A method and system for estimating above-ground biomass of corn

Abstract

The application discloses a corn above-ground biomass estimation method and system, comprising the following steps: S1, obtaining multispectral and thermal infrared image data of corn plants in a corn field to be measured; S2, measuring the SPAD value of the corn canopy; S3, inputting the obtained multispectral and thermal infrared image data and the SPAD value of the corn canopy into an AGB estimation model to obtain an AGB estimation value; the system comprises a multispectral sensor, a thermal infrared sensor, a SPAD value acquisition module and an AGB estimation model; the application can make up for the problem of insufficient information of a single sensor through multi-source unmanned aerial vehicle sensor data fusion, is favorable for improving estimation accuracy, can reduce the influence of sensor deviation on AGB estimation by combining sensor data with the SPAD value, further improves estimation accuracy, and the CatBoost algorithm model has strong universality and high accuracy, thereby providing a brand-new and effective technical means for corn AGB estimation.

Description

Technical Field

[0001] This invention relates to the field of biomass monitoring technology, and more specifically to a method and system for estimating aboveground biomass in maize. Background Technology

[0002] Currently, sensors mounted on UAV platforms have demonstrated high accuracy in assessing phenotypic parameters such as crop AGB, leaf area index, and nitrogen content. However, different sensors acquire different data. For example, multispectral sensors quantitatively estimate crop growth parameters through the interactions between different wavelengths of light and plants, such as reflection, absorption, and transmission; thermal infrared sensors are unaffected by light and are mainly used to measure crop canopy temperature and crop responses to water stress.

[0003] However, when assessing crop traits, single-sensor data provides limited information and results in lower estimation accuracy. The fusion of multi-source sensor data can reflect crop growth information from different perspectives, effectively improving estimation accuracy. For example, in winter wheat yield estimation, fusing UAV multispectral, thermal infrared, and RGB data significantly improved estimation results compared to using only a single sensor. In potato AGB estimation, fusing hyperspectral and RGB data yielded better results than using only hyperspectral data. Furthermore, incorporating remote sensing-aided information, such as crop nitrogen content, crop water content, and crop height, into UAV remote sensing data can further improve the estimation accuracy of crop growth parameters. For instance, incorporating crop height information into hyperspectral data further improved the estimation accuracy of wheat AGB. SPAD values, while reflecting crop growth status well, can serve as an effective parameter for crop phenotypic estimation; however, current research on combining multi-source UAV sensor data with SPAD values ​​is limited.

[0004] In recent years, with the rapid development of computer science and artificial intelligence, machine learning has been widely used in crop phenotyping research. For example, when building crop trait evaluation models, PLSR, MLR and SVR have shown high prediction accuracy and robustness. However, most existing studies use single learner models. Single learner models have limited performance, high requirements for datasets, and poor generalization ability when applied to small datasets or fluctuating datasets.

[0005] Ensemble learning is a type of machine learning method that trains several weak learners and combines them using certain strategies to obtain a more comprehensive strong learner to complete the modeling task. Currently, ensemble learning is mainly divided into Bagging framework and Boosting framework. RFR and CatBoost are ensemble learning methods based on Bagging and Boosting, respectively. RFR has been successfully applied to crop trait assessment and has shown better estimation results. The CatBoost algorithm has shown strong usability in hydrological and structural engineering research, but its application in crop phenotypic assessment research is relatively limited.

[0006] Therefore, how to provide a method and system for estimating maize aboveground biomass that integrates multi-source UAV sensor data with SPAD values ​​and combines it with the CatBoost method to achieve maize AGB estimation is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0007] In view of this, the present invention provides a method and system for estimating aboveground biomass of maize to solve the problems mentioned in the background art.

[0008] To achieve the above objectives, the present invention adopts the following technical solution:

[0009] A method for estimating aboveground biomass in maize includes:

[0010] S1. Acquire multispectral and thermal infrared image data of corn plants in the cornfield to be tested;

[0011] S2. Determine the SPAD value of the maize canopy;

[0012] S3. Input the obtained multispectral and thermal infrared image data and the SPAD value of the maize canopy into the AGB estimation model to obtain the AGB estimated value.

[0013] Preferably, the methods for obtaining the AGB estimation model include:

[0014] S31. Obtain multispectral and thermal infrared image data of maize plants in each block of the experimental base and preprocess them to extract the spectral reflectance and canopy temperature of the maize canopy.

[0015] S32. Simultaneously measure the SPAD and AGB values ​​of maize plants in each block;

[0016] S33. Extract feature datasets based on spectral reflectance, canopy temperature, and SPAD value;

[0017] S34. Construct the CatBoost model;

[0018] S35. Divide the feature dataset into training and testing sets, and train and test the CatBoost model using cross-validation.

[0019] Preferably, the preprocessing in S31 includes: image stitching, radiometric correction, and soil pixel separation.

[0020] Preferably, the specific content of soil pixel separation includes: determining the NDVI thresholds corresponding to corn and soil pixels through multispectral and thermal infrared image data; obtaining the corresponding corn pixel mask through masking based on the NDVI thresholds; applying the mask to the image data to remove soil pixels; and finally extracting the spectral reflectance and canopy temperature of the corn canopy in each block using the region of interest.

[0021] Preferably, the method for obtaining the SPAD value in S32 is as follows: n corn plants are randomly selected from each block, and the upper, middle and lower parts of the top leaves of the plant canopy are measured using a chlorophyll meter. The average value is taken as the SPAD value of the corn plant canopy, and then the average value of the SPAD values ​​of the n plants is calculated as the SPAD value of the corn canopy in each block.

[0022] Preferably, the method for obtaining the AGB value in S32 is as follows: n uniformly growing corn plants are randomly selected from each block as samples, weighed, and then placed in a forced-air drying oven to dry until the mass is constant. The dry weight of the samples is weighed, and the AGB per unit area of ​​cornfield is calculated based on the dry weight of the samples and the population density as the measured value of AGB.

[0023] Preferably, the method for extracting the feature dataset in S33 is to calculate the vegetation index and crop water stress index based on spectral reflectance and canopy temperature, and then fuse them with the SPAD value as the feature dataset.

[0024] The vegetation index and crop water stress index are as follows:

[0025] NDVI = (NIR - R) / (NIR + R)

[0026] GNDVI = (NIR - G) / (NIR + G)

[0027] RNDVI = (EDGE - R) / (EDGE + R)

[0028] SAVI=(1+L)*(NIR-R) / (NIR+R+L), L=0.5

[0029] EVI=2.5*(NIR-R) / (NIR+6*R-7.5*B+1)

[0030] RVI = NIR / R

[0031] TVI = 60 * (NIR - G) - 100 (RG)

[0032] CWSI = [T canopy -T min ] / [T max -T min ]

[0033] Where R represents the red band, G the green band, B the blue band, EDGE the red-edge band, NIR the near-infrared band, and T... canopy The average temperature of the corn canopy in the block is T. min T is the minimum canopy temperature. max This represents the maximum canopy temperature.

[0034] Preferably, the specific content of the CatBoost model is as follows: A sample set is constructed using multi-source UAV data and SPAD values ​​as input features, and measured biomass as the output feature. The CatBoost algorithm is called through the CatBoost library for modeling. CatBoost consists of multiple base models. For training each base model M, the original samples are randomly sorted to generate a [1, n] permutation σ. Based on σ, n different models M1, M2, ..., M are initialized. n Where σ represents the reordering of the original samples, which include measured biomass data (AGB value, spectral reflectance, canopy temperature, and SPAD value), and Mi is the base model trained using the first i samples in the random permutation of σ. In each iteration, the base model M is used to... j-1 The unbiased gradient estimate of the j-th sample is obtained, and the unbiased gradient estimates of all samples are obtained. Based on the unbiased gradient estimate obtained from the base model, the weights of the next base model are optimized to obtain a new unbiased gradient estimate. The training of all base models is completed iteratively, and the prediction results of all base models are weighted and averaged to obtain the final prediction result of CatBoost.

[0035] Preferably, S35 specifically includes: randomly and evenly dividing the data into 5 groups, of which 4 groups are used for model training and 1 group is used for validation. This process is repeated iteratively 5 times, and finally the average of the coefficient of determination R2 and the relative root mean square error rRMSE of the 5 validation results is used as the accuracy evaluation index.

[0036]

[0037]

[0038] Where, x i The measured AGB value for corn is y. i y is the predicted value of maize AGB, y is the average value of the measured maize AGB, and n is the number of samples in the test set.

[0039] A maize aboveground biomass estimation system includes a multispectral sensor, a thermal infrared sensor, a SPAD value acquisition module, and an AGB estimation model;

[0040] Multispectral and thermal infrared sensors are used to acquire multispectral and thermal infrared image data of corn plants in the cornfield under test.

[0041] The SPAD value acquisition module is used to determine the SPAD value of the maize canopy.

[0042] The AGB estimation model is used to output AGB estimates based on the input multispectral and thermal infrared image data and the SPAD values ​​of the maize canopy.

[0043] As can be seen from the above technical solution, compared with the prior art, the present invention discloses a method and system for estimating aboveground biomass (AGB) of maize. By fusing multi-source UAV sensor data with SPAD values ​​and combining them with a novel machine learning algorithm, CatBoost, the AGB of maize is estimated. The fusion of multi-source UAV sensor data can compensate for the lack of information from a single sensor, which is beneficial to improving the estimation accuracy. Combining sensor data with SPAD values ​​can reduce the impact of sensor bias on AGB estimation, further improving the estimation accuracy. The CatBoost algorithm model has strong universality and high accuracy, providing a new and effective technical means for estimating AGB of maize. It can meet the needs of low-cost, large-scale, and repeated observation of AGB monitoring, and help agricultural decision-makers to make timely management measures and meet the needs of precision agriculture. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0045] Figure 1 The attached figure is a schematic diagram of the aboveground biomass estimation method for maize provided by the present invention;

[0046] Figure 2 The attached figure is a schematic diagram illustrating the distribution of SPAD values ​​in maize as provided by this invention.

[0047] Figure 3 The attached figure is a schematic diagram of the AGB distribution measurement provided by the present invention;

[0048] Figure 4 The attached figure is a schematic diagram of the CatBoost model implementation process provided by this invention;

[0049] Figure 5The attached figure is a schematic diagram of the five-fold cross-validation process provided by the present invention;

[0050] Figure 6 The attached figure is a schematic diagram of the statistical list of AGB estimation accuracy for different reproductive stages, different data combinations, and different algorithms provided in Embodiment 1 of the present invention;

[0051] Figure 7 The attached figure is a schematic diagram of the AGB estimation accuracy of SVR, RFR and CatBoost algorithms under different reproductive stages and different data combinations provided in Embodiment 1 of the present invention;

[0052] Figure 8 The attached figure is a scatter plot of AGB estimation based on multi-sensor data and SPAD values ​​for the jointing stage, trumpet stage, and tasseling stage provided in Embodiment 1 of the present invention.

[0053] Figure 9 The attached figure shows the AGB estimation R for different reproductive stages and different algorithms provided in Embodiment 2 of the present invention. 2 Comparison diagram;

[0054] Figure 10 The attached figure is a schematic diagram comparing the AGB estimation rRMSE of different reproductive stages and different algorithms provided in Embodiment 2 of the present invention;

[0055] Figure 11 The attached figure is a statistical t-test diagram of the AGB estimated value and the measured value of the fusion of multi-source sensor data and SPAD value during the jointing stage and CatBoost provided in Embodiment 3 of the present invention.

[0056] Figure 12 The attached figure is a statistical T-test diagram of the AGB estimated value and the measured value of the fusion of multi-source sensor data and SPAD value during the trumpet-mouth period provided in Embodiment 3 of the present invention and CatBoost.

[0057] Figure 13 The attached figure is a schematic diagram of the statistical T-test of the AGB estimated value and the measured value of the fusion of multi-source sensor data and SPAD value during the male ejaculation period and CatBoost, provided in Embodiment 3 of the present invention. Detailed Implementation

[0058] 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, and 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.

[0059] This invention discloses a method for estimating aboveground biomass in maize, such as... Figure 1,include:

[0060] S1. Acquire multispectral and thermal infrared image data of corn plants in the cornfield to be tested;

[0061] S2. Determine the SPAD value of the maize canopy;

[0062] S3. Input the obtained multispectral and thermal infrared image data and the SPAD value of the maize canopy into the AGB estimation model to obtain the AGB estimated value.

[0063] In this embodiment, a DJM210 UAV is used as a remote sensing platform, equipped with a multispectral sensor RedEdge MX and a thermal infrared sensor ZENMUSE XT for multi-sensor data acquisition.

[0064] To further implement the above technical solution, the methods for obtaining the AGB estimation model include:

[0065] S31. Obtain multispectral and thermal infrared image data of maize plants in each block of the experimental base and preprocess them to extract the spectral reflectance and canopy temperature of the maize canopy.

[0066] S32. Simultaneously measure the SPAD and AGB values ​​of maize plants in each block;

[0067] S33. Extract feature datasets based on spectral reflectance, canopy temperature, and SPAD value;

[0068] S34. Construct the CatBoost model;

[0069] S35. Divide the feature dataset into training and testing sets, and train and test the CatBoost model using cross-validation.

[0070] In this embodiment, the experimental site is the Xinxiang Comprehensive Experimental Base of the Chinese Academy of Agricultural Sciences in Xinxiang County, Henan Province, China (113°45'42"E, 35°08'05"N). Figure 1 An experiment was conducted, selecting 10 maize varieties and setting up 4 gradient fertilizer treatments (N0: 0 kg / hm²). 2 N1: 80 kg / hm 2 N2: 120 kg / hm 2 N3: 160 kg / hm 2 Each maize variety was replicated three times in each fertilizer treatment, for a total of 120 plots, with each plot measuring 8m². 2To ensure the reliability of AGB data for the corn plantation, after the corn seedlings emerge, transplant seedlings to fill in any gaps and remove seedlings where there are too many. Irrigation management is carried out according to local high-yield field standards, and pests, diseases, and weeds are controlled. Flight missions are conducted during the corn jointing, whooping, and tasseling stages on sunny, cloudless days with good lighting conditions to collect multispectral and thermal infrared image data of the corn plants. Whiteboard data is collected before and after each flight for subsequent radiometric correction.

[0071] To further implement the above technical solution, the preprocessing in S31 includes: image stitching, radiometric correction and soil pixel separation.

[0072] To further implement the above technical solution, the specific content of soil pixel separation includes: determining the NDVI thresholds corresponding to corn and soil pixels through multispectral and thermal infrared image data; obtaining the corresponding corn pixel mask through masking based on the NDVI thresholds; applying the mask to the image data to remove soil pixels; and finally extracting the spectral reflectance and canopy temperature of the corn canopy in each block using the region of interest.

[0073] To further implement the above technical solution, the method for obtaining the SPAD value in S32 is as follows: n corn plants are randomly selected from each block, and the upper, middle and lower parts of the top leaves of the plant canopy are measured using a chlorophyll meter. The average value is taken as the SPAD value of the corn plant canopy, and then the average value of the SPAD values ​​of the n plants is calculated as the SPAD value of the corn canopy in each block.

[0074] To further implement the above technical solution, the method for obtaining the AGB value in S32 is as follows: n uniformly growing corn plants are randomly selected from each block as samples, weighed, and then placed in a forced-air drying oven to dry until the mass is constant. The dry weight of the samples is weighed, and the AGB per unit area of ​​cornfield is calculated based on the dry weight of the samples and the population density as the measured value of AGB.

[0075] In this embodiment, three maize plants are randomly selected from each block. A SPAD-502 chlorophyll meter is used to measure the upper, middle, and lower parts of the leaves at the top of the plant canopy. The average value is taken as the SPAD value of that plant's canopy. The average of the three plant SPAD values ​​is then calculated as the SPAD value of the maize canopy in that block. Figure 2 Three uniformly growing corn plants were randomly selected from each plot as samples. After weighing, they were placed in a drying oven and dried until their mass became constant. The dry weight of the samples was then measured. Finally, the average annual density (AGB) per unit area of ​​cornfield was calculated based on the dry weight of the samples and the population density. Figure 3 As can be seen from the figure, excessive fertilizer treatment does not lead to higher AGB. The high AGB plots in the three growth stages are mainly concentrated under the N1 and N2 treatments. The AGB of the N3 treatment, which further increases fertilizer, actually decreases.

[0076] To further implement the above technical solution, the method for extracting the feature dataset in S33 is to calculate the vegetation index and crop water stress index based on spectral reflectance and canopy temperature, and then fuse them with the SPAD value as the feature dataset.

[0077] The vegetation index and crop water stress index are as follows:

[0078] NDVI = (NIR - R) / (NIR + R)

[0079] GNDVI = (NIR - G) / (NIR + G)

[0080] RNDVI = (EDGE - R) / (EDGE + R)

[0081] SAVI=(1+L)*(NIR-R) / (NIR+R+L), L=0.5

[0082] EVI=2.5*(NIR-R) / (NIR+6*R-7.5*B+1)

[0083] RVI = NIR / R

[0084] TVI = 60 * (NIR - G) - 100 (RG)

[0085] CWSI = [T canopy -T min ] / [T max -T min ]

[0086] Where R represents the red band, G the green band, B the blue band, EDGE the red-edge band, NIR the near-infrared band, and T... canopy The average temperature of the corn canopy in the block is T. min T is the minimum canopy temperature. max This represents the maximum canopy temperature.

[0087] To further implement the above technical solutions, such as Figure 4 The specific content of the CatBoost model is as follows: A sample set is constructed using multi-source UAV data and SPAD values ​​as input features, and measured biomass as the output feature. The CatBoost algorithm is called through the CatBoost library for modeling. CatBoost consists of multiple base models. For training each base model M, the original samples are randomly sorted to generate a [1, n] permutation σ. Based on σ, n different models M1, M2, ..., Mn are initialized. nWhere σ represents the reordering of the original samples, which include measured biomass data (AGB value, spectral reflectance, canopy temperature, and SPAD value), and Mi is the base model trained using the first i samples in the random permutation of σ. In each iteration, the base model M is used to... j-1 The unbiased gradient estimate of the j-th sample is obtained, and the unbiased gradient estimates of all samples are obtained. Based on the unbiased gradient estimate obtained from the base model, the weights of the next base model are optimized to obtain a new unbiased gradient estimate. The training of all base models is completed iteratively, and the prediction results of all base models are weighted and averaged to obtain the final prediction result of CatBoost.

[0088] To further implement the above technical solutions, such as Figure 5 The specific content of S35 includes: randomly and evenly dividing the data into 5 groups, of which 4 groups are used for model training and 1 group is used for validation. This process is repeated iteratively 5 times. Finally, the average of the coefficient of determination R2 and the relative root mean square error rRMSE of the 5 validation results is used as the accuracy evaluation index.

[0089]

[0090]

[0091] Where, x i The measured AGB value for corn is y. i y is the predicted value of maize AGB, y is the average value of the measured maize AGB, and n is the number of samples in the test set.

[0092] A maize aboveground biomass estimation system includes a multispectral sensor, a thermal infrared sensor, a SPAD value acquisition module, and an AGB estimation model;

[0093] Multispectral and thermal infrared sensors are used to acquire multispectral and thermal infrared image data of corn plants in the cornfield under test.

[0094] The SPAD value acquisition module is used to determine the SPAD value of the maize canopy.

[0095] The AGB estimation model is used to output AGB estimates based on the input multispectral and thermal infrared image data and the SPAD values ​​of the maize canopy.

[0096] Example 1

[0097] Multi-source sensor data and SPAD values ​​are fused for AGB estimation:

[0098] Using various data combination schemes from the jointing stage to the estrus stage, including SPAD, MS, TIR, SPAD+MS, MS+TIR, SPAD+TIR, and SPAD+MS+TIR, AGB estimation was performed based on SVR, RFR, and CatBoost algorithms. Figure 6 .

[0099] Compared to single-sensor data (MS or TIR), multi-source sensor data fusion (MS+TIR) improves the estimation accuracy of all three algorithms to varying degrees, particularly in the RV of SVR from the jointing stage to the tasseling stage. 2 The values ​​ranged from 0.450 to 0.661, and the rRMSE ranged from 31.67% to 22.55%. The RFR... 2 The R² values ​​were 0.464–0.691, and the rRMSE was 31.96%–21.57%. For CatBoost, the R² values ​​were 0.510–0.698, and the rRMSE was 30.53%–21.37%. Figure 7 .

[0100] Adding SPAD data (SPAD+MS, SPAD+TIR, and SPAD+MS+TIR) to the sensor data further improves estimation accuracy. Among them, the estimation accuracy of multi-source sensor data fusion with SPAD values ​​(SPAD+MS+TIR) is the highest, with the R value of SVR from the jointing stage to the tasseling stage being the highest. 2 The values ​​ranged from 0.461 to 0.703, and rRMSE ranged from 31.0% to 20.93%; the RFR... 2 The values ​​ranged from 0.510 to 0.719, and the rRMSE ranged from 30.54% to 20.42%; CatBoost's R... 2 The values ​​ranged from 0.547 to 0.735, and the rRMSE ranged from 29.17% to 19.97%. Figure 8 .

[0101] The vegetation index in MS data is composed of spectral reflectance from different bands and is an important parameter for crop growth analysis. NDVI quantifies vegetation by calculating the difference between NIR and R bands, reflecting the health of vegetation. SAVI reduces the impact of soil on crop monitoring and can still effectively detect vegetation in areas with high bare soil and low vegetation. GNDVI modifies NDVI, replacing the red band with the green band, making it less susceptible to spectral saturation in high vegetation areas and a good indicator of crop growth status. The CT and CWSI in TIR data are the result of the combined effects of crop genetic characteristics and environmental conditions. They are closely related not only to the functional period of leaves and transpiration rate but also to the growth of crop stem organs and starch synthesis, and have a strong correlation with AGB. There is heterogeneity between MS and TIR data. Combining the two can form complementary information, which is beneficial to improving the accuracy of AGB estimation.

[0102] SPAD values, as a health indicator of plant growth and development and leaf nitrogen content, effectively reflect crop nutritional status and senescence processes, thus helping to reduce sensor bias. Incorporating SPAD values ​​(SPAD+MS, SPAD+TIR, and SPAD+MS+TIR) into sensor data further improves AGB prediction accuracy compared to using sensor data alone. This is because SPAD values ​​are closely related to the strength of crop photosynthesis, which can explain changes in crop AGB and yield; therefore, adding SPAD data helps improve estimation accuracy. Compared to other remote sensing auxiliary data such as crop moisture content and soil factors, SPAD values ​​are easily obtained in field experiments using the SPAD-502 chlorophyll meter. Even those without agronomic knowledge can easily operate it, offering advantages such as low cost, simple operation, and non-destructive sampling, making it suitable for a wide range of applications in crop parameter inversion.

[0103] Example 2

[0104] Comparison of AGB estimations at different reproductive stages and using different algorithms:

[0105] Statistical analysis of R2 and rRMSE for SVR, RFR, and CatBoost under different reproductive stages and different data combinations, such as... Figure 9 and Figure 10 The results showed that the estimation accuracy of the three algorithms gradually increased with the progression of the reproductive period (the mean R² of SVR from the jointing stage to the estrus stage increased from 0.399 to 0.629, and the rRMSE decreased from 33.35% to 23.46%; the R² of RFR increased from 0.399 to 0.629). 2 The mean increased from 0.426 to 0.661, while rRMSE decreased from 33.03% to 22.49%; CatBoost: R 2 The mean value increased from 0.459 to 0.679, and the rRMSE decreased from 32.04% to 21.94%, with the accuracy of estimating the ejaculation period reaching its highest level.

[0106] Compared to a single learner (SVR), the R values ​​of the two ensemble learning methods (RFR and CatBoost) in the box plots are... 2 CatBoost has higher accuracy and lower rRMSE, resulting in better estimation performance. Among the two ensemble learning methods, CatBoost's estimation accuracy is significantly higher than that of RFR. The dataset contains actual AGB data of maize under different growth stages, fertilizer treatments, and varieties. Compared with the three machine learning algorithms, CatBoost shows the best estimation accuracy in various feature combination schemes and can better invert AGB under various conditions.

[0107] Example 3

[0108] Multi-source data fusion combined with CatBoost AGB estimation:

[0109] To verify the AGB estimation performance of multi-source sensor data (MS+TIR) fusion with SPAD values ​​combined with CatBoost, a T-test was used to evaluate whether there was a significant difference between the estimated and measured AGB values. Figures 11-13 The results showed that at the 0.05 level, there was no significant difference between the estimated and measured values ​​of AGB for the three reproductive stages, indicating that the method has scientific statistical significance for AGB estimation and proving the effectiveness of the fusion of multi-source sensor data (MS+TIR) and SPAD values ​​combined with CatBoost for AGB estimation.

[0110] Figures 11-13 The spatial distribution of AGB estimates is shown in the figure. The changes in maize AGB under different fertilizer treatments are clearly visible in the graph. Maize AGB initially increases and then decreases with increasing fertilizer application, which is consistent with the measured maize AGB. Figure 4 This indicates that fusing multi-source sensor data (MS+TIR) with SPAD values ​​and combining it with the CatBoost algorithm can yield reliable AGB estimates.

[0111] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0112] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for estimating aboveground biomass in maize, characterized in that, include: S1. Acquire multispectral and thermal infrared image data of corn plants in the cornfield to be tested; S2. Determine the SPAD value of the maize canopy; S3. Input the obtained multispectral and thermal infrared image data and the SPAD value of the maize canopy into the AGB estimation model to obtain the AGB estimated value; Methods for obtaining the AGB estimation model include: S31. Obtain multispectral and thermal infrared image data of maize plants in each block of the experimental base and preprocess them to extract the spectral reflectance and canopy temperature of the maize canopy. S32. Simultaneously measure the SPAD and AGB values ​​of maize plants in each block; S33. Extract feature datasets based on spectral reflectance, canopy temperature, and SPAD value; S34. Construct the CatBoost model; S35. Divide the feature dataset into training and testing sets, and train and test the CatBoost model using cross-validation; The method for extracting the feature dataset in S33 is to calculate the vegetation index and crop water stress index based on spectral reflectance and canopy temperature, and then fuse them with the SPAD value as the feature dataset.

2. The method for estimating aboveground biomass of maize according to claim 1, characterized in that, Preprocessing in S31 includes: image stitching, radiometric correction, and soil pixel separation.

3. The method for estimating aboveground biomass of maize according to claim 2, characterized in that, The specific steps of soil pixel separation include: determining the NDVI thresholds corresponding to maize and soil pixels using multispectral and thermal infrared image data; obtaining the corresponding maize pixel mask through masking based on the NDVI thresholds; applying the mask to the image data to remove soil pixels; and finally extracting the spectral reflectance and canopy temperature of the maize canopy in each block using the region of interest.

4. The method for estimating aboveground biomass of maize according to claim 1, characterized in that, The method for obtaining the SPAD value in S32 is as follows: n corn plants are randomly selected from each block, and the upper, middle and lower parts of the top leaves of the plant canopy are measured using a chlorophyll meter. The average value is taken as the SPAD value of the corn plant canopy. Then the average value of the SPAD values ​​of the n plants is calculated as the SPAD value of the corn canopy in each block.

5. The method for estimating aboveground biomass of maize according to claim 1, characterized in that, The method for obtaining the AGB value in S32 is as follows: n uniformly growing corn plants are randomly selected from each block as samples, weighed, and then placed in a forced-air drying oven to dry until the mass is constant. The dry weight of the samples is weighed, and the AGB per unit area of ​​cornfield is calculated based on the dry weight of the samples and the population density as the measured value of AGB.

6. The method for estimating aboveground biomass of maize according to claim 1, characterized in that, The vegetation index and crop water stress index are as follows: NDVI = (NIR - R) / (NIR + R) GNDVI = (NIR - G) / (NIR + G) RNDVI = (EDGE - R) / (EDGE + R) SAVI=(1+L)*(NIR-R) / (NIR+R+L), L=0.5 EVI=2.5*(NIR-R) / (NIR+6*R-7.5*B+1) RVI = NIR / R TVI = 60 * (NIR - G) - 100 (RG) CWSI=[T canopy -T min ] / [T max -T min ] Where R represents the red band, G the green band, B the blue band, EDGE the red-edge band, NIR the near-infrared band, and T... canopy The average temperature of the corn canopy in the block is T. min T is the minimum canopy temperature. max This represents the maximum canopy temperature.

7. The method for estimating aboveground biomass of maize according to claim 1, characterized in that, The CatBoost model works as follows: It constructs a sample set using multi-source UAV data and SPAD values ​​as input features, and measured biomass as the output feature. The CatBoost algorithm is called through the CatBoost library for modeling. CatBoost consists of multiple base models. For training each base model M, the original samples are randomly sorted to generate a [1, n] permutation σ. Based on σ, n distinct models M1, M2, ..., Mn are initialized. n Where σ represents the reordering of the original samples, which include measured biomass data (AGB value, spectral reflectance, canopy temperature, and SPAD value), and Mi is the base model trained using the first i samples in the random permutation of σ. In each iteration, the base model M is used to... j-1 The unbiased gradient estimate of the j-th sample is obtained, and the unbiased gradient estimates of all samples are obtained. Based on the unbiased gradient estimate obtained from the base model, the weights of the next base model are optimized to obtain a new unbiased gradient estimate. The training of all base models is completed iteratively, and the prediction results of all base models are weighted and averaged to obtain the final prediction result of CatBoost.

8. The method for estimating aboveground biomass of maize according to claim 7, characterized in that, The specific content of S35 includes: randomly and evenly dividing the data into 5 groups, of which 4 groups are used for model training and 1 group is used for validation. This process is repeated iteratively 5 times. Finally, the average of the coefficient of determination R2 and the relative root mean square error rRMSE of the 5 validation results is used as the accuracy evaluation index. in, This is the measured AGB value for corn. This is the AGB forecast value for corn. This represents the average of the measured AGB values ​​for corn. This represents the number of samples in the test set.

9. A maize aboveground biomass estimation system, based on the maize aboveground biomass estimation method according to any one of claims 1-8, characterized in that, Includes a multispectral sensor, a thermal infrared sensor, a SPAD value acquisition module, and an AGB estimation model; Multispectral and thermal infrared sensors are used to acquire multispectral and thermal infrared image data of corn plants in the cornfield under test. The SPAD value acquisition module is used to determine the SPAD value of the maize canopy. The AGB estimation model is used to output AGB estimates based on the input multispectral and thermal infrared image data and the SPAD values ​​of the maize canopy.

Images