CN-122024052-A - Continuous space-time dynamic analysis and community population quantitative evaluation method and system for soybean individual plant field phenotype

CN122024052ACN 122024052 ACN122024052 ACN 122024052ACN-122024052-A

Abstract

The invention provides a continuous space-time dynamic analysis method and system for soybean individual plant field phenotypes, which solve the problems that individual plant heterogeneity is ignored in group phenotype evaluation in field crop phenotype data extraction, individual plant segmentation is restricted by canopy constraint, and unmanned aerial vehicle phenotype data accuracy is defective. The method comprises the steps of carrying out full-growth period data acquisition through an unmanned aerial vehicle carrying multiple sensors, carrying out multi-source space-time registration based on ground mark points, identifying single plants from seedling images, dividing the single plants to obtain independent single plant images, fusing visible light, multispectral and laser radar data, extracting single plant canopy height, coverage and normalized vegetation indexes, and constructing a multi-level evaluation system by calculating variation coefficients of single plant phenotype parameters in a group so as to realize dynamic quantitative evaluation from single plant heterogeneity analysis to group growth uniformity.

Inventors

ZHAO ZHENQING
YANG MINGLIANG
FAN ZHIHUI
CHEN QINGSHAN
LI XIAO
Feng Jiaao
LI XIAOQIAN
ZHAO XIANGYU

Assignees

东北农业大学

Dates

Publication Date: 20260512
Application Date: 20260127

Claims (10)

1. A continuous space-time dynamic analysis and community group quantitative evaluation method for soybean individual plant field phenotype is characterized by comprising the following steps: Carrying a visible light camera, a multispectral camera and a laser radar camera through an unmanned aerial vehicle platform, acquiring remote sensing data of a target field according to preset unified flight parameters in the whole growth period of crops, arranging a plurality of ground mark points in the field, and simultaneously acquiring longitude and latitude coordinates of the ground mark points and the boundary of a target cell by using RTK measuring equipment; Performing space-time registration on visible light data, multispectral data and laser radar data acquired in the whole growth period based on longitude and latitude coordinates of the ground marker points, and performing standardized preprocessing on various registered data; Generating a cell vector image layer based on longitude and latitude coordinates of the boundary of the target cell, and performing batch space clipping on the standardized preprocessed visible light data, multispectral data and laser radar data by utilizing the cell vector image layer to obtain cell image data; step four, based on a trained soybean seedling emergence recognition model, recognizing the position of a single plant from the seedling stage visible light data of the image data of the cell, obtaining the pixel coordinates of the mass center of the single plant, generating a mask based on the coordinates, and dividing to obtain the independent image of the single plant; extracting phenotype parameters of each individual plant based on the individual plant independent images, wherein the phenotype parameters comprise individual plant canopy height extracted based on laser radar data, individual plant canopy coverage extracted based on visible light data and normalized vegetation index calculated based on multispectral data; Step six, calculating the variation coefficient of all individual plant phenotype parameters in the cell according to each growth period, and establishing a multi-level evaluation system according to the canopy height, canopy coverage and variation coefficient of normalized vegetation index, wherein the multi-level evaluation system is used for evaluating the uniformity of soybean population growth in each period.
2. The method according to claim 1, wherein, in step one, The unified flight parameters comprise flight height, flight speed, transverse overlapping rate and longitudinal overlapping rate, and before data acquisition, the standard color card is used for carrying out color correction on the visible light camera, and the standard reflecting plate is used for carrying out radiation calibration on the multispectral camera.
3. The method according to claim 1, wherein, in step two, The standardized preprocessing specifically comprises the steps of splicing images with overlapping areas into orthographic images with field dimensions based on a feature matching algorithm for visible light data and multispectral data, and removing outlier noise points for laser radar data by adopting a method of combining statistical filtering and downsampling structure optimization, wherein the rule of the statistical filtering is that when the deviation of the elevation value of a certain point and the elevation mean value of a neighborhood point exceeds 2 times of standard deviation, the point is judged to be the noise point and is removed.
4. The method according to claim 1, wherein the fourth step is specifically: Step four, processing visible light data at a seedling stage by using a crop seedling emergence recognition model, and outputting normalized boundary frame coordinates of a single plant; Converting the normalized boundary frame coordinates into four-vertex coordinates under an image pixel coordinate system; And fourthly, performing arithmetic average on the four-vertex coordinates, and calculating to obtain pixel coordinates of the mass center of the single plant, wherein the calculation mode of the pixel coordinates of the mass center of the single plant is as follows: Wherein, the 、 Is the centroid pixel coordinate; 、 is the vertex pixel coordinate; and fourthly, generating a single plant region mask by using a segmentation algorithm based on texture and color characteristics based on pixel coordinates of the single plant centroid, and further segmenting to obtain independent single plant images.
5. The method according to claim 1, wherein in step five, The calculation mode of the canopy height of the single plant is as follows; Wherein, the Is the height of the canopy of the single plant, Is the height of crops in a single plant area, Is bare ground height.
6. The method according to claim 1, wherein in step five, The single plant canopy coverage calculating mode is as follows: wherein CC is canopy coverage; the number of pixels identified as vegetation; is the total number of pixels in a single cell.
7. The method according to claim 1, wherein in step five, The normalized vegetation index is calculated by: the NDVI is normalized vegetation index, the NIR is reflectivity value of near infrared band, and the Red is reflectivity value of Red band.
8. The method according to claim 1, wherein in step six, The multistage evaluation system specifically comprises the steps of judging that the population growth is uniform if all the canopy height, canopy coverage and variation coefficients of normalized vegetation indexes are smaller than a first threshold value, judging that the population growth is uniform if all the variation coefficients of all the indexes are smaller than a second threshold value and at least one index is larger than or equal to the first threshold value, judging that the population growth is non-uniform if the variation coefficient of any index is larger than or equal to the second threshold value, wherein the first threshold value is 10% and the second threshold value is 20%.
9. The method of claim 1, wherein step six further comprises: analyzing the relationship between the individual plant development dynamics and the colony interaction by analyzing the dynamic change track of the colony growth uniformity in the whole growth period.
10. A continuous space-time dynamic analysis and community population quantification evaluation system for soybean individual plant field phenotypes, which is characterized by comprising: The data acquisition and geographic information acquisition module carries a visible light camera, a multispectral camera and a laser radar camera through an unmanned plane platform, carries out remote sensing data acquisition on a target field block according to preset unified flight parameters in the whole growth period of crops, a plurality of ground mark points are distributed in the field, and meanwhile, longitude and latitude coordinates of the ground mark points and the boundary of a target cell are obtained by using RTK measuring equipment; The data processing and registering module is used for carrying out space-time registration on visible light data, multispectral data and laser radar data acquired in the whole growth period based on longitude and latitude coordinates of the ground mark points, and carrying out standardized preprocessing on various registered data; The data clipping module generates a cell vector image layer based on longitude and latitude coordinates of the boundary of the target cell, and performs batch space clipping on the standardized preprocessed visible light data, multispectral data and laser radar data by utilizing the cell vector image layer to obtain cell image data; The individual plant identification and segmentation module is used for identifying the position of an individual plant from the seedling stage visible light data of the district image data based on the trained soybean seedling emergence identification model, acquiring the pixel coordinates of the mass center of the individual plant, generating a mask based on the coordinates, and segmenting to obtain an individual plant independent image; A phenotype parameter extraction module for respectively extracting phenotype parameters of each individual plant based on the individual plant independent images, wherein the phenotype parameters comprise individual plant canopy height extracted based on laser radar data, individual plant canopy coverage extracted based on visible light data and normalized vegetation index calculated based on multispectral data; The group evaluation and analysis module calculates the variation coefficient of all individual plant phenotype parameters in the cell according to each growth period, and establishes a multi-stage evaluation system according to the canopy height, canopy coverage and variation coefficient of the normalized vegetation index, and is used for evaluating the uniformity of soybean group growth in each period.

Description

Continuous space-time dynamic analysis and community population quantitative evaluation method and system for soybean individual plant field phenotype Technical Field The invention relates to the technical field of intelligent breeding, in particular to a soybean single plant phenotype continuous space-time dynamic analysis and community group quantitative evaluation method and system. Background In intelligent breeding and precise agricultural research, the method is a key basis for realizing acceleration of fine variety breeding and implementation of fine cultivation management, and the method is used for obtaining the phenotypic data of field crops in a high-throughput, lossless and precise manner. Unmanned aerial vehicle remote sensing technology has become the important instrument of field crop phenotype high flux acquisition with its nimble, high-efficient, advantage that spatial resolution is high. However, the existing unmanned aerial vehicle platform-based phenotype acquisition and evaluation technology still has the following defects in practical application, especially in the scale of a breeding cell of close-planted crops: The population phenotype evaluation ignores single plant heterogeneity, and the current mainstream unmanned aerial vehicle phenotype analysis method mostly regards the whole breeding cell as a homogenized integral unit, and evaluates the whole breeding cell by extracting the average spectrum index or integral canopy structure parameter of the cell. The method is efficient, but completely covers the space-time heterogeneity of individual plants in the cell in growth vigor, plant height, biomass and the like. The individual difference is a direct expression of population competition, microenvironment variation and genotype difference, and has important indication significance for breeding selection and cultivation regulation. Neglecting population scale evaluation of individual heterogeneity creates systematic bias between phenotype data and true genetic performance and accurate agronomic decision requirements. Crown canopy closure restricts the realization of single plant segmentation, and for high-density planted crops such as soybeans, the canopy is rapidly closed in the middle and later stages of growth, and leaves among plants are mutually staggered and adhered seriously. The method leads to the common technical problem of single plant boundary blurring and high feature overlapping when single plant segmentation is carried out based on the visible light or the spectral image of the unmanned aerial vehicle. The traditional image processing methods such as threshold segmentation and edge detection fail in the scene, and high-throughput and high-precision single plant separation and identification are difficult to realize, so that a technical path for carrying out phenotype analysis from a single plant scale is blocked. The unmanned aerial vehicle phenotype data accuracy defect, unmanned aerial vehicle flight attitude change, sensor self error, environment illumination fluctuation and other factors all can introduce noise, influence the accuracy and the comparability of data for the phenotypic parameter precision of extraction is difficult to stably reach the quantization standard required by breeding and cultivation management. In summary, in the prior art, a continuous space-time dynamic analysis method capable of breaking through the canopy closure limit to achieve accurate identification of a single plant, integrating multi-source remote sensing data to improve phenotype extraction precision and finally achieving analysis from heterogeneity of the single plant to scientific evaluation of population uniformity is needed. Disclosure of Invention The invention provides a continuous space-time dynamic analysis method and system for soybean individual plant field phenotypes, which aims to solve the problems that in the prior art, individual plant heterogeneity is ignored in population phenotype evaluation, individual plant segmentation is restricted by canopy constraint, and unmanned aerial vehicle phenotype data accuracy is defective. According to the invention, through integrating the unmanned aerial vehicle multi-platform sensor data and combining with the computer vision and geospatial analysis technology, the full-flow analysis from the precise identification of single plants, the extraction of phenotype parameters and the dynamic evaluation of the group growth uniformity is realized, and the method comprises the following steps: Carrying a visible light camera, a multispectral camera and a laser radar camera through an unmanned aerial vehicle platform, acquiring remote sensing data of a target field according to preset unified flight parameters in the whole growth period of crops, arranging a plurality of ground mark points in the field, and simultaneously acquiring longitude and latitude coordinates of the ground mark points and the boundary of a target cell by using RTK