Method for estimating the seed setting rate of multi-varieties of rice based on dynamic multi-spectral vegetation index

Abstract

This invention provides a method for estimating the seed setting rate of multiple rice varieties based on dynamic multispectral vegetation indices. The method includes: collecting image information of different rice varieties from heading to maturity using a UAV multispectral camera; constructing a digital surface model (DSM) to obtain rice plant height information; and calculating the vegetation index (VI) using data from each band. By analyzing the relative plant height and VI of the rice canopy at a single time point and combining this with an ordered set of images from heading to maturity, a dataset of the variation in the population photosynthetic efficiency (Pn) of different rice varieties from heading to maturity is established. The dynamics of Pn are fitted, and the parameters and maximum and minimum values ​​of the fitted curve are extracted. The mean value of Pn from heading to maturity is then fused, and a machine learning algorithm is used to estimate the seed setting rate of multiple rice varieties. This invention can be applied to multiple rice varieties, using dynamic fitting and machine learning to estimate the seed setting rate of rice by dynamically varying the population photosynthetic efficiency (Pn) from heading to maturity.

Description

Technical Field

[0001] This invention relates to the field of agricultural remote sensing technology, and in particular to a method for estimating the seed setting rate of multiple rice varieties based on dynamic multispectral vegetation index. Background Technology

[0002] The seed setting rate of rice is of great significance to yield because it directly affects the yield per unit area. A high seed setting rate means that the plant can utilize nutrients and growing conditions more efficiently, producing more grains and thus increasing yield.

[0003] In recent years, low-altitude unmanned aerial vehicles (UAVs) equipped with various sensors have been able to provide high-resolution images, particularly utilizing RGB, multispectral, and hyperspectral sensors. Currently, UAV remote sensing and digital surface models (DSMs) are widely used in rice plant height studies and other phenotypic research. However, research on estimating rice seed setting rate using UAV spectral data is still limited. Furthermore, due to differences between different rice varieties and significant variations in seed setting rate at different growth stages, applying this method to multiple rice varieties, especially in breeding work, remains a challenge. Summary of the Invention

[0004] In view of this, the purpose of this invention is to propose a method for estimating the seed setting rate of multiple rice varieties based on dynamic multispectral vegetation index, so as to realize the estimation of the seed setting rate of different rice varieties.

[0005] To achieve the above objectives, the technical solution of the present invention is as follows: a method for estimating the seed setting rate of multiple rice varieties based on dynamic multispectral vegetation index, comprising: Using a drone multispectral camera, images of different rice varieties from the heading stage to maturity were collected. A digital surface model (DSM) was constructed to obtain rice plant height information, and the vegetation index (VI) was calculated using data from each band. By analyzing the relative plant height and VI of the rice canopy at a single time point, and combining this with an ordered image set of rice from heading to maturity, a dataset of the variation of photosynthetic efficiency Pn of different rice varieties from heading to maturity was established, where Pn = a + DSM·VI 2 / (1+e -b·(VI - c) ), where a is a constant term that can provide a basic offset for Pn, b represents the influence coefficient of VI, and c is the threshold of VI; The dynamics of Pn are fitted, the parameters and maximum and minimum values ​​of the fitted curve are extracted, the mean of Pn from the beginning of heading to maturity is integrated, and the seed setting rate of multiple rice varieties is estimated using machine learning algorithms.

[0006] Furthermore, image information of different rice varieties from the heading stage to maturity was collected using a drone multispectral camera, including: Before rice was planted, an orthophoto of rice was taken using a drone camera. From the initial heading of a rice variety in the field until the rice matures and is harvested, orthophotos of the rice are collected using drone cameras once a week. The collected images were stitched together using DJI Terra to obtain an image of the entire field. Then, Python was used to mask and cut the entire field to obtain images of each variety plot.

[0007] A digital surface model (DSM) was constructed to obtain rice plant height information, and the vegetation index (VI) was calculated using data from various wavebands, including: Using Python to extract DSM and single-band data of rice from images; When processing data elevation model images from different periods, the height of the rice canopy is calculated based on the relative height of bare soil (without rice planting). The VI is obtained by further calculation of single-band data.

[0008] The dynamics of Pn are fitted, and the parameters and maximum and minimum values ​​of the fitted curve are extracted. The mean value of Pn from the beginning of heading to maturity is then integrated, including: The ordered dataset of Pn from the beginning of heading to maturity was fitted using a quadratic polynomial; Based on the dynamic fitting results of Pn, the goodness of fit R is selected. 2 The highest VI and its corresponding Pn; According to R 2 The highest dynamic curve of Pn is obtained, and the mean, maximum, minimum values ​​and curve parameters are extracted.

[0009] Estimating the seed setting rate of multiple rice varieties using machine learning algorithms, including: The seed setting rate of rice was estimated by using the XGBoost algorithm and combining the dynamic characteristics of Pn from the beginning of heading to maturity of different rice varieties.

[0010] Compared with the prior art, the advantages of the present invention are as follows: the collected images are stitched, masked and cut in sequence to realize the division of different rice varieties; Pn of each rice variety from the beginning of heading to maturity is calculated by DSM and single-band data; dynamic fitting is performed on them; and the seed setting rate of different rice varieties can be estimated by using the XGBoost algorithm and combining the dynamic characteristics of Pn from the beginning of heading to maturity. Attached Figure Description

[0011] Figure 1 This is a flowchart of the technical solution of the present invention; Figure 2 This is a field overview and a plotted plot after cutting, as described in this invention. Figure 3The R-value of Pn dynamic fitting based on each dynamic VI in this invention is... 2 and RMSE; Figure 4 This is the Pn dynamic fitting curve based on RVI of the present invention; Figure 5 The performance of the model of this invention is evaluated. Detailed Implementation

[0012] Typical embodiments embodying the features and advantages of the present invention will be described in detail in the following description. It should be understood that the present invention can have various variations in different embodiments without departing from the scope of the present invention, and the descriptions and illustrations herein are for illustrative purposes only and not intended to limit the present invention.

[0013] Taking a multi-variety rice trial (60 main varieties promoted in southern regions) as an example, images were acquired using drones equipped with multispectral sensors one week before transplanting, when the field was dry after tillage; in addition, drone flights were conducted once a week from the emergence of one variety in the field until all varieties matured.

[0014] Each collected image was stitched together using DJI Terra software, and then Python was used for cell segmentation, such as... Figure 2 As shown.

[0015] Furthermore, the digital surface model (DSM) and single-band data of each cell at each time point are extracted, and the vegetation indices VI and Pn are calculated.

[0016] The time series Pn of each cell was fitted using a quadratic polynomial, and the R² and RMSE of the Pn fitting based on each VI were obtained. It was found that the Pn fitting based on RVI had the best effect, such as... Figure 3 As shown, the fitting curves for each cell are as follows: Figure 4 As shown.

[0017] Subsequently, the mean, maximum, minimum values, and curve parameters of the dynamic fitting of each cell were extracted and used as input data for the seed set rate estimation.

[0018] The input data for estimating the seed setting rate was first standardized. Then, the standardized data (the dataset was divided into training and test sets in a 7:3 ratio) was combined with the XGBoost algorithm to estimate the seed setting rate of rice. To reliably evaluate the generalization ability of the model, 10x cross-validation was also used.

[0019] Finally, the model performance was evaluated using the coefficient of determination R² and the root mean square error RMSE, and the results are as follows: Figure 5 As shown.

[0020] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various variations or substitutions within the technical scope disclosed in the present invention, and these should all be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for estimating the seed setting rate of multiple rice varieties based on dynamic multispectral vegetation index, characterized in that, include: Using a multispectral camera from an unmanned aerial vehicle (UAV) to collect image information of different rice varieties from the beginning of heading to maturity, a digital surface model (DSM) was constructed to obtain the relative plant height of the rice canopy, and the vegetation index (VI) was calculated using data from each band. By analyzing the relative plant height and VI of the rice canopy at a single time point, and combining this with an ordered image set of rice from heading to maturity, a dataset of the variation of photosynthetic efficiency Pn of different rice varieties from heading to maturity was established, where Pn = a + DSM·VI 2 / (1+e -b·(VI - c) ), where a is a constant term that can provide a basic offset for Pn, b represents the influence coefficient of VI, c is the threshold of VI, and DSM represents the relative height of the rice canopy. The dynamics of Pn are fitted, the parameters and maximum and minimum values ​​of the fitted curve are extracted, the mean of Pn from the beginning of heading to maturity is integrated, and the seed setting rate of multiple rice varieties is estimated using machine learning algorithms.

2. The method for estimating the seed setting rate of multiple rice varieties based on dynamic multispectral vegetation index according to claim 1, characterized in that, Image information of different rice varieties from the heading stage to maturity was collected using a drone's multispectral camera, including: Before rice was planted, an orthophoto of rice was taken using a drone camera. From the initial heading of a rice variety in the field until the rice matures and is harvested, orthophotos of the rice are collected using drone cameras once a week. DJI Terra was used to stitch the collected images to obtain an image of the entire field. Then, Python was used to mask and cut the entire field to obtain an image of each variety plot.

3. The method for estimating the seed setting rate of multiple rice varieties based on dynamic multispectral vegetation index according to claim 1, characterized in that, A digital surface model (DSM) was constructed to obtain the relative plant height of the rice canopy, and the vegetation index (VI) was calculated using data from various wavebands, including: Using Python to extract DSM and single-band data of rice from images; When processing data elevation model images from different periods, the relative height of the rice canopy is calculated based on the relative height of the bare soil. The VI is obtained by further calculation of single-band data.

4. The method for estimating the seed setting rate of multiple rice varieties based on dynamic multispectral vegetation index according to claim 1, characterized in that, The dynamics of Pn are fitted, and the parameters and maximum and minimum values ​​of the fitted curve are extracted. The mean value of Pn from the beginning of heading to maturity is then integrated, including: The ordered dataset of Pn from the beginning of heading to maturity was fitted using a quadratic polynomial; Based on the dynamic fitting results of Pn, the goodness of fit R is selected. 2 The highest VI and its corresponding Pn; According to R 2 The highest dynamic curve of Pn is obtained, and the mean, maximum, minimum values ​​and curve parameters are extracted.

5. The method for estimating the seed setting rate of multiple rice varieties based on dynamic multispectral vegetation index according to claim 1, characterized in that, Estimating the seed setting rate of multiple rice varieties using machine learning algorithms, including: The seed setting rate of rice was estimated by using the XGBoost algorithm and combining the dynamic characteristics of Pn from the beginning of heading to maturity of different rice varieties.

Images