Canopy morphology 3D feature analysis method based on lightweight unmanned aerial vehicle and point cloud singular value decomposition

Abstract

The invention discloses a canopy morphology 3D feature analysis method based on a lightweight unmanned aerial vehicle and point cloud singular value decomposition, and belongs to the field of plant morphology analysis. The method comprises the following steps: carrying out high-precision three-dimensional reconstruction on a soybean canopy structure by adopting an unmanned aerial vehicle to carry an RGB lens and combining with an SfM-MVS algorithm, introducing a singular value decomposition (SVD) method for a point cloud coordinate matrix based on a label method, and establishing nine canopy morphological 3D features under multiple scales as new explainable features of a crop plot level. Results show that the global canopy 3D features have significant correlation (plt; plt) with the yield, lodging, canopy height and vegetation index of soybeans; = 0.05). More detailed structure information of the canopy surface can be extracted based on 3D feature extraction of a local range, and characterization of differences of more growth periods of soybeans is facilitated.

Description

Technical Field

[0001] The present invention belongs to the field of plant morphological analysis, and specifically relates to a canopy morphological 3D feature analysis method based on a lightweight unmanned aerial vehicle and point cloud singular value decomposition. Background Art

[0002] Hundreds or even thousands of breeding materials usually need to be surveyed during crop breeding. Traditional field surveys are time-consuming, labor-intensive and costly, and the surveyed traits are limited. With its high throughput and adaptability, drone platforms have been increasingly used in the study of soybean (Glycine max (L.) Merr.) breeding phenotypes. The rapid development of drone proximity platforms and information technology can effectively monitor crop phenotypes and assist breeding decisions (Zhou et al., 2022; Ren et al., 2023). By equipping with RGB, multispectral, hyperspectral and lidar sensors, phenotypic characteristics such as vegetation index, canopy cover, texture and grain (Mardanisamani et al., 2019; Yue et al., 2019) are extracted, which can be used for the estimation of biomass (Ogawa et al., 2021; Che et al., 2022), yield (Fei et al., 2023), nitrogen content (Ding et al., 2022) and chlorophyll content (Hu et al., 2010), as well as early diagnosis of nutrients (Wiwart et al., 2009) and disease identification at the plot scale (Cao et al., 2018).

[0003] Compared with two-dimensional spectral data with limited resolution and occlusion problems, three-dimensional spatial data provides more comprehensive structural information of crops (Herrero-Huerta et al., 2020; Sun et al., 2024). The methods for obtaining high-precision three-dimensional data of crops are mainly divided into Light Detection and Ranging (LiDAR) and photogrammetry. LiDAR has been successfully used in the field of crop phenotyping, but the high cost and low convenience of LiDAR limit its application in agriculture (Paulus, 2019; Xiao et al., 2023). Photogrammetry is equipped with a consumer-grade high-resolution RGB camera, based on the fusion of motion structure (SFM) and multi-view stereo vision (MVS), to effectively generate dense point clouds and color information from sparse point clouds (Westoby et al., 2012; Wu, 2013). Due to terrain restrictions and crop occlusion (Xiao et al., 2024), in complex field environments, traditional straight-line oblique photography cannot obtain a sufficient number of viewpoints (Sun et al., 2024). Based on the UAV CCO imaging method, the flight altitude and shooting angle are adjusted according to specific crop conditions to obtain more comprehensive surface information and higher-quality point cloud information economically and efficiently (Xiao et al., 2023; Zhu et al., 2023; Sun et al., 2024).

[0004] Based on 3D data, the property characteristics of the canopy in spatial structure can be effectively extracted (Jimenez-Berni et al., 2018; Deery et al., 2020; Herrero-Huerta et al., 2020). At present, there is an urgent need for fast and automated three-dimensional analysis methods to meet the efficient extraction of crop phenotypes, and then associate them with genotypes to improve breeding efficiency (Duan et al., 2016). Canopy structural characteristics can be obtained from 3D point clouds, such as: 3D voxel index (3DVI) and 3D silhouette coefficient (3DPI) (Jimenez-Berni et al., 2018; Deery et al., 2020), plant heights at different height percentages (Luo et al., 2021), etc. Although they have been proven to be effective in estimating parameters such as crop biomass, canopy 3D features are still scarce in the face of hundreds of crop breeding materials, and it is difficult to meet the ability to distinguish trait differences between multiple genotypes (Li et al., 2021; Ren et al., 2023). According to the feature extraction strategy, point cloud feature description methods are divided into: statistical feature-based and deep learning-based methods (Han et al., 2018). Deep learning-based methods preprocess the original 3D point cloud data into a structured, fixed-size representation that is effectively processed by different network structures (Qi et al., 2017; Zhou and Tuzel, 2017). However, the features obtained by this method have difficulty in explaining the trait differences between different varieties of crops, and the data needs to be labeled (Xiao et al., 2024). The statistical feature-based method can intuitively reflect the characteristics of crop canopy point cloud data, making the feature extraction process easier to understand and explain.

[0005] Methods based on statistical features are mainly divided into label-based methods and statistical histogram-based methods (Han et al., 2018). The statistical histogram-based method simply adds the geometric statistics in each area to effectively describe the local features in the form of a histogram, but the bin width of the histogram needs to be carefully adjusted (Rusu et al., 2009). When the data distribution is uneven, some global information will be lost based on the statistical histogram (Flint et al., 2007). The label-based method calculates the geometric information in the local area of ​​the feature point for encoding to reflect the characteristics of the local area (Pauly et al., 2003; West et al., 2004), and has been widely used in urban scene classification (Weinmann et al., 2015; Hackel et al., 2016), cotton boll identification and yield estimation (Sun et al., 2020; Xiao et al., 2024) and other fields. This method has high classification accuracy in pattern recognition, but in the field of crop phenotyping, the extraction of local information ignores the overall three-dimensional structural information of the crop canopy, and local calculations require a large amount of memory space for storage.

[0006] In summary, there is an urgent need for fast and automated three-dimensional analysis methods to efficiently extract crop phenotypic characteristics, and then associate them with genotypes to optimize breeding programs. However, for hundreds of varieties of breeding materials, there is a lack of features to characterize the 3D traits of the canopy in field plots, making it difficult to effectively identify trait differences between multiple genotypes. Summary of the invention

[0007] The purpose of the present invention is to solve the problem of lack of point cloud feature indicators in the field of agricultural point cloud applications, and to provide a canopy morphology 3D feature analysis method based on lightweight unmanned aerial vehicles and point cloud singular value decomposition. The method combines cross-circle (CCO) route oblique photography with the SFM-MVS algorithm to obtain high-quality, low-cost canopy 3D point clouds of different varieties of crops. The label-based method of the present invention introduces the singular value decomposition (SVD) method for the point cloud matrix, and establishes 9 canopy morphology 3D features at global and local scales. The feature extraction process is automated, and the overall features of the canopy are efficiently extracted at a global scale, and information on the surface details of the canopy is obtained at a local scale. In addition, the significance test of the features was performed under different planting densities, lodging grades, and yield grades. At the same time, the 3D features of the canopy are also used in phenotypic estimation, such as yield estimation and lodging discrimination, to test its effectiveness.

[0008] To achieve the above purpose, the technical solution adopted by the present invention is as follows:

[0009] A canopy morphology 3D feature analysis method based on a lightweight UAV and point cloud singular value decomposition comprises the following steps:

[0010] Step 1: Field data collection:

[0011] In the field data collection, the canopy height, bottom pod height, main stem node number, branch number, number of pods per plant, number of grains per plant, 100-grain weight, and grain yield of crops of different plot varieties were measured; the crop lodging grade was accurately judged through agronomic knowledge, and the crops in each plot were harvested, threshed, and dried separately, and the crops were weighed and the yield was measured when the moisture content was less than 13%;

[0012] Step 2: UAV multi-sensor image acquisition:

[0013] UAVs can be equipped with high-resolution RGB to conduct aerial surveys at different stages of crops;

[0014] Step 3: Raw data preprocessing:

[0015] After the drone aerial operation is completed, the captured data is geometrically corrected and stitched to generate a digital elevation model and a digital orthophoto; the geometric correction and stitching processing specifically includes image alignment, GCP import and dense point cloud generation;

[0016] Step 4: Extraction of canopy spectrum, point cloud structure, and texture information

[0017] Step 41: Extraction of canopy spectrum and texture information: Use high-resolution orthophotos to establish the CIVE index, which is used to segment plants and soil, and then extract soybean canopy coverage; use gray-level co-occurrence matrix to extract RGB texture feature information;

[0018] Step 42: Extract global scale and local scale features based on soybean 3D canopy point cloud;

[0019] Step 5: Based on the global and local scale characteristics of the canopy point cloud calculated in step 4, a time series analysis box plot is constructed according to the yield grade and lodging grade, and a significance test is performed; in addition, the correlation between the characteristics of different growth periods is calculated to characterize the differences within the characteristics;

[0020] Step 6: In phenotypic estimation, a machine learning model was constructed. A regression model was used for yield estimation, and a classification model was used for lodging discrimination. The maximum number of models was set to 10, the maximum running time was set to 60 seconds, and 5-fold cross validation was used to improve the stability and reliability of the evaluation.

[0021] Furthermore, in step 2, the drone adopts two flight modes:

[0022] (1) When the flight altitude is 15m, the shooting mode is oblique photography, the conventional route is adopted, and the overlap rate between images is 85%;

[0023] (2) The flight altitude is 5m, a cross-circling flight route is adopted, and the overlap rate between images is 50%.

[0024] Furthermore, in step three, for the images obtained by CCO route photography, the multi-view images captured along the CCO path are stitched, and the multi-view stereo and motion structure algorithms are used to reconstruct the point cloud of the soybean canopy; the image data collected on the CCO route are image aligned to generate a dense point cloud for subsequent modeling; a vector layer is established to segment a single plot of land, and the RGB orthophoto map, plant height and point cloud data of the plot are obtained; in the process of point cloud data processing, outliers and noise points are removed, and then the RANSAC method is used to fit the ground area, retaining high-precision point cloud data related to the soybean canopy.

[0025] Furthermore, in step 42, in the preprocessing part, the soybean three-dimensional canopy point cloud is screened by color information and ground height information, and the point cloud in the ground area is eliminated; then, the noise points of the point cloud are eliminated by the remove radius outlier method of open3d, in which the neighborhood radius is set to 0.1m and the minimum number of neighborhood points is set to 20; finally, the point cloud is normalized by the center value.

[0026] Furthermore, in step 42, in the global scale canopy 3D feature extraction, the three-dimensional coordinates of the global point cloud are used to construct a matrix NP, and the eigenvectors and eigenvalues ​​of the matrix NP are obtained through singular value decomposition.

[0027] Furthermore, in step 42, in the local scale canopy 3D feature extraction, by traversing each point cloud, using 20, 50, 100, 200, and 500 point clouds of ROI within the K nearest neighbor search area, the small-scale point clouds of different levels are subjected to SVD to obtain eigenvalues ​​and eigenvectors, thereby obtaining L λs ,P λs ,S λs ,E λs ,Σ λs ,C λs ,V λs ,A λs ,T λs , traverse each point cloud in turn and calculate the local canopy 3D features, and finally calculate the average value to get L λ ,P λ ,S λ ,E λ ,Σλ,C λ ,V λ ,A λ,T λ .

[0028] Compared with the prior art, the present invention has the following advantages:

[0029] This paper uses a small light drone combined with a CCO flight line to reconstruct the crop canopy point cloud with high precision. At the same time, the SVD method for the point cloud coordinate matrix is ​​introduced to establish 9 canopy morphological 3D features at different scales and evaluate their potential for phenotypic estimation.

[0030] 1. The global canopy 3D features are significantly correlated with crop yield, lodging, canopy height and vegetation index (p <= 0.05). When the planting density, lodging grade and yield grade are used as the classification basis, the characteristic entropy (E λ ), characteristics and (Σ λ ) and the total variance (V λ ) index was significantly different from the initial pod stage to the full pod stage (p<=0.05), and the ball quality (S λ ), curvature (C λ ) and anisotropy (A λ ) index showed significant difference in the initial grain to full grain initial maturity stage (p<=0.01).

[0031] 2. Based on the global canopy 3D features, it mainly focuses on characterizing the differences in different yield levels of crops from the initial grain stage to the full grain initial maturity stage. Based on the local range of 3D feature extraction, more detailed structural information can be extracted, which is helpful to characterize the differences of crops in more growth stages and improve the ability of phenotypic estimation.

[0032] 3. Compared with features based on only one type, multimodal feature fusion can effectively improve the accuracy of phenotypic estimation. When all features are used for fusion, the yield estimation accuracy is the best (RMSE = 362.10 kg ha -1 , RMAE = 278.31 kgha -1 , rRMSE%=14.31, r=0.63). The accuracy of the four-category lodging discrimination reached the best (Accuracy=0.575, F1-score=0.569), and the accuracy of the two-category lodging discrimination reached (Accuracy=0.901, F1-score=0.892).

[0033] In summary, the 3D features of the canopy at different scales have great potential, and the results can help accelerate the process of crop breeding. In addition, in the future, it is necessary to collect data for many years and multiple time points, and use data representing variety or genotype differences to improve the applicability and stability of the 3D features of the canopy. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1Schematic diagram of the geographical location and site design of the test site; (a) Test area in Heihe City, Heilongjiang Province, China; (b) Orthophoto generated by stitching RGB images; (c) Using DJI Phantom 4RTK, the cross-surround CCO route method is used to obtain the 3D crop canopy RGB image; (d) The 3D crop canopy is reconstructed using the SFM-MVS algorithm, and then the ground point cloud is removed and cropped to obtain the point cloud of a single plot.

[0035] Figure 2 This is a flow chart for the calculation of the canopy morphology 3D feature method. The content is divided into two sections: global scale and local scale feature extraction. 1. In the global scale calculation, all point clouds are directly input, and then the eigenvalues ​​(λ1, λ2, λ3) are obtained through SVD to calculate the global 3D features; 2. In the local scale calculation, by searching for adjacent point clouds, K nearest neighbors are used to search for different numbers of point clouds around each point cloud for feature calculation. By traversing all point clouds, the mean value represents the local 3D feature.

[0036] Figure 3 The correlation matrix of point cloud structure characteristics at different periods, the characteristic factors are yield, lodging grade, global canopy morphology 3D characteristics, voxel ratio, plant height, vegetation index (EXR, EXG, EXGR). Red ellipses indicate positive correlations, and blue ellipses indicate negative correlations. * indicates that P < = 0.05 has a significant correlation. (a) indicates the initial flowering stage of crops; (b) indicates the full flowering stage of crops; (c) indicates the initial pod stage of crops; (d) indicates the full pod stage of crops; (e) indicates the initial grain stage of crops; (f) indicates the full grain and initial maturity stage of crops.

[0037] Figure 4 Figure 2 shows the box plot of the global canopy 3D characteristics at different growth stages under different planting densities. (a) Linear (L λ ); (b) planarity (P λ ); (c) sphericity (S λ ); (d) characteristic entropy (E λ ); (e) characteristic sum (Σ λ ); (f) curvature (C λ ); (g) total variance (V λ ); (h) anisotropy (A λ ); (i) trace (T λ ). D1 indicates low-density planting (green); D2 indicates medium-density planting (orange); D3 indicates high-density planting (red). * indicates p<=0.05, with significant difference. ** indicates p<=0.01, with very significant difference. *** indicates p<=0.001, with extremely significant difference. The purple line between the box plots indicates the connecting line of the mean value.

[0038] Figure 5 The box plots of the global canopy 3D characteristics at different growth stages under different lodging levels. (a) Linear (L λ ); (b) planarity (P λ ); (c) sphericity (S λ ); (d) characteristic entropy (E λ ); (e) characteristic sum (Σ λ ); (f) curvature (C λ ); (g) total variance (V λ ); (h) anisotropy (A λ ); (i) trace (T λ ). Among them, L1 indicates no lodging (green), L3 indicates mild lodging (yellow), L5 indicates moderate lodging (orange), and L7 indicates severe lodging (red). * indicates p<=0.05, and there is a significant difference. ** indicates p<=0.01, and there is a very significant difference. *** indicates p<=0.001, and there is an extremely significant difference. The purple line between the box plots indicates the connecting line of the mean value.

[0039] Figure 6 The box plots of the global canopy 3D characteristics at different growth stages under different yield levels are shown in Figure 2. The yield level is divided into four intervals according to 25%, 50%, 75%, and 100%. (a) Linear (L λ ); (b) planarity (P λ ); (c) sphericity (S λ ); (d) characteristic entropy (E λ ); (e) characteristic sum (Σ λ ); (f) curvature (C λ ); (g) total variance (V λ ); (h) anisotropy (A λ ); (i) trace (T λ ). Y1 represents lower yield (red), Y2 represents low-medium yield (orange), Y3 represents medium-high yield (yellow), and Y4 represents high yield (green). * indicates p<=0.05, and there is a significant difference. ** indicates p<=0.01, and there is a very significant difference. *** indicates p<=0.001, and there is an extremely significant difference. The purple line between the box plots represents the connecting line of the mean value.

[0040] Figure 7Box plots are calculated for linear indicators at different scales, with production levels and periods as the basis for division. (a) The number of local point clouds is 20; (b) The number of local point clouds is 50; (c) The number of local point clouds is 100; (d) The number of local point clouds is 200; (e) The number of local point clouds is 500; (f) All point clouds in the global scale participate in the calculation. * indicates p < = 0.05, there is a significant difference. ** indicates p < = 0.01, there is a very significant difference. *** indicates p < = 0.001, there is an extremely significant difference. The purple line between the box plots represents the connecting line of the mean.

[0041] Figure 8 Box plots are calculated for planarity indicators at different scales, with production levels and periods as the basis for division. (a) The number of local point clouds is 20; (b) The number of local point clouds is 50; (c) The number of local point clouds is 100; (d) The number of local point clouds is 200; (e) The number of local point clouds is 500; (f) All point clouds in the global scale participate in the calculation. * indicates p < = 0.05, there is a significant difference. ** indicates p < = 0.01, there is a very significant difference. *** indicates p < = 0.001, there is an extremely significant difference. The purple line between the box plots represents the connecting line of the mean value.

[0042] Fig. 9 Box plots are calculated for sphericity indicators at different scales, with production levels and periods as the basis for division. (a) The number of local point clouds is 20; (b) The number of local point clouds is 50; (c) The number of local point clouds is 100; (d) The number of local point clouds is 200; (e) The number of local point clouds is 500; (f) All point clouds in the global scale participate in the calculation. * indicates p < = 0.05, there is a significant difference. ** indicates p < = 0.01, there is a very significant difference. *** indicates p < = 0.001, there is an extremely significant difference. The purple line between the box plots represents the connecting line of the mean value.

[0043] Fig.10 Box plots are calculated for the characteristic entropy index at different scales, with production level and period as the basis for division. (a) The number of local point clouds is 20; (b) The number of local point clouds is 50; (c) The number of local point clouds is 100; (d) The number of local point clouds is 200; (e) The number of local point clouds is 500; (f) All point clouds in the global scale participate in the calculation.

[0044] Fig.11Box plots are calculated for features and indicators at different scales, divided by production level and period. (a) The number of local point clouds is 20; (b) The number of local point clouds is 50; (c) The number of local point clouds is 100; (d) The number of local point clouds is 200; (e) The number of local point clouds is 500; (f) All point clouds in the global scale participate in the calculation.

[0045] Fig.12 Box plots of curvature index calculations at different scales, divided by production level and period. (a) The number of local point clouds is 20; (b) The number of local point clouds is 50; (c) The number of local point clouds is 100; (d) The number of local point clouds is 200; (e) The number of local point clouds is 500; (f) All point clouds in the global scale participate in the calculation.

[0046] Fig.13 Box plots of total variance indices at different scales are calculated, with production levels and periods as the basis for division. (a) The number of local point clouds is 20; (b) The number of local point clouds is 50; (c) The number of local point clouds is 100; (d) The number of local point clouds is 200; (e) The number of local point clouds is 500; (f) All point clouds in the global scale participate in the calculation.

[0047] Fig.14 Box plots of anisotropy index calculations at different scales, divided by production level and period. (a) The number of local point clouds is 20; (b) The number of local point clouds is 50; (c) The number of local point clouds is 100; (d) The number of local point clouds is 200; (e) The number of local point clouds is 500; (f) All point clouds in the global scale participate in the calculation.

[0048] Fig.15 Box plots of trace index calculations for different scales, divided by production level and period. (a) The number of local point clouds is 20; (b) The number of local point clouds is 50; (c) The number of local point clouds is 100; (d) The number of local point clouds is 200; (e) The number of local point clouds is 500; (f) All point clouds in the global scale participate in the calculation.

[0049] Fig.16 The yield estimation results of different feature combinations at different growth stages of crops. The results were obtained by 5-fold validation using H2O-AutoML. ALL means that all features are used for modeling. (a) RMSE; (b) MAE; (c) rRMSE%; (d) Pearson's.

[0050] Fig.17The results of different feature combinations for distinguishing four types of lodging at different growth stages of crops. The results were obtained by 5-fold validation using H2O-AutoML. ALL means that all features are used for modeling. (a) Accuracy; (b) Precision; (c) Recall; (d) F1-score.

[0051] Fig.18 The results of different feature combinations for different classifications of lodging at different growth stages of crops. The results were obtained by 5-fold validation using H2O-AutoML. ALL means that all features are used for modeling. (a) Accuracy; (b) Precision; (c) Recall; (d) F1-score. DETAILED DESCRIPTION

[0052] The technical solution of the present invention is further described below in conjunction with the accompanying drawings, but is not limited thereto. Any modification or equivalent replacement of the technical solution of the present invention without departing from the spirit and scope of the technical solution of the present invention should be included in the protection scope of the present invention.

[0053] The present invention uses an RGB lens mounted on an unmanned aerial vehicle combined with the SfM-MVS algorithm to perform high-precision three-dimensional reconstruction of the soybean canopy structure. At the same time, a label-based method introduces the singular value decomposition (SVD) method for the point cloud coordinate matrix, and establishes 9 canopy morphological 3D features at multiple scales as a new interpretable feature at the crop plot level. The results show that the global canopy 3D features are significantly correlated with soybean yield, lodging, canopy height and vegetation index (p<=0.05). 3D feature extraction based on the local range can extract more detailed structural information on the canopy surface, which helps to characterize the differences in more growth periods of soybeans. Finally, when the vegetation index, texture and canopy 3D features of different scales are fused, the yield estimation accuracy using H2O-AutoML is the best (RMSE=362.10kg ha -1 , rRMSE%=14.31), and the accuracy of lodging 4 and 2 categories reached the best (4 categories: Accuracy=0.575, F1-score=0.569, 2 categories: Accuracy=0.901, F1-score=0.892). In summary, the high-precision 3D reconstruction canopy 3D feature extraction technology based on the UAV platform developed in this study shows great potential and provides fast and reliable data support for breeding research.

[0054] The present invention provides a canopy morphology 3D feature analysis method based on a lightweight unmanned aerial vehicle and point cloud singular value decomposition. Taking soybean material as an example, the method comprises the following steps:

[0055] 1. Test location and experimental setup

[0056] The test site is located in Heihe City, Heilongjiang Province, China (47°42′-51°03′N, 124°45′-129°18′E). Heihe City is located at the eastern end of the Greater Khingan Range and the northern end of the Lesser Khingan Range ( Figure 1 a). It belongs to the cold temperate continental monsoon climate, with warm spring and windy weather and abundant rainfall in summer. Figure 1 b) The crops are all soybeans, the ridge length is 3.1m, the width is 0.6m, and the area is about 1.9m 2 The combination of three ridges is the area of ​​a small plot of one variety, the spacing between ridges is about 0.3m, and the area of ​​the small plot is about 5.9m 2 .

[0057] A total of 186 soybean varieties were planted in the study area, with a total of 558 planting plots, of which the low-density (D1) planting density was 3.0×10 5 Seed hm -1 The medium density (D2) planting density is 3.5×10 5 Seed hm -1 High density (D3) is 4.2×10 5 seed -1 All soybean varieties were planted on May 14, 2023, sown on May 30, 2023, and harvested on October 19, 2023. Field management was carried out according to conventional specifications.

[0058] 2. Data Collection

[0059] 2.1 Field data collection

[0060] The canopy height, bottom pod height, main stem node number, branch number, number of pods per plant, number of grains per plant, 100-grain weight, grain yield and lodging grade of soybeans in 558 plots were quantitatively evaluated. The lodging grade was determined by manual visual inspection and divided into 4 grades: no lodging (grade 1), mild lodging (grade 3), moderate lodging (grade 5) and severe lodging (grade 7). Each plot was harvested, threshed and dried separately. The soybeans were weighed when the moisture content was less than 13%. The yield was expressed in kg ha -1 Indicates that, as shown in Table 1, the field statistics.

[0061] Table 1 Descriptive statistics of field measurement parameters

[0062]

[0063]

[0064] 2.2 Methods of UAV data collection

[0065] The drone survey was conducted in multiple stages on July 1, 2023 (first flowering), July 10 (full flowering), July 20 (first pods), July 28 (full pods), August 14 (first grains), and September 2 (full grains and first maturity). The drone model is DJI Phantom 4RTK, and the RGB image resolution obtained is 4608×3456 pixels. The drone adopts two flight modes: (1) When operating at an altitude of 15 meters, oblique photography is used to capture images with 85% forward and lateral overlap. The spatial resolution of the RGB image is 0.38cm / pixel. (2) In order to obtain high-precision canopy point cloud data ( Figure 1 c), the UAV flight altitude was 5m, and the cross-circle (CCO) route (Xiao et al., 2023) was adopted. The overlap rate was 50%, and the camera tilt angle was 45°. All UAV image collection was carried out at noon (11:00-14:00), with clear weather and no wind.

[0066] 2.2. Raw data preprocessing

[0067] After completing the drone orthophoto shooting mission, the captured data was geometrically corrected and stitched using Agisoft PhotoScan Professional software (Agisoft LLC, St. Petersburg, Russia). This includes image alignment, GCP import, and dense point cloud generation. Finally, a digital elevation model and a digital orthophoto were generated. For the images acquired by CCO route photography, Agisoft PhotoScan Professional was used in the study to stitch the multi-view images captured along the CCO path, and the multi-view stereo and motion structure (SFM-MVS) algorithm was used to reconstruct the point cloud of the soybean canopy. Agisoft PhotoScan Professional software performed image alignment on the image data collected along the CCO route to generate a dense point cloud ( Figure 1 d) for subsequent modeling. The vector layer was created using QGIS 3.24 software to segment individual plots ( Figure 1 d), obtain the RGB orthophoto map, plant height and point cloud data of the plot. In the process of point cloud data processing ( Figure 1 d), first remove outliers and noise points, and then use the RANSAC method to fit the ground area (Yang et al., 2022). This allows the ground and soybean plants in the data to be separated. Ultimately, high-precision point cloud data related to the soybean canopy is retained.

[0068] 3. Modeling methods and phenotypic information extraction

[0069] 3.1. Extraction of canopy spectrum and texture information

[0070] The orthophoto image constructed by the drone consists of three bands: red, green and blue. Based on these three bands, the following Figure 2 The vegetation index shown is used for yield estimation and lodging discrimination. At the same time, the CIVE index is established using high-resolution orthophotos to segment plants and soil, and then extract soybean canopy cover (CC). At the same time, in the high-throughput plant phenotyping monitoring of unmanned aerial vehicles in the field, texture information has been proven to have a good effect on yield and lodging monitoring (Nichol and Sarker, 2011). Therefore, the present invention uses the gray-level co-occurrence matrix (GLCM) (Haralick and Shanmugam, 1973) to extract RGB texture feature information. After converting the RGB image into a grayscale image, homogeneity (HO), contrast (CO), dissimilarity (DI), characteristic entropy (EN), second-order moment (SE) and correlation (CO) are extracted respectively, as shown in Table 2.

[0071] Table 2 Extraction of vegetation index and texture features

[0072]

[0073] 3.2. Extraction of canopy point cloud structure information

[0074] The canopy point cloud structure information established by the present invention mainly includes canopy morphological 3D features, canopy height and voxel ratio features. The point cloud surface normal vector is an important attribute of the geometric surface. In the present invention, the problem of estimating the surface normal is converted into calculating the eigenvectors and eigenvalues ​​of the covariance matrix. In the present invention, the global scale and local scale feature extraction is constructed based on the soybean three-dimensional canopy point cloud, such as Figure 2 As shown in Algorithm 1. In the preprocessing part, the soybean 3D canopy point cloud is screened by color information and ground height information, and the ground area point cloud is removed. Then, the noise points of the point cloud are removed by the open3d noise point removal method, where the neighborhood radius is set to 0.1m and the minimum number of neighborhood points is set to 20. Finally, the point cloud is normalized by the center value.

[0075] In the global scale canopy 3D feature extraction ( Figure 2 The global scale) is used to construct the matrix NP of the three-dimensional coordinates of the global point cloud, and the eigenvectors and eigenvalues ​​of the matrix NP are obtained through singular value decomposition (SVD). First, the covariance matrix C of the global point cloud is calculated cov , and then find C cov The eigenvalues ​​(λ1, λ2, λ3) and eigenvectors (u1, u2, u3) of are shown in Equations 1 and 2. The 3D feature linearity (L λ ), planarity (P λ ), sphericity (S λ), characteristic entropy (E λ ), characteristics and (Σ λ ), curvature (C λ ), total variance (V λ ), anisotropy (A λ ), trace (T λ )(Weinmann et al., 2015), the calculation method is shown in Table 3.

[0076]

[0077] C cov =NP T ·NP#(2)

[0078] In the local scale canopy 3D feature extraction ( Figure 2 Local scale), which is different from the global scale in that it takes into account the local characteristic information of the soybean canopy. By traversing each point cloud, using 20, 50, 100, 200, and 500 point clouds of ROI (full name first) within the K nearest neighbor search area, the small-scale point clouds of different levels are subjected to SVD to obtain the eigenvalues ​​(λ 1s ,λ 2s ,λ 3s ) and the eigenvector (u 1s ,u 2s ,u 3s ), thus obtaining L λs ,P λs ,S λs ,E λs ,Σ λs ,C λs ,V λs ,A λs ,T λs . Traverse each point cloud in turn and calculate the local canopy 3D features, and finally calculate the average value to obtain L λ ,P λ ,S λ ,E λ ,Σλ,C λ ,V λ ,A λ ,T λ .

[0079] The plant height of soybean at different growth stages and the point cloud voxel ratio are also important information. The present invention obtains high-precision canopy point clouds of soybeans through the CCO route, and subtracts the ground point cloud elevation from the canopy point cloud elevation to obtain canopy height information. The canopy height information extraction is divided into the point cloud height average, maximum, and cumulative height 95% quantile. The point cloud voxel ratio is an indicator used to evaluate the sparsity of point cloud data. The voxel ratio is generally defined as the ratio of the number of actual points in the point cloud to the number of non-empty (containing points) voxels in the voxels after the point cloud is voxelized (Liu et al., 2021). In the study, the sizes of the voxelized grids were set to (0.05, 0.10, 0.20, 0.30, 0.40, 0.50) for calculation.

[0080] Table 33D feature extraction method

[0081]

[0082]

[0083] 3.3 Model building method and environment

[0084] In the present invention, the development language is Python 3.7.5, the GPU model is Tesla V100, the CPU is 2 Cores, and the memory is 32 GB. The 3D features of the canopy point cloud are calculated using Open3d 3.7 and Numpy 1.25.0 libraries. In phenotypic estimation, a machine learning model was constructed using H2O-AutoML 3.46 (LeDell and Poirier, 2020), a regression model was used for yield estimation, and a classification model was used for lodging discrimination. The maximum number of models is set to 10, and the maximum running time is 60s. In order to reduce the randomness of phenotypic estimation and improve the generalization ability of the model, 5-fold cross validation was used in the study to improve the stability and reliability of the evaluation.

[0085] 3.4 Statistical methods

[0086] 3.4.1 Regression Indicators

[0087] The root mean square error (RMSE), root mean square error (RMAE), relative RMSE (rRMSE%), and correlation coefficient r are used to evaluate the estimation accuracy of soybean yield, etc., as shown in formula 3-6. i and They are the measured and estimated values ​​of soybean yield respectively. is the average value of the measured soybean yield, and n is the total number of samples in the test set.

[0088]

[0089] 3.4.2 Lodging classification indicators

[0090] The accuracy, precision, recall, and F1-score are used to evaluate the accuracy of crop lodging and measure the training quality of the model, as shown in formulas 7-10. TP is the number of true positive samples, FP is the number of false positive samples, and FN is the number of false negative samples.

[0091]

[0092] 3.4.3ANOVA test

[0093] In order to explore the performance and applicability of canopy 3D features in different planting densities, lodging grades and yield grades, a one-way analysis of variance (ANOVA) was performed using the significant difference (HSD) Tukey test (α = 0.001) to test the significant differences in canopy 3D features between different planting densities, different lodging grades and different yield grades. The determination of yield grade is based on the measured data being divided into 4 grades, which are composed of the 25%, 50%, 75% and 100% quantiles of yield as intervals. 0-25% is low yield (Y1), 25-50% is low-medium yield (Y2), 50%-75% is medium-high yield (Y3), and 75-100% is high yield (Y4).

[0094] 4. Results

[0095] 4.1. Multi-period correlation of different characteristic factors

[0096] At different growth stages, yield, lodging grade, global canopy morphological 3D features, voxel ratio, plant height, and vegetation index (EXR, EXG, EXGR) were used to construct the correlation matrix, e.g. Figure 3 As shown in the figure, * indicates a significant correlation with p <= 0.05 in the Pearson's calculation. In the analysis of different growth periods, the point cloud structure characteristics showed a certain correlation with yield, lodging grade and vegetation index. λ ) as an example, in the initial flowering to full grain and initial maturity stage, it showed a significant negative correlation with yield (p <= 0.05). λ The higher the value, the lower the yield. In the initial flowering-full flowering stage, since lodging has not yet occurred, there is no obvious correlation. In the initial grain-full grain early maturity period, soybeans have already lodged, so there is a significant positive correlation with the lodging grade (Pearson's initial grain: 0.40; full grain early maturity: 0.44; p <= 0.05). This shows that the higher the lodging grade, the more serious the lodging phenomenon, S λ In addition, during the initial flowering to full grain maturity stage, S λThere was a significant positive correlation with canopy height (H95%, Hmean, Hmax) (Pearson's: 0.08-0.49, p <= 0.05). λ There was also a significant correlation with VIs (Pearson's: 0.08-0.38, p <= 0.05), and in the stage of full pods to full grains, there was a significant correlation with S λ It showed a weak correlation (Pearson's: -0.02-0.15, p <= 0.05). For the point cloud voxel ratio feature, it showed a positive correlation in the early flowering-full flowering period, and a weak correlation in the early pod-full grain and early maturity period. For the point cloud extracted canopy plant height feature (H95%, Hmean, Hmax), it showed a significant positive correlation with yield in the early flowering-full pod period (Pearson's: 0.12-0.36, p <= 0.05). In the early grain-full grain and early maturity period, with the occurrence of leaf shedding and lodging of early soybean varieties, plant height showed a weak negative correlation with yield (Pearson's: -0.28--0.09, p <= 0.05), and showed a significant positive correlation with lodging grade (Pearson's: 0.33-0.55, p <= 0.05).

[0097] 4.2. Time series results of global canopy 3D features under different planting densities

[0098] The global canopy 3D feature time series results were evaluated at different planting densities ( Figure 4 ). For linear (L λ ) and planarity (P λ )index( Figure 4 ab), in the initial flowering to full flowering stage, D1, D2 and D3 were significantly different (p <= 0.05). In the early growth stage of soybean, the higher the planting density, the higher the L λ The value is relatively high, while P λ For sphericity (S λ ), curvature (C λ ) and anisotropy (A λ )index( Figure 4 cfh), and there were significant differences in the initial pod, initial grain and full grain initial maturity stages. Especially in the full grain initial maturity stage, there were extremely significant differences among D1, D2 and D3 (p<=0.001). With the increase of planting density, the point cloud features of soybeans under D3 condition showed higher divergence and curvature, S λ , C λ The value increased relatively, while A λ The value is relatively reduced. For the characteristic entropy (E λ ), characteristics and (Σ λ ) and the total variance (Vλ )index( Figure 4 deg), mainly in the primary pod-primary grain stage, there was a significant difference (p<=0.001). The higher the planting density, the higher the E λ There is an increase, Σ λ With V λ It is shown as a decrease. For the trace (T λ )index( Figure 4 i) There were significant differences in the period from full pod to initial grain (p<=0.01). The higher the planting density, the higher the T λ The lower.

[0099] 4.3. Global canopy 3D feature time series results at different lodging levels

[0100] The global canopy 3D feature time series results were evaluated at different lodging levels ( Figure 5 ). For linear (L λ ) and planarity (P λ )index( Figure 5 ab), the differences in different lodging grades were small during the initial flowering to full grain maturity period. λ ), curvature (C λ ) and anisotropy (A λ )index( Figure 5 cfh), in the initial grain-full grain initial maturity stage, L1, L3, L5 and L7 have extremely significant differences (p <= 0.001). As the soybean lodging becomes more serious, the point cloud features show higher divergence and curvature, S λ , C λ The value increased significantly, and A λ The value is relatively reduced. λ ), characteristics and (Σ λ ) and the total variance (V λ )index( Figure 5 deg), mainly in the initial pod-full pod stage, there was a significant difference (p <= 0.05). λ There is an increase, Σ λ With V λ It is shown as a decrease. For the trace (T λ )index( Figure 5 i) There were significant differences in lodging grades at the pod-filling stage (p <= 0.01). The higher the planting density, the higher the T λ The lower.

[0101] 4.4. Global canopy 3D feature time series results at different yield levels

[0102] Global canopy 3D feature time series results were evaluated at different yield levels ( Figure 6). For linear (L λ ) and planarity (P λ )index( Figure 6 ab), the differences in different lodging grades were small during the initial flowering to full grain maturity period. λ ), curvature (C λ ) and anisotropy (A λ )index( Figure 6 cfh), in the initial grain-full grain initial maturity stage, there were extremely significant differences between Y1 and Y2, Y3, and Y4 (p <= 0.001). The lower the soybean plot yield, the higher the divergence, curvature, and anisotropy of the canopy 3D characteristics. λ , C λ The value increased significantly, and A λ The value is relatively reduced. λ ), characteristics and (Σ λ ) and the total variance (V λ )index( Figure 6 deg), mainly in the primary grain stage, there was a significant difference (p <= 0.01). λ It shows a gradual increase, while Σ λ With V λ It is shown as a decrease. For the trace (T λ )index( Figure 6 i) There was a significant difference between Y1 and Y4 at the initial seed stage (p <= 0.001). The higher the yield of the plot, the higher the yield of T λ The lower the value.

[0103] 4.5. Time series results of local extraction of 3D features of different canopies

[0104] The 3D characteristics of the canopy at different scales were evaluated. The yield grade was used as the classification in the study, and the significant differences in the 3D characteristics of the canopy at different stages were statistically analyzed. For the linearity and planarity index ( Figure 7-8 ), although the feature differences at the global scale are not significant ( Figure 7 f), but in the primary stage, when the local range ROI = (50, 100, 200) ( Figure 7 ad, Figure 8 be), which can obtain more detailed characteristics of soybeans, so there are significant differences (p <= 0.01). Yield levels Y1 and (Y2, Y3, Y4) show significant differences. As the yield level increases, L λ With P λ It shows an upward trend. λ )index( Fig.15 ), the difference between local features and global features is small.

[0105] For sphericity (S λ ), curvature (C λ ) and anisotropy (A λ )index( Fig. 9 , Fig.12 , Fig.14 ), different significant differences appeared at different scales. At the global scale, Y1 showed significant differences with other yield grades (p<=0.001) when concentrated in the first grain-full grain first maturity stage. When ROI=(20, 50, 100), there was a significant difference between the low yield grade and Y4 at the first grain stage (p<=0.01). When ROI=500, more detailed information was obtained in the local range, so that the first pod-full grain first maturity stage showed significant differences (p<=0.05). S λ Index in global scope and local scope ( Fig. 9 af), as the production level continues to increase, S λ All showed a downward trend.

[0106] For the characteristic entropy (E λ ), characteristics and (Σ λ ) and the total variance (V λ )index( Fig.10 , Fig.11 , Fig.13 ), E λ With Σ λ Under local feature extraction, the differences between different yield grades are small. λ When ROI = (20, 50), local feature extraction shows more significant differences, such as increasing the difference between Y4 and other yield levels. At the same time, local feature extraction and global feature extraction show an opposite upward trend ( Fig.13 ), although at the global scale, the higher the production level, the higher the V λ The lower the value, the higher the yield level will be, because the local extraction can obtain more detailed information. λ The higher the status.

[0107] 4.6 Results of yield estimation and lodging identification using different strategies

[0108] The accuracy of yield estimation of different soybean trait combination strategies was evaluated at different growth stages ( Fig.16 ). At the pod-filling stage, when only the canopy 3D (global) features were used, the yield estimation RMSE was 428.05 kg ha -1 When the 3D (global + local) features are combined with local information, the yield estimation RMSE is relatively reduced by 5.16 kgha -1 . When using VIs and 3D (global + local) combined features ( Fig.16VIs+3D (global+local), yield estimation RMSE = 376.10 kgha -1 , RMAE = 285.63 kgha -1 When different features were combined (Fig. 16ALL), the yield estimation accuracy was the best (RMSE = 362.10 kg ha -1 ,RMAE=278.31kg ha -1 , rRMSE%=14.31, r=0.63).

[0109] The results of the four types of lodging discrimination with different combinations of characteristics at different growth stages are as follows: Fig.17 As shown in the figure. With the advancement of the soybean growth period, the lodging phenomenon becomes more and more obvious, and the F1-score of the four-category discrimination of lodging is constantly improving. At the full grain and initial maturity stage, the F1-score of the four-category discrimination using VIs is 0.438, while the F1-score is increased by 0.024 when the canopy 3D (global) features are used, and the F1-score is increased by 0.092 when the 3D (global + local) features are used. This shows that combining the 3D features of the canopy can effectively improve the accuracy of lodging discrimination. When VIs is combined with 3D (global + local), the accuracy of the four-category discrimination of lodging reaches the best (Accuracy = 0.575, Precision = 0.571, Recall = 0.575, F1-score = 0.569).

[0110] The results of different feature combinations for distinguishing the two types of lodging at different growth stages are as follows: Fig.18 As shown. From the initial flowering stage to the full grain and initial maturity stage of soybean, the accuracy of the two types of lodging discrimination gradually improved. In the full grain and initial maturity stage of soybean, when only VIs was used for lodging discrimination, F1-score = 0.855, while when 3D (global + local) features were used, F1-score increased by 0.021. When all features were combined, the accuracy was improved to the highest (Accuracy = 0.901, Precision = 0.903, Recall = 0.901, F1-score = 0.892).

[0111] 5.1 Global interpretation of 3D features of canopy morphology

[0112] Although 3D features of canopy morphology have been named (Pauly et al., 2003; West et al., 2004; Weinmann et al., 2015), their implications in crop phenotyping still need further evaluation. Figure 3 ), linear (L λ ), planarity (P λ ) and trace (Tλ ) index showed a significant correlation. The correlation test was performed with the lodging grade at the initial flowering, initial pod and full pod stages (p <= 0.05), but there was no significant correlation at the initial grain-full grain initial maturity stage, indicating that L λ , P λ , T λ It is helpful to characterize the differences in early crop lodging. In addition, (Dittrich et al., 2017) showed that linear (L λ ) and planarity (P λ ) is susceptible to noise, which makes the classification task vulnerable.

[0113] For sphericity (S λ ), curvature (C λ ) and anisotropy (A λ ) index, in the correlation test, S λ With C λ There was a significant positive correlation (Pearson's = 0.98-1.00, P < = 0.05) with A λ There was a significant negative correlation (Pearson's = -1.00--0.98, P < = 0.05). In the first grain to full grain early maturity period, the soybean lodging condition was more obvious, so it showed a more significant positive correlation with the lodging grade ( Figure 3 ef). Through the significance test of different lodging grades and yield grades ( Figure 5-6 ), indicating that the stronger the soybean lodging grade, the stronger the surface curvature and structural divergence of the point cloud, S λ , C λ The value increases, A λ The value decreases. As the soybean yield level increases, the curvature and divergence of the point cloud surface decrease. λ , C λ and A λ It has the potential to be used to describe soybean lodging traits and estimate yield. Figure 4 efh), the increase of planting density will also lead to the increase of curvature and divergence of point cloud structure.

[0114] For the characteristic entropy (E λ ), characteristics and (Σ λ ) and the total variance (V λ ) index, in the correlation test, E λ and (Σ λ , V λ ) showed a significant negative correlation (Pearson's=-1.00--0.90, P<=0.05). λIt is defined by Shannon entropy (Shannon, 1948), which measures the complexity of the point cloud information it contains (Weinmann et al., 2015; Dittrich et al., 2017). In the soybean pod-early stage, there is a significant correlation with yield and lodging ( Figure 3 de). At the same time, a significance analysis was performed on different planting densities ( Figure 4 ), D1 and D3 showed significant differences. λ The lower the value, the more significant the structure in the point cloud data is under low density, while under high density, the point cloud structure is more complex and tends to be chaotic. λ The value is lower, showing significant differences from L5 and L7 (p <= 0.001), indicating that for soybean plots without lodging, the point cloud structure is stronger, while for plots with moderate and severe lodging, the point cloud structure is more chaotic and disordered. Therefore, it is consistent with the study of (Dittrich et al., 2017) that the curvature (C λ ) and characteristic entropy (E λ ) features are more suitable for classification tasks.

[0115] 5.2 Effect of scale on canopy 3D features

[0116] In photogrammetry, computer vision and other technologies, the selection of scale is important for the automatic analysis of canopy 3D point clouds (Dittrich et al., 2017). (Pauly et al., 2003) The selection of neighborhood scales depends on the covariance characteristics of local surface changes. (Demantké et al., 2012) Covariance characteristics based on linearity, planarity and divergence are used for dimensional scale selection. (Weinmann et al., 2015) Covariance characteristics based on feature entropy are used for eigen-feature entropy-based scale selection. The latter two methods regard the scale corresponding to the minimum Shannon entropy as the optimal scale. However, these scale selection methods are mainly used for semantic classification tasks of large-scale point clouds (Weinmann et al., 2015; Hackel et al., 2016). In the phenotyping of multi-variety soybeans, especially at the plot scale, the calculation of canopy 3D features is limited to independent plots. The use of global point cloud computing can effectively estimate yield and distinguish lodging ( Figure 16-18 ), and characterize the differences in different planting densities, lodging grades, and yield grades ( Figure 4-6In addition, compared with the global scale, the local scale 3D feature extraction can obtain more detailed structural information (Weinmann et al., 2015; Hackel et al., 2016), which is helpful to characterize the differences in other growth stages ( Figure 7-15 ). For example, linear features are not very useful at the global scale ( Fig. 9 f) No difference was found, while when ROI = (20, 50, 100, 200) ( Fig. 9 ad), there were significant differences in different yield levels at the primary grain stage (p<=0.01). Fig. 9 f), mainly manifested in significant differences in yield levels at the initial grain to full grain initial maturity stage. However, at the local scale, when ROI = 500 ( Fig. 9 e) There were significant differences in different yield grades from the initial pod to full grain and initial maturity stages.

Claims

1. A canopy morphology 3D feature analysis method based on lightweight UAV and point cloud singular value decomposition, characterized by The method comprises the following steps: Step 1: Field data collection: In the field data collection, the canopy height, bottom pod height, main stem node number, branch number, number of pods per plant, number of grains per plant, 100-grain weight, and grain yield of crops of different plot varieties were measured; the crop lodging grade was accurately judged through agronomic knowledge, and the crops in each plot were harvested, threshed, and dried separately, and the crops were weighed and the yield was measured when the moisture content was less than 13%; Step 2: UAV multi-sensor image acquisition: UAVs can be equipped with high-resolution RGB to conduct aerial surveys at different stages of crops; Step 3: Raw data preprocessing: After the drone aerial operation is completed, the captured data is geometrically corrected and stitched to generate a digital elevation model and a digital orthophoto; the geometric correction and stitching processing specifically includes image alignment, GCP import and dense point cloud generation; Step 4: Extraction of canopy spectrum, point cloud structure, and texture information Step 41: Extraction of canopy spectrum and texture information: Use high-resolution orthophotos to establish the CIVE index, which is used to segment plants and soil, and then extract soybean canopy coverage; use gray-level co-occurrence matrix to extract RGB texture feature information; Step 42: Extract global scale and local scale features based on soybean 3D canopy point cloud; Step 5: Based on the global and local scale characteristics of the canopy point cloud calculated in step 4, a time series analysis box plot is constructed according to the yield grade and lodging grade, and a significance test is performed; in addition, the correlation between the characteristics of different growth periods is calculated to characterize the differences within the characteristics; Step 6: In phenotypic estimation, a machine learning model was constructed. A regression model was used for yield estimation, and a classification model was used for lodging discrimination. The maximum number of models was set to 10, the maximum running time was set to 60 seconds, and 5-fold cross validation was used to improve the stability and reliability of the evaluation.

2. The canopy morphology 3D feature analysis method based on lightweight unmanned aerial vehicle and point cloud singular value decomposition according to claim 1 is characterized in that: In step 2, the drone uses two flight modes: (1) When the flight altitude is 15m, the shooting mode is oblique photography, the conventional route is adopted, and the overlap rate between images is 85%; (2) The flight altitude is 5m, a cross-circling flight route is adopted, and the overlap rate between images is 50%.

3. The canopy morphology 3D feature analysis method based on lightweight unmanned aerial vehicle and point cloud singular value decomposition according to claim 1 is characterized by: In step three, for the images obtained by CCO route photography, the multi-view images captured along the CCO path are stitched, and the multi-view stereo and motion structure algorithms are used to reconstruct the point cloud of the soybean canopy; the image data collected on the CCO route are aligned to generate a dense point cloud for subsequent modeling; a vector layer is established to segment a single plot of land, and the RGB orthophoto map, plant height and point cloud data of the plot are obtained; in the process of point cloud data processing, outliers and noise points are removed, and then the RANSAC method is used to fit the ground area, retaining the high-precision point cloud data related to the soybean canopy.

4. The canopy morphology 3D feature analysis method based on lightweight unmanned aerial vehicle and point cloud singular value decomposition according to claim 1 is characterized by: Step 42: In the preprocessing part, the soybean three-dimensional canopy point cloud is screened by color information and ground height information, and the point cloud in the ground area is eliminated; then, the noise points of the point cloud are eliminated by the remove radius outlier method of open3d, where the neighborhood radius is set to 0.1m and the minimum number of neighborhood points is set to 20; finally, the point cloud is normalized by the center value.

5. The canopy morphology 3D feature analysis method based on lightweight unmanned aerial vehicle and point cloud singular value decomposition according to claim 1, characterized in that: Step 42: In the global scale canopy 3D feature extraction, the three-dimensional coordinates of the global point cloud are used to construct a matrix NP, and the eigenvectors and eigenvalues ​​of the matrix NP are obtained through singular value decomposition.

6. The canopy morphology 3D feature analysis method based on lightweight unmanned aerial vehicle and point cloud singular value decomposition according to claim 1, characterized in that: Step 42: In the local scale canopy 3D feature extraction, by traversing each point cloud, using 20, 50, 100, 200, and 500 point clouds of ROI within the K nearest neighbor search area, the small-scale point clouds of different levels are subjected to SVD to obtain eigenvalues ​​and eigenvectors, thereby obtaining L λs ,P λs ,S λs ,E λs ,Σ λs ,C λs ,V λs ,A λs ,T λs , traverse each point cloud in turn and calculate the local canopy 3D features, and finally calculate the average value to get L λ ,P λ ,S λ ,E λ ,Σλ,C λ ,V λ ,A λ ,T λ .

Images