A new three-dimensional index construction method for estimating crop yield based on point cloud and spectrum fusion

Abstract

The application discloses a new three-dimensional index construction method for estimating crop yield based on point cloud and spectrum fusion, first, a vegetation index is constructed through a crop unmanned aerial vehicle multi-spectral image, and a linear regression model is established with a canopy chlorophyll; then, crop point cloud is acquired through an unmanned aerial vehicle laser radar, the point cloud is fused with the multi-spectral image by using a multi-dimensional matrix fusion method, and crop multi-spectral point cloud is generated; simultaneously, the linear regression model of the canopy chlorophyll is applied to the crop multi-spectral point cloud, so that the three-dimensional spatial distribution of the crop canopy chlorophyll is generated; finally, the three-dimensional spatial distribution of the canopy chlorophyll is vertically layered and the 75% quantile (CCC_P 75th ) of the vertical layer is counted to estimate the yield. Results show that the new three-dimensional index (CCC_P 75th ) constructed by the application contains the structure and physiological characteristics of crops, has better estimation ability on yield compared with traditional indexes, and has important guiding significance for crop high-yield plant type screening.

Description

Technical Field

[0001] This invention belongs to the field of non-destructive monitoring of crop life information in precision agriculture, and relates to a new index construction method for crop yield estimation, specifically a new three-dimensional index construction method for estimating crop yield based on point cloud and spectral fusion. Background Technology

[0002] Most current research uses agronomic traits or remote sensing indicators to indicate crop yield phenotypes, thereby screening for high-yielding plant types. However, these yield prediction indicators are often limited in form and lack novel indicators that can simultaneously describe crop structure and physiological traits. While many researchers have used machine learning methods to fuse crop structure and physiological traits to estimate yield, machine learning requires a large number of training samples and lacks robust mechanistic understanding. The aforementioned indicators and methods cannot meet the requirements for high-yielding breeding screening of ideal plant types with complex population photosynthetic physiology. Therefore, there is an urgent need for a new indicator that can simultaneously reflect crop structure and physiological information, providing a simple, rapid, and high-throughput indication of crop yield information.

[0003] Traditional passive optical remote sensing can only acquire two-dimensional images or spectral information, failing to characterize the three-dimensional structural information of vegetation. LiDAR, as an active sensing technology, records the three-dimensional structure and morphology of objects by generating point clouds through laser pulses. However, the spectral information provided by LiDAR is limited and cannot be directly used alone to acquire the physiological parameters of vegetation. The fusion of LiDAR point clouds and spectral images can simultaneously provide both the three-dimensional structure and spectral characteristics of vegetation. Therefore, constructing a novel yield estimation index that combines both structural and physiological characteristics through point cloud and spectral fusion is worthy of exploration. Summary of the Invention

[0004] The purpose of this invention is to provide a new three-dimensional index construction method for estimating crop yield based on point cloud and spectral fusion. This method utilizes the fusion of crop lidar point cloud and multispectral imagery to construct a novel three-dimensional index that can simultaneously describe crop structure and physiological traits, thereby improving the accuracy of crop yield estimation.

[0005] To achieve the above objectives, the technical solution adopted by this invention is: a new three-dimensional index construction method for estimating crop yield based on point cloud and spectral fusion, which includes the following steps:

[0006] Step 1: Collect multispectral images of crops using a drone equipped with a multispectral camera;

[0007] Step 2: Collect 3D point cloud data of crops using a drone equipped with a lidar.

[0008] Step 3: Obtain crop canopy chlorophyll data;

[0009] Step 4: Fuse the crop point cloud with the spectrum using a multi-dimensional matrix fusion method;

[0010] Step 5: Obtain multispectral reflectance and construct a vegetation index for estimating canopy chlorophyll;

[0011] Step 6: Establish a linear regression model between vegetation index and crop canopy chlorophyll;

[0012] Step 7: Obtain crop multispectral point cloud;

[0013] Step 8: Apply the crop canopy chlorophyll linear regression model to the multispectral point cloud to generate the three-dimensional spatial distribution of crop canopy chlorophyll;

[0014] Step 9: Vertically stratify the crop canopy chlorophyll to obtain the average chlorophyll content of each layer, and perform statistical analysis to obtain the quantiles of these chlorophyll contents;

[0015] Step 10: Obtain the quantiles of the new three-dimensional indicator chlorophyll content and establish a linear regression model with crop yield, and verify the accuracy.

[0016] Furthermore, in step 1, a drone equipped with an AIRPHEN multispectral camera is used to acquire multispectral images of crops.

[0017] Furthermore, in step 2, a UAV equipped with a Riegl miniVUX-3 lidar system is used to acquire crop lidar point clouds.

[0018] Furthermore, in step 3, the field-measured chlorophyll content of the canopy is obtained. Specifically, this is achieved by randomly selecting a certain number of leaves from each leaf layer of each plot using a Dualex scientific chlorophyll meter, measuring them, and calculating the average value. Then, the Dualex readings of the crop leaves are converted into chlorophyll content using the following formula:

[0019] LCC value =1.49 * Dualex value -3.60

[0020] Among them, LCC value Represents leaf chlorophyll content (μg / cm³) 2 Dualex value This represents the Dualex reading.

[0021] Furthermore, in step 4, the method for fusing the crop point cloud with the spectrum is as follows:

[0022] First, the LiDAR point cloud is exported as a Digital Surface Model (DSM), with the DSM pixel size matching that of the multispectral image. Then, the multispectral image is spatially registered with the DSM as a reference surface.

[0023] Then, the lidar point cloud is divided into a matrix composed of columns that are horizontally aligned with the multispectral pixel grid. The highest point in each column is selected, and other points are excluded. The highest point in each column is projected onto the multispectral image, so that the coordinates (x, y, z) of the point cloud are fused with the reflectivity of multiple bands of the image in a multidimensional matrix.

[0024] Finally, the point cloud projection is removed to restore its three-dimensional structure, thus completing the fusion of point cloud and spectral reflectance.

[0025] Furthermore, in step 5, the band reflectance of the UAV multispectral image is used to calculate a vegetation index suitable for estimating the chlorophyll content of the crop canopy.

[0026] Furthermore, in step 6, a linear regression model is established between the constructed vegetation index and the canopy chlorophyll content; the specific model formula is as follows:

[0027] y = 0.37x - 0.57

[0028] Where y represents the canopy chlorophyll content (g / m²) 2 ), where x represents the vegetation index CI. green .

[0029] Furthermore, in step 7, the crop multispectral point cloud is the point cloud data that incorporates the crop multispectral reflectance.

[0030] Furthermore, in step 8, the spectral reflectance of each point in the crop multispectral point cloud is used to construct a vegetation index, and then a linear regression model of canopy chlorophyll content is applied to each point to generate the three-dimensional spatial distribution of crop canopy chlorophyll.

[0031] Furthermore, in step 9, based on the three-dimensional spatial distribution of chlorophyll in the crop canopy, the point cloud is vertically layered at 0.05m height intervals, and the average chlorophyll content of each layer is calculated. The 75th percentile of these chlorophyll contents, i.e., CCC_P, is obtained through statistical analysis. 75th .

[0032] Furthermore, in step 10, a linear regression model is established between the 75th percentile of the vertical stratification of canopy chlorophyll and yield, and the accuracy of the model is verified using independent year data, with the coefficient of determination R0 being used. 2 The results are comprehensively evaluated using the root mean square error (RMSE) and the relative root mean square error (rRMSE), as shown in the following formula:

[0033]

[0034] Where y i Let y represent the measured value of sample i. i ′ represents the estimated value of sample i. This represents the average of all measurements, and n represents the total number of samples.

[0035] The beneficial effects of this invention are as follows: Based on the fusion of UAV lidar point cloud and multispectral imagery, this invention constructs a novel three-dimensional index that simultaneously incorporates crop structural and physiological characteristics. Compared to traditional indicators, it has a better ability to estimate yield, which is of significant guiding importance for screening high-yielding crop plant types. Validation using independent datasets revealed that the novel three-dimensional index has high yield estimation accuracy across different crop growth stages, demonstrating the high accuracy and universality of the method presented in this invention.

[0036] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0037] Figure 1 This is a flowchart of the method for constructing a new three-dimensional index for estimating crop yield according to the present invention.

[0038] Figure 2 This is a schematic diagram of the fusion of lidar point cloud and multispectral imagery.

[0039] Figure 3 It is the optimal vegetation index CI green Scatter plot of linear regression with output.

[0040] Figure 4 This is a schematic diagram of the three-dimensional spatial and vertical distribution of chlorophyll content in crop canopy.

[0041] Figure 5 This is a graph verifying the yield estimation results. Detailed Implementation

[0042] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0043] like Figure 1 As shown, a novel three-dimensional index construction method for estimating crop yield based on point cloud and spectral fusion mainly includes the following steps:

[0044] S101: Acquire multispectral images of crops using a drone equipped with a multispectral camera;

[0045] S102: Collect 3D point clouds of crops using a drone equipped with a lidar;

[0046] S103: Obtain crop canopy chlorophyll data;

[0047] S104: Fusing crop point clouds and spectra using a multi-dimensional matrix fusion method;

[0048] S105: Obtain multispectral reflectance to construct a vegetation index for estimating canopy chlorophyll;

[0049] S106: Establish a linear regression model between vegetation index and canopy chlorophyll;

[0050] S107: Acquire crop multispectral point cloud;

[0051] S108: Apply the linear regression model of crop canopy chlorophyll to multispectral point clouds to generate the three-dimensional spatial distribution of crop canopy chlorophyll;

[0052] S109: Based on the three-dimensional spatial distribution of chlorophyll in the crop canopy, the point cloud was vertically layered at 0.05m height intervals, and the average chlorophyll content of each layer was calculated. The 75th percentile (CCC_P) of these chlorophyll contents was obtained through statistical analysis. 75th ).

[0053] S110: Incorporate the new three-dimensional index CCC_P 75th A linear regression model was established with crop yield, and its accuracy was verified.

[0054] The specific details of the above steps are as follows:

[0055] S101: Uses a drone equipped with an AIRPHEN multispectral camera to acquire multispectral images of crops at different growth stages (heading stage, heading stage, flowering stage, and grain-filling stage).

[0056] Eight winter wheat varieties were selected for the experiment: Jimai 22 (V1), Yangmai 16 (V2), Ningmai 13 (V3), Yangfumai 4 (V4), Yangmai 23 (V5), Huaimai 33 (V6), Yannong 19 (V7), and Zhenmai 12 (V8); three nitrogen levels were used: 0 kg / hm². 2 (N0), 150 kg / hm 2 (N1), 300 kg / hm 2 (N2), with 50% of the nitrogen application occurring at the sowing stage and 50% at the jointing stage; two planting density levels: row spacing 33cm (sowing density 0.9×10⁻⁶). 6 Plant / hm -2 Row spacing 25cm (sowing density 1.2×10⁻⁶) 6 Plant / hm 2 The experiment was repeated once, with a total of 96 test plots, each plot measuring 1 square meter. 2 (1m×1m).

[0057] The specific method for acquiring multispectral images of wheat was as follows: The UAV flew at an altitude of 20 meters and a speed of 3 meters per second. After acquiring the UAV images, Agisoft Photoscan Professional software was used for image stitching, geometric correction, and radiometric calibration. Six circular plates with a radius of 25 centimeters were used as ground control points (GCPs) for geometric correction. Radiometric calibration was performed using a 1m² area... 2 A rectangular gray panel with a reflectance of 8% converts image grayscale values ​​into reflectance.

[0058] S102: A drone equipped with a Riegl miniVUX-3 lidar system was used to acquire wheat lidar point clouds. Specifically, the drone flew at an altitude of 15 meters and a speed of 3 meters per second. Data processing software LiDAR360 was used for preprocessing, including data stitching, calibration, noise reduction, ground point cloud separation, and point cloud height normalization.

[0059] S103: Obtain the field-measured chlorophyll content of wheat canopy. The specific method was as follows: using a Dualex scientific chlorophyll meter (Force-A, Orsay, France), 10 leaves were randomly selected from each leaf layer (upper, middle, and lower) of each plot for measurement (measurement positions were 1 / 3, 1 / 2, and 2 / 3 of the distance from the leaf base, respectively), and the average value was calculated. Then, using a previously proposed model, the Dualex readings of the crop leaves were converted into chlorophyll content, as shown in the following formula:

[0060] LCC value =1.49 * Dualex value -3.60

[0061] Among them, LCC value Represents leaf chlorophyll content (μg / cm³) 2 Dualex value This represents the Dualex reading.

[0062] This invention uses a plant canopy analyzer, LAI-2200C (LI-COR Inc., Lincoln, NE, USA), to obtain the leaf area index (LAI) of crops. Measurements are taken in the late afternoon when sunlight is diffuse, and the lens is covered with a 45° viewing angle cover to avoid operator interference with the light. For each plot, two adjacent rows of crops, each 1 meter long, are selected. One A value is measured above this position using the LAI-2200C, and four B values ​​are then measured evenly along the diagonal on the ground below. These values ​​are used to calculate the LAI for that measurement. To eliminate random errors during measurement, a representative area in the middle of each plot is selected, and the measurement is repeated three times.

[0063] Finally, the product of the average chlorophyll content of crop leaves and the canopy LAI in each plot was taken as the canopy chlorophyll content (CCC (g / m²)) of that plot. 2 ))

[0064] S104: Wheat point cloud and spectrum are fused using a multidimensional matrix fusion method. The specific method is as follows: First, the LiDAR point cloud is exported as a Digital Surface Model (DSM), with the DSM pixel size matching the multispectral image pixel size (1 cm). Using the DSM as a reference plane, the multispectral image is spatially registered with it. The overall registration accuracy is less than 0.5 cm (half a pixel). Then, the LiDAR point cloud is divided into a matrix of columns (0.01 × 0.01 × 2 m). 3 The point cloud coordinates are horizontally aligned with the multispectral pixel grid. The highest point in each column is selected, and other points are excluded. The highest point in each column is projected onto the multispectral image, and the point cloud coordinates (x, y, z) are fused with the reflectance of the image's six bands (b1, b2, b3, b4, b5, b6) using a multidimensional matrix. Finally, the point cloud projection is removed to restore its three-dimensional structure, completing the fusion of point cloud and spectral reflectance. Figure 2 ).

[0065] S105: Twelve commonly used vegetation indices suitable for estimating chlorophyll in the canopy were calculated using six bands of UAV multispectral imagery. Then, linear regression models were established between these 12 vegetation indices and wheat canopy chlorophyll. Finally, the coefficient of determination (R²) was used to calculate the chlorophyll content. 2 The optimal vegetation index CI was selected. green The specific formula is as follows:

[0066] CI green =R 850 / R 530 -1

[0067] Among them, R 530 and R 850 These represent the reflectance at 530 and 850 nm in the multispectral image, respectively.

[0068] S106: Optimal vegetation index CI green Establish a linear regression model with canopy chlorophyll ( Figure 3 The specific model formula is as follows:

[0069] y = 0.37x - 0.57

[0070] Where y represents the canopy chlorophyll content (g / m²) 2 ), where x represents the vegetation index CI. green .

[0071] S107: Obtain point cloud data that incorporates multispectral reflectance, which is wheat multispectral point cloud data.

[0072] S108: Construct a vegetation index CI from the spectral reflectance of each point in the crop multispectral point cloud. green Then, a linear regression model of canopy chlorophyll content was applied to each point to generate the three-dimensional spatial distribution of wheat canopy chlorophyll. Figure 4 ).

[0073] S109: Based on the three-dimensional spatial distribution of wheat canopy chlorophyll, the point cloud was vertically layered at 0.05m height intervals, and the average canopy chlorophyll value for each layer was calculated. A linear regression model was established between these statistical parameters of canopy chlorophyll (5%, 25%, 50%, 75%, and 95% quantiles, maximum, minimum, mean, and standard deviation) and yield, and the coefficient of determination (R²) was selected. 2 The largest 75th percentile (CCC_P) 75th It is used as a new three-dimensional indicator to estimate output.

[0074] S110: Add the new three-dimensional index CCC_P 75th A linear regression model was established with wheat yield, and its accuracy was validated. CCC_P values ​​obtained at the heading, flowering, and grain-filling stages were analyzed. 75th To estimate crop yield, and through the coefficient of determination (R²) 2 The results are comprehensively evaluated using the root mean square error (RMSE) and the relative root mean square error (rRMSE), as shown in the following formula:

[0075]

[0076] Where y i Let y represent the measured value of sample i. i ′ represents the estimated value of sample i. This represents the average of all measurements, and n represents the total number of samples.

[0077] like Figure 5 The R-values ​​of the verification results 2 The accuracy was 0.65-0.71, RMSE was 0.39-0.43 t / ha, and rRMSE was 17-19.1%, with the best estimation accuracy, especially during the grouting period (R). 2 =0.71, RMSE=0.39t / ha, rRMSE=17.00%).

[0078] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the above embodiments do not limit the scope of protection of the present invention in any way, and all technical solutions obtained by equivalent substitution or other means fall within the scope of protection of the present invention. Parts not covered in this invention are the same as or can be implemented using existing technology.

Claims

1. A novel three-dimensional index construction method for estimating crop yield based on point cloud and spectral fusion, characterized in that... Includes the following steps: Step 1: Collect multispectral images of crops using a drone equipped with a multispectral camera; Step 2: Collect 3D point cloud data of crops using a drone equipped with a lidar. Step 3: Obtain crop canopy chlorophyll data; Step 4: Fuse the crop point cloud with the spectrum using a multi-dimensional matrix fusion method; Step 5: Obtain multispectral reflectance and construct a vegetation index for estimating canopy chlorophyll; Step 6: Establish a linear regression model between vegetation index and crop canopy chlorophyll; Step 7: Obtain crop multispectral point cloud; Step 8: Apply the crop canopy chlorophyll linear regression model to the multispectral point cloud to generate the three-dimensional spatial distribution of crop canopy chlorophyll; Step 9: Vertically stratify the crop canopy chlorophyll to obtain the average chlorophyll content of each layer, and perform statistical analysis to obtain the quantiles of these chlorophyll contents; Step 10: Obtain the quantiles of the new three-dimensional index chlorophyll content and establish a linear regression model with crop yield, and verify the accuracy.

2. The method for constructing a new three-dimensional index for estimating crop yield based on point cloud and spectral fusion as described in claim 1, characterized in that, In step 1, a drone equipped with an AIRPHEN multispectral camera is used to acquire multispectral images of crops; in step 2, a drone equipped with a Riegl miniVUX-3 lidar system is used to acquire lidar point clouds of crops.

3. The method for constructing a new three-dimensional index for estimating crop yield based on point cloud and spectral fusion according to claim 1, characterized in that, In step 3, the field-measured chlorophyll content of the canopy is obtained. Specifically, a certain number of leaves are randomly selected from each leaf layer of each plot using a Dualex scientific chlorophyll meter for measurement, and the average value is calculated. Then, the Dualex readings of the crop leaves are converted into chlorophyll content using the following formula: in, Represents leaf chlorophyll content, in μg / cm³ 2 ; This represents the Dualex reading.

4. The method for constructing a new three-dimensional index for estimating crop yield based on point cloud and spectral fusion according to claim 1, characterized in that, In step 4, the method for fusing crop point clouds and spectra is as follows: First, the LiDAR point cloud is exported as a Digital Surface Model (DSM), and the pixel size of the DSM is consistent with the pixel size of the multispectral image. Using the DSM as the reference plane, multispectral images are spatially registered with it; Then, the lidar point cloud is divided into a matrix composed of columns that coincides horizontally with the multispectral pixel grid. The highest point in each column is selected, and other points are excluded. The highest point in each column is projected onto the multispectral image, so that the coordinates (x, y, z) of the point cloud are fused with the reflectivity of multiple bands of the image in a multidimensional matrix. Finally, the point cloud projection is removed to restore its three-dimensional structure, thus completing the fusion of point cloud and spectral reflectance.

5. The method for constructing a new three-dimensional index for estimating crop yield based on point cloud and spectral fusion according to claim 1, characterized in that, In step 5, the band reflectance of the UAV multispectral image is used to calculate a vegetation index suitable for estimating the chlorophyll content of the crop canopy.

6. The method for constructing a new three-dimensional index for estimating crop yield based on point cloud and spectral fusion according to claim 1, characterized in that, In step 6, a linear regression model is established between the constructed vegetation index and the canopy chlorophyll content; the specific model formula is as follows: y = 0.37x - 0.57 Where y represents the canopy chlorophyll content, in g / m². 2 ; x represents the vegetation index CI green .

7. The method for constructing a new three-dimensional index for estimating crop yield based on point cloud and spectral fusion according to claim 1, characterized in that, In step 7, the crop multispectral point cloud is the point cloud data that incorporates the crop multispectral reflectance.

8. The method for constructing a new three-dimensional index for estimating crop yield based on point cloud and spectral fusion according to claim 1, characterized in that, In step 8, the spectral reflectance of each point in the crop multispectral point cloud is used to construct a vegetation index, and then a linear regression model of canopy chlorophyll content is applied to each point to generate the three-dimensional spatial distribution of crop canopy chlorophyll.

9. A method for constructing a new three-dimensional index for estimating crop yield based on point cloud and spectral fusion as described in claim 1, characterized in that, In step 9, based on the three-dimensional spatial distribution of chlorophyll in the crop canopy, the point cloud is vertically layered at 0.05m height intervals, and the average chlorophyll content of each layer is calculated. The 75th percentile of these chlorophyll contents, i.e., CCC_P, is obtained through statistical analysis. 75th .

10. A method for constructing a new three-dimensional index for estimating crop yield based on point cloud and spectral fusion according to claim 1, characterized in that, In step 10, a linear regression model is established between the 75th percentile of the vertical stratification of canopy chlorophyll and yield. The accuracy of the model is then validated using independent year data, with the coefficient of determination R0 being used. 2 The results are comprehensively evaluated using the root mean square error (RMSE) and the relative root mean square error (rRMSE), as shown in the following formula: in This represents the measurement value of sample i. This represents the estimated value of sample i. This represents the average of all measurements, and n represents the total number of samples.

Images