FAPAR for that field is calculated to be 0.407. Similarly, an image containing multiple fields can be obtained, and the FAPAR result can be predicted, as shown below. Figure 9 As shown: Figure 9 (a) is a multispectral true-color image of several fields taken by an unmanned aerial vehicle. Figure 9 (b) is a graph showing the FAPAR inversion results.

[0141] The implementation of the various embodiments of the present invention is based on programmed processing through a system with processor functionality. Therefore, in practical engineering, the technical solutions and functions of the various embodiments of the present invention are encapsulated into various modules. Based on this reality, and building upon the above embodiments, the embodiments of the present invention provide a crop FAPAR inversion system based on adaptive spectral unmixing, which is used to execute a crop FAPAR inversion method based on adaptive spectral unmixing from the above method embodiments.

[0142] See Figure 10 The system includes:

[0143] The system comprises the following modules: a reflectance calculation module, used to acquire near-field multispectral images and full-coverage multispectral images of crop sample fields, and obtain the reflectance of each pixel in each multispectral image; a simplified spectral library construction module, used to construct a simplified spectral library containing foreground and background endmember spectra based on the pixel-by-pixel reflectance of the near-field multispectral images using an adaptive endmember extraction method; a crop-oriented pixel-by-pixel unmixing model construction module, used to construct a crop-oriented pixel-by-pixel unmixing model based on a multilinear model; a global unmixing module, used to solve the full-coverage multispectral images pixel-by-pixel using the simplified spectral library and the crop-oriented pixel-by-pixel unmixing model based on an exhaustive cross-pairing strategy, and obtain the foreground abundance unmixing results pixel-by-pixel; and a FAPAR inversion module, used to fit a linear regression model of foreground abundance and FAPAR based on the measured FAPAR of the crop sample fields and the foreground abundance unmixing results of the full-coverage multispectral images, and invert the FAPAR of the crops in the sample fields based on this linear regression model.

[0144] It should be noted that the system embodiments provided by the present invention are used not only to implement the methods in the above method embodiments, but also to implement the methods in other method embodiments provided by the present invention. The only difference is that corresponding functional modules are set. The principle is basically the same as that of the above system embodiments provided by the present invention. As long as those skilled in the art can improve the modules in the above system embodiments by referring to the specific technical solutions in other method embodiments and combining technical features to obtain corresponding technical means and technical solutions composed of these technical means, on the basis of the above system embodiments, and on the premise of ensuring the practicality of the technical solutions, they can obtain corresponding system-like embodiments for implementing the methods in other method-like embodiments.

[0145] The method in this embodiment of the invention is implemented using an electronic device; therefore, it is necessary to introduce the relevant electronic device. For this purpose, embodiments of the present invention provide an electronic device, such as... Figure 11 As shown, the electronic device includes: at least one processor, a communication interface, at least one memory, and a communication bus, wherein the at least one processor, the communication interface, and the at least one memory communicate with each other via the communication bus. The at least one processor invokes logical instructions stored in the at least one memory to execute all or part of the steps of the methods provided in the foregoing method embodiments.

[0146] Furthermore, when the logical instructions in at least one of the aforementioned memories are implemented as software functional units and sold or used as independent products, they are stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, is 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 (a personal computer, server, or network device) to execute all or part of the steps of the methods described in the various method embodiments of the present invention. The aforementioned storage medium includes: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks—various media for storing program code.

[0147] The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, located in one place, or distributed across multiple network units. The purpose of this embodiment is achieved by selecting some or all of the modules according to actual needs. Those skilled in the art will understand and implement this without any inventive effort.

[0148] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0149] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0150] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0151] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0152] Based on the same technical concept as the foregoing embodiments, the present invention provides a non-transitory computer-readable storage medium storing computer instructions that cause the computer to execute a crop FAPAR inversion method based on adaptive spectral unmixing.

[0153] In summary, addressing the limitations of traditional VI inversion FAPAR retrieval methods due to the complex and dynamically changing structure of crop canopies, and the simultaneous interference from soil and other background elements caused by multiple light transmission and scattering, this invention proposes a crop FAPAR retrieval method based on adaptive spectral unmixing for multi-layered mixed growth scenarios of crops such as sorghum and rice. This method extracts clean pixels from UAV multispectral near-field images using the R-NIR feature space, establishes a variable endmember spectral library using k-means and IES algorithms, performs pixel-by-pixel unmixing based on a multiple linear unmixing model, and selects the optimal endmember combination to construct a FAPAR retrieval model with unmixed foreground abundance as a parameter. Compared to traditional methods, this invention effectively quantifies the contribution of the crop canopy (foreground) to the spectral signal, alleviates the saturation phenomenon and soil interference problems inherent in VI, and helps address the spectral variability, dynamic changes, and scene complexity commonly found in crop growth scenarios. This provides a more accurate, efficient, and generalized new FAPAR retrieval method for precision agriculture.

[0154] 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 or all of the technical features therein; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the technical solutions of the embodiments of the present invention.

Claims

1. A crop FAPAR inversion method based on adaptive spectral unmixing, characterized in that, include: Acquire close-up multispectral images and full-coverage multispectral images of crop sample fields, and obtain the reflectance of each pixel in each multispectral image; Based on the pixel-wise reflectance of near-field multispectral images, a simplified spectral library containing foreground and background endmember spectra is constructed using an adaptive endmember extraction method. A pixel-by-pixel unmixing model for crops is constructed based on a multilinear model; Based on the exhaustive cross-pairing strategy, a simplified spectral library and a crop-oriented pixel-by-pixel unmixing model are used to solve the full-coverage multispectral image pixel by pixel to obtain the foreground abundance unmixing results pixel by pixel. Based on the measured FAPAR of crop sample fields and the unmixing results of pixel-by-pixel foreground abundance from full-coverage multispectral images, a linear regression model of foreground abundance and FAPAR is fitted, and the FAPAR of crops in the sample fields is inverted based on this linear regression model.

2. The crop FAPAR inversion method based on adaptive spectral unmixing as described in claim 1, characterized in that, The steps for obtaining the pixel-by-pixel reflectance in each multispectral image include: The average pixel DN value of the center band of the multispectral image under the reference standard reflectance is obtained, and the linear regression relationship between the pixel DN value and reflectance under different center bands is fitted piecewise based on the least squares method. Based on the linear regression relationship between pixel DN values ​​and reflectance under different center bands, the reflectance of each pixel in the near-field multispectral image and the far-field multispectral image is obtained by solving the pixel-by-pixel method.

3. The crop FAPAR inversion method based on adaptive spectral unmixing as described in claim 1, characterized in that, The steps for constructing the simplified spectral library include: Define the crop canopy as the foreground, select the two central bands with the greatest difference between the foreground and background in the near-field multispectral image, and use them as the horizontal and vertical coordinates to construct a two-dimensional triangular feature space, and extract the pure pixel spectrum in the near-field multispectral image by category; Based on K-means clustering and iterative endmember selection, foreground and background endmember spectra are adaptively extracted from the pure pixel spectra of different categories in near-field multispectral images to construct a simplified spectral library.

4. The crop FAPAR inversion method based on adaptive spectral unmixing as described in claim 1, characterized in that, The mathematical representation of the crop-oriented pixel-by-pixel demixing model is as follows: ； ； in, Indicates pixel reflectivity; , , , They represent the first The abundance, albedo, reflectance, and photon re-collision probability of land features, where p is the number of land feature endmembers and P is the photon re-collision probability.

5. The crop FAPAR inversion method based on adaptive spectral unmixing as described in claim 1, characterized in that, Based on an exhaustive cross-pairing strategy, the steps for solving the full-coverage multispectral image pixel by pixel using a simplified spectral library and a crop-oriented pixel-by-pixel unmixing model include: Based on a simplified spectral library, an exhaustive cross-pairing strategy is used to obtain several endmember spectral combinations. Each endmember spectral combination consists of a foreground endmember spectrum and a background endmember spectrum. The pixel reflectance of the full-coverage multispectral image is combined with several endmember spectra and input into the crop-oriented pixel-by-pixel unmixing model to solve for the corresponding foreground abundance and background abundance. The endmember spectra combination that results in the lowest unmixing residual is selected as the best-fit endmember spectra combination for the pixel. The best-fit endmember spectra combination and the corresponding foreground abundance are output pixel by pixel as the foreground abundance unmixing result.

6. The crop FAPAR inversion method based on adaptive spectral unmixing as described in claim 5, characterized in that, The fitting process of the linear regression model between foreground abundance and FAPAR includes: FAPAR was obtained from actual measurements of each crop sample plot; The foreground abundance of the most suitable endmember spectral combination corresponding to the pixel-by-pixel correspondence of the full-coverage multispectral image was matched with the FAPAR of each crop plot, and a linear regression model of foreground abundance and FAPAR was constructed using a simple linear fitting method.

7. The crop FAPAR inversion method based on adaptive spectral unmixing as described in claim 1, characterized in that, The process of retrieving FAPAR from crops in the sample field includes: For the crop to be predicted in the crop sample field, the pixel position of the crop in the full-coverage multispectral image is obtained, and the foreground abundance of the corresponding pixel is used to solve the linear regression model of foreground abundance and FAPAR to obtain the FAPAR of the pixel as the FAPAR of the crop to be predicted.

8. A crop FAPAR inversion system based on adaptive spectral unmixing, characterized in that, include: The reflectance calculation module is used to acquire close-up multispectral images and full-coverage multispectral images of crop sample fields, and to obtain the reflectance of each pixel in each multispectral image. The simplified spectral library construction module is used to construct a simplified spectral library containing foreground and background endmember spectra based on the pixel-by-pixel reflectance of near-field multispectral images using an adaptive endmember extraction method. A module for building a crop-oriented pixel-by-pixel unmixing model is used to build a crop-oriented pixel-by-pixel unmixing model based on a multilinear model. The global unmixing module is used to solve the full-coverage multispectral image pixel by pixel based on an exhaustive cross-pairing strategy, using a simplified spectral library and a crop-oriented pixel-by-pixel unmixing model, and obtain the foreground abundance unmixing results pixel by pixel. The FAPAR inversion module is used to unmix the measured FAPAR of crop sample fields with the pixel-by-pixel foreground abundance of full-coverage multispectral images, fit a linear regression model between foreground abundance and FAPAR, and invert the FAPAR of crops in the sample fields based on this linear regression model.

9. An electronic device, characterized in that, The method includes a memory and a processor, the memory storing program instructions that are executed by the processor, the processor invoking the program instructions to perform the method according to any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium stores computer instructions that cause the computer to perform the method described in any one of claims 1 to 7.

Images