A method and device for identifying the traits of safflower DUS test leaves

Abstract

This invention proposes a method and apparatus for leaf trait discrimination in safflower DUS testing. The method includes: acquiring image data of the first and sixth leaves of safflower, constructing a cross-scale three-dimensional reconstruction model, and generating a three-dimensional morphological feature set; acquiring dynamic growth data of safflower and generating a stability discrimination parameter set; constructing an association network using genome-phenome joint analysis, fusing chloroplast genome SNP data and lateral vein density variation coefficients to generate multimodal association feature vectors; classifying using a Transformer+CNN hybrid architecture model, generating classification results and statistical significance values ​​using a dynamic hyperparameter tuning module; and generating a comprehensive discrimination conclusion for DUS testing through a three-dimensional validation system. This invention uses cross-scale three-dimensional reconstruction technology to quantify leaf characteristics, combines a time-series three-dimensional scanning system to capture growth data, and solves the problems of traditional manual observation; it uses a hybrid architecture model for classification and introduces a tuning module to improve accuracy.

Description

Technical Field

[0001] This invention relates to the field of plant trait discrimination technology, and in particular to a method and apparatus for discriminating leaf traits using the DUS test for safflower. Background Technology

[0002] In current plant DUS testing, leaf trait identification mainly relies on traditional morphological analysis and basic image processing techniques. The industry commonly uses manual visual inspection combined with simple measuring tools (such as calipers and protractors) to record phenotypic parameters such as leaf margin serrations and vein distribution, supplemented by two-dimensional image software (such as Photoshop and ImageJ) for basic quantification of area, perimeter, etc. Some research institutions have begun to apply 3D scanning technology to obtain leaf contour data, but this is mostly limited to static modeling at a single time point, and dynamic growth monitoring still relies primarily on periodic manual sampling. At the data analysis level, classic statistical methods such as analysis of variance (ANOVA) and principal component analysis (PCA) are widely used. Genomics-phenomic association analysis is still in its early stages of exploration, and deep integration of chloroplast genome SNP data with phenotypic characteristics is less common. The overall technical system exhibits a characteristic of "emphasizing morphological description while neglecting dynamic association."

[0003] Traditional methods rely on manual observation and measurement, which suffer from high subjectivity and poor repeatability. Different observers may make different judgments on key indicators such as the leaf margin serration depth / width ratio and leaf vein branching angle, resulting in insufficient reliability of test results. Summary of the Invention

[0004] This invention aims to at least solve the technical problem of insufficient reliability of test results in the prior art, and innovatively proposes a method and device for judging the leaf characteristics of safflower DUS test.

[0005] To achieve the above-mentioned objectives of this invention, this invention provides a method for identifying leaf traits in safflower using the DUS test, the method comprising: S1. Collect image data of the first and sixth leaves of the safflower; S2. Based on the image data of the first leaf and the sixth leaf, construct a cross-scale three-dimensional reconstruction model and generate a three-dimensional morphological feature set including the leaf margin serration depth / width ratio and the standard deviation of the leaf vein branch angle. S3. Based on the three-dimensional morphological feature set, dynamic growth data from the seedling stage to maturity of safflower are collected through a time-series three-dimensional scanning system, and a stability discrimination parameter set is generated using the dynamic model of leaf vein branch angle change rate and leaf area index. S4. Based on the three-dimensional morphological feature set and stability discrimination parameter set, a genotype-phenotype association network is constructed using genome-phenome joint analysis, and a multimodal association feature vector is generated by fusing chloroplast genome SNP data and lateral vein density variation coefficient through cosine similarity algorithm. S5. Based on the multimodal associated feature vector, a hybrid architecture model based on Transformer+CNN is used for feature alignment and classification. The dynamic hyperparameter tuning module is used to generate the classification results of leaf margin entire / serrated / deeply lobed and the statistical significance value of leaf vein density difference. S6. Based on the classification results and statistical significance values, the specificity, consistency and stability discrimination logic is executed through the three-dimensional verification system to generate a comprehensive discrimination conclusion for the DUS test.

[0006] In another aspect, the present invention also provides a safflower DUS test leaf morphology discrimination device, the system comprising: processor; Memory used to store processor-executable instructions; The processor is configured to implement the safflower DUS test leaf morphology discrimination method when executing the executable instructions.

[0007] The beneficial effects of this invention are as follows: This invention achieves precise quantification of leaf morphological characteristics through cross-scale three-dimensional reconstruction technology, and dynamically captures growth change data from seedling stage to maturity using a time-series three-dimensional scanning system, effectively solving the technical bottlenecks of traditional manual observation, which is characterized by strong subjectivity and poor repeatability. By constructing a deep association network through genome-phenome joint analysis, and integrating chloroplast genome SNP data with phenotypic features, it reveals the genetic basis of phenotypic differences at the molecular level, overcoming the limitations of the superficial phenotypic-genotype association analysis in traditional methods. A Transformer+CNN hybrid architecture model is used for feature alignment and classification, and a dynamic hyperparameter tuning module is introduced to improve the accuracy of feature extraction and the scientific validity of classification results. This method, through multi-dimensional data fusion and dynamic growth process monitoring, significantly enhances the objectivity and reliability of the "specificity-consistency-stability" discrimination in DUS testing, providing more scientific and internationally compliant technical support for variety identification, and effectively improving the credibility of DUS test conclusions.

[0008] 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

[0009] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which: Figure 1This is a flowchart of a method for identifying leaf characteristics using the DUS test in safflower according to the present invention. Detailed Implementation

[0010] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0011] Example 1 like Figure 1 As shown, a method for identifying leaf traits in safflower using the DUS test, the method comprising: S1. Collect image data of the first and sixth leaves of the safflower; In step S1, it is necessary to explain in detail that, firstly, safflower plants with consistent growth status and free from pests, diseases, and mechanical damage should be selected as test samples. The first leaf specifically refers to the first true leaf that grows after the cotyledons have fully unfolded (seedling stage, seedling height about 5-8cm), and the sixth leaf specifically refers to the sixth functional leaf that unfolds after the plant enters the mature stage (the stage when the leaf area reaches its maximum value and the morphology is stable). Secondly, a high-resolution digital camera equipped with a 50-megapixel CMOS sensor and a macro lens is used, combined with a white backlight and a standard calibration ruler (accuracy 0.01mm), under diffused natural light conditions (light intensity controlled at 800-1200 lux), with the leaf's main vein perpendicular to the lens, to ensure that the leaf is fully presented in the field of view without shadows. Finally, the acquired raw images are preprocessed, including removing noise through Gaussian filtering, extracting leaf areas through adaptive threshold segmentation, and using the calibration ruler to complete the pixel-to-actual size calibration conversion, generating standardized image data containing the complete outline of the leaf, texture information, and size calibration parameters.

[0012] S2. Based on the image data of the first leaf and the sixth leaf, construct a cross-scale three-dimensional reconstruction model and generate a three-dimensional morphological feature set including the leaf margin serration depth / width ratio and the standard deviation of the leaf vein branch angle. S3. Based on the three-dimensional morphological feature set, dynamic growth data from the seedling stage to maturity of safflower are collected through a time-series three-dimensional scanning system, and a stability discrimination parameter set is generated using the dynamic model of leaf vein branch angle change rate and leaf area index. S4. Based on the three-dimensional morphological feature set and stability discrimination parameter set, a genotype-phenotype association network is constructed using genome-phenome joint analysis, and a multimodal association feature vector is generated by fusing chloroplast genome SNP data and lateral vein density variation coefficient through cosine similarity algorithm. S5. Based on the multimodal associated feature vector, a hybrid architecture model based on Transformer+CNN is used for feature alignment and classification. The dynamic hyperparameter tuning module is used to generate the classification results of leaf margin entire / serrated / deeply lobed and the statistical significance value of leaf vein density difference. S6. Based on the classification results and statistical significance values, the specificity, consistency and stability discrimination logic is executed through the three-dimensional verification system to generate a comprehensive discrimination conclusion for the DUS test.

[0013] The principle of a leaf trait discrimination method for safflower using DUS testing in this embodiment is as follows: First, high-resolution images of the first leaf during the seedling stage and the sixth leaf during the mature stage of safflower are acquired through standardized acquisition and preprocessing to provide a reliable data foundation for the accurate quantification of leaf morphological characteristics. Then, a cross-scale 3D reconstruction model is used to extract core phenotypic features such as the leaf margin serration depth / width ratio and the standard deviation of the leaf vein branch angle, achieving a 3D digital representation of the leaf morphology. Next, a time-series 3D scanning system is used to capture dynamic data throughout the entire growth cycle. A stability discrimination parameter set is constructed using a dynamic model of the leaf vein branch angle change rate and leaf area index to effectively exclude… This method addresses the interference of environmental factors on phenotypic observations. It then integrates chloroplast genome SNP data with phenotypic features using a genome-phenome joint analysis framework, employing a cosine similarity algorithm to generate multimodal association feature vectors, revealing the genetic basis of phenotypic differences at the molecular level. Subsequently, a Transformer+CNN hybrid architecture model is used for feature alignment and classification, incorporating a dynamic hyperparameter tuning module to improve classification accuracy. Statistical tests are used to quantify the significance of differences in leaf vein density. Finally, a three-dimensional validation system integrates specificity, consistency, and stability discrimination logic to generate a comprehensive DUS test conclusion that meets the requirements of the international UPOV guidelines. This method, through multi-dimensional data fusion and cross-scale dynamic monitoring, transforms the discrimination of safflower leaf traits from traditional subjective observation to digital, intelligent, and precise methods, significantly improving the objectivity and reliability of DUS testing.

[0014] As an optional embodiment of the present invention, optionally, in step S2, based on the image data of the first leaf and the sixth leaf, a cross-scale three-dimensional reconstruction model is constructed to generate a three-dimensional morphological feature set including the leaf margin serration depth / width ratio and the standard deviation of the leaf vein branch angle, including: S201. Based on the image data of the first blade and the sixth blade, multi-view optical slices and three-dimensional point cloud data are acquired by combining micro-CT and laser confocal scanning microscopy to generate an original three-dimensional dataset containing the blade surface contour and leaf vein network structure. In step S201, it is necessary to explain in detail that a high-resolution micro-CT system is used, with a scanning voltage of 50kV, a current of 100μA, and a scanning slice thickness of 5μm. Continuous tomographic scanning is performed on formaldehyde-fixed leaf samples to obtain a two-dimensional slice image sequence with an axial resolution of 3μm. Three-dimensional reconstruction is performed using NRecon software, and the slice sequence is denoised, thresholded, and stitched together to obtain volume data containing the internal leaf vein network (main vein, lateral veins, and fine vein branches) and the external contour of the leaf, with a voxel size of 2×2×5μm³. Simultaneously, a laser confocal microscope is used, selecting a 488nm argon ion laser to excite the autofluorescence of leaf epidermal cells. The scanning step size is set to 1μm and the lateral resolution to 0.5μm, and three-dimensional point cloud data of the micro-texture of the leaf surface (such as the curvature of serrated edges and the arrangement of epidermal cells) is acquired, with a point cloud density ≥1000 points / mm². The multi-view point clouds are registered and fused using LASX software to generate a three-dimensional model of the leaf surface with an accuracy of 1μm. Finally, the internal leaf vein data reconstructed by micro-CT and the surface point cloud model generated by laser confocal microscopy were imported into Geomagic DesignX software. Cross-scale data alignment was completed using leaf vein intersections and leaf edge serration vertices as feature matching points. The final result was an original 3D dataset containing macroscopic morphological contours (such as leaf aspect ratio), microscopic surface textures, and internal leaf vein network structures. The spatial resolution of the dataset covers the range from 1μm (microscopic) to 1mm (macroscopic).

[0015] S202. Based on the original three-dimensional dataset, a three-dimensional reconstruction model is generated by combining a three-dimensional reconstruction algorithm based on voxel color assignment with a sparse reconstruction algorithm based on calibration constraints, and optimizing the matching degree between leaf texture and contour through the SFNet feature descriptor network. In step S202, it is necessary to explain in detail that the internal leaf vein voxels reconstructed by micro-CT are mapped to depth color levels according to gray values ​​(0-255) (gray value ≤ 50 corresponds to blue, representing the deep structure of leaf veins; gray value ≥ 200 corresponds to red, representing the surface branches of leaf veins). At the same time, the surface point cloud obtained by laser confocal microscopy is mapped to texture color gradients according to fluorescence intensity (0-1000 a.u.) (intensity ≤ 300 corresponds to green, representing normal areas of epidermal cells; intensity ≥ 800 corresponds to yellow, representing high curvature areas of serrated edges), thus achieving color information fusion at the voxel level. Six non-coplanar calibration feature points from the original 3D dataset (including the intersection of the three scale lines of the calibration ruler, the two endpoints of the leaf midrib, and the vertex of the largest serration) are selected. Using the known actual coordinates of these points as constraints, an objective function for minimizing projection error is constructed, and the camera parameter matrix for sparse reconstruction is solved to control the reconstruction error within 2 μm. Next, the matching degree is optimized using the SFNet feature descriptor network: Local feature blocks from the original 3D dataset are input, and the SFNet network extracts multi-scale features through three convolutional layers, generating a 128-dimensional feature descriptor. The cosine similarity of the feature blocks from different viewpoints is calculated, and a similarity threshold of 0.95 is set. Regions with matching degrees below the threshold (such as misaligned edges of leaf serrations) are locally adjusted to correct alignment deviations between the contour and texture. Finally, the above processing results are integrated to generate a 3D reconstruction model, and the accuracy is verified using a standard leaf model: the reconstructed model is compared with a standard red flower leaf solid model of known size, and the root mean square error (RMSE) is calculated to ensure RMSE ≤ 0.5μm, thus completing the model's validity verification.

[0016] S203. Based on the three-dimensional reconstruction model, the leaf edge serrated area is segmented using the watershed algorithm and the serrated depth / width ratio is calculated. The straight vein features of the leaf vein are detected by Hough transform and the standard deviation of the main vein width and the branch angle of the lateral vein is calculated. The leaf edge contour curve is fitted by the least squares method in combination with the three-dimensional point cloud data to generate quantified three-dimensional morphological feature parameters. In step S203, it is necessary to explain in detail that firstly, the leaf margin region in the 3D reconstruction model is segmented. A watershed algorithm combined with local gradient information is used to extract the serrated boundaries, ensuring sub-pixel level segmentation accuracy. For each serrated unit, its maximum depth and base width are calculated, and the serration depth / width ratio is mapped to the [0,1] interval through standardization to reflect the complexity of the leaf margin morphology. Next, Hough transform is used to detect the straight vein features, extracting the midrib axis and its width distribution. Simultaneously, the branch angles of the lateral veins are statistically analyzed, and their standard deviation is calculated as a key parameter characterizing the stability of the vein network. Furthermore, based on 3D point cloud data, the leaf margin contour curve is fitted using the least squares method to extract curvature variation features, generating a quantitative 3D morphological feature parameter set containing the leaf margin serration depth / width ratio, the mean midrib width, and the standard deviation of the lateral vein branch angles. These parameters not only accurately describe the macroscopic morphological characteristics of the leaf but also effectively capture subtle differences in the microstructure.

[0017] S204. Based on the three-dimensional morphological feature parameters, the reconstruction accuracy is verified by comparing with manual measurement values ​​and the reconstruction parameters are dynamically adjusted to generate a three-dimensional morphological feature set that meets the quantification requirements of DUS testing.

[0018] In step S204, it is necessary to explain in detail that, to ensure the accuracy of the three-dimensional morphological feature parameters, a method of comparing and verifying the results of manual measurement and automated reconstruction is adopted. Specifically, 50 standard red safflower leaf samples are selected, and manual measurements are performed using a high-precision digital caliper and an optical microscope to obtain the actual values ​​of key parameters such as leaf margin serration depth, base width, and midrib width. Simultaneously, the corresponding parameters output by the three-dimensional reconstruction model of these samples are paired with the manually measured values ​​for analysis, calculating the mean absolute error (MAE) and relative error (RE). If the error exceeds a preset threshold (e.g., MAE > 0.1 mm or RE > 5%), the key parameters in the three-dimensional reconstruction algorithm are dynamically adjusted, including the gradient threshold of the watershed algorithm, the resolution of the Hough transform, and the number of iterations for least squares fitting, until the error meets the accuracy requirements. The final generated three-dimensional morphological feature set must contain at least 20 quantitative indicators, covering leaf margin serration morphology, leaf vein network structure, and overall contour features, and ensure good repeatability and consistency under different environmental conditions.

[0019] As an optional embodiment of the present invention, optionally, in step S3, based on the three-dimensional morphological feature set, dynamic growth data from the seedling stage to maturity of safflower are collected through a time-series three-dimensional scanning system, and a stability discrimination parameter set is generated using a dynamic model of leaf vein branch angle change rate and leaf area index, including: S301. Based on the three-dimensional morphological feature set, a time-series three-dimensional scanner equipped with a ring light source and a polarization filter is used to collect three-dimensional morphological data of leaves from seedling stage to maturity stage at n-day intervals, and environmental temperature, humidity and light intensity parameters are recorded simultaneously to generate an original growth dataset containing dynamic changes in leaf morphology and environmental correlation data. In step S301, a high-precision time-series 3D scanner is employed. This device is equipped with a ring LED light source and a polarization filter to reduce ambient light interference and improve the accuracy of data acquisition. The scanning interval is set to n days (typically ranging from 5 to 7 days), from the seedling stage of red flowers to the end of the maturity stage, ensuring coverage of key stages of leaf morphological change. During each scan, environmental parameters are recorded simultaneously, including air temperature (accuracy ±0.1℃), relative humidity (accuracy ±1%), and light intensity (accuracy ±5 lux). These data are stored in real time to the scanning system database via a built-in sensor module. During the scanning process, the leaf sample must be fixed on a rotating platform, which automatically rotates 360° in 0.5° increments. The scanner collects 3D point cloud data of the leaf surface and edges from multiple angles, with a point cloud density of no less than 5000 points / cm². After scanning, the accompanying software is used to register and fuse the multi-view point clouds, generating a complete 3D model of a single scan, and extracting basic morphological parameters such as leaf length, width, and area. Finally, the 3D models at each time point are integrated with the corresponding environmental parameters to form an original growth dataset containing information on the dynamic growth trajectory of leaves and the correlation with environmental factors.

[0020] S302. Based on the original growth dataset, the leaf vein branch angle change rate model is combined with the leaf area index dynamic model to calculate the leaf vein branch angle change rate, leaf area index coefficient of variation and growth rate standard deviation, and generate intermediate stability quantification parameters. In step S302, it is necessary to explain in detail that, firstly, a time series analysis is performed on the leaf vein branching angles in the original growth dataset. By calculating the ratio of the angle difference between adjacent time points to the time interval, the rate of change of leaf vein branching angles is obtained. This parameter reflects the dynamic adjustment ability of the leaf vein network structure during leaf growth. Next, based on the leaf area index dynamic model, leaf area data at each time point is extracted, and its coefficient of variation is calculated to assess the stability during leaf expansion. Simultaneously, the growth rate is calculated using the ratio of the increase in leaf length and width to the time interval, and its standard deviation is further obtained as an indicator of growth uniformity. These intermediate stability quantification parameters together constitute the basic dataset describing the dynamic characteristics of safflower leaf growth.

[0021] The aforementioned dynamic leaf area index (LAI) model is a mathematical model based on nonlinear regression analysis. Its core lies in revealing the dynamic patterns during the growth process by fitting a curve showing the change in leaf area over time. Specifically, the model uses a Logistic growth function as its basic framework, combined with weighting coefficients of environmental factors, to predict and correct the LAI. The model expression is: Y(t) = K / (1 + exp(-r(t-t0))), where Y(t) represents the LAI at time t, K is the maximum LAI, r is the growth rate parameter, and t0 is the inflection point time. Through training and validation with multiple batches of experimental data, the optimal estimates of each parameter are determined, and correction terms for environmental variables such as temperature, humidity, and light intensity are introduced to improve the model's adaptability under different conditions. The final generated dynamic LAI model accurately reflects the expansion pattern of safflower leaves throughout their entire growth cycle.

[0022] Meanwhile, special attention must be paid to the importance of data smoothing during the calculation process. Since the actual collected leaf vein branching angle and leaf area data may contain noise interference, the raw data must be preprocessed using a Savitzky-Golay filter before calculating the rate of change and coefficient of variation. This filter can effectively remove high-frequency noise while preserving data characteristics, ensuring the accuracy of subsequent analysis results. The smoothed data sequence will be used as input to calculate various intermediate stability quantification parameters, thereby ensuring the scientific rigor and reliability of the entire analysis process.

[0023] S303. Based on the intermediate stability quantification parameters, verify the consistency of the X-year field trial data, combine chloroplast genome haplotype analysis to exclude environmental interference, and generate a preliminary stability discrimination parameter set containing statistically significant indicators. In step S303, it is necessary to explain in detail that, firstly, the consistency of the intermediate stability quantification parameters is verified using field trial data. Observational data from multi-location field trials over X years are selected, and the degree of parameter variability among different experimental sites is assessed through analysis of variance. The ratio of between-group mean square to within-group mean square is calculated to determine the spatial stability of the parameters. Simultaneously, chloroplast genome haplotype analysis is combined to screen sample populations with the same genetic background, eliminating phenotypic interference caused by environmental factors. Specifically, the whole chloroplast genome sequence of leaf samples is extracted, and a haplotype network map is constructed using SNP loci to identify the main haplotype groups and their distribution frequencies. Haplotype information is correlated with the intermediate stability quantification parameters, eliminating outliers significantly affected by environmental factors and retaining the core parameter set with genetic stability. Finally, non-parametric tests are used to calculate the statistical significance indices of each parameter, including p-values ​​and effect sizes, generating a preliminary stability discrimination parameter set containing significance level markers.

[0024] S304. Based on the preliminary stability discrimination parameter set, the leaf area index dynamic model is verified by comparing with the manually measured values ​​and optimized, and finally a stability discrimination parameter set that meets the DUS stability test requirements is generated.

[0025] In step S304, it is necessary to explain in detail that, to ensure the reliability of the stability discrimination parameter set, a combination of manual measurement and model prediction is used for verification and optimization. Specifically, 100 safflower leaf samples are selected, and actual leaf area data are obtained at different growth stages using a high-precision leaf area meter. These data are then compared with the predicted values ​​of the leaf area index dynamic model. The root mean square error (RMSE) and coefficient of determination (R²) are calculated to evaluate the goodness of fit of the model. If the RMSE exceeds a preset threshold (e.g., 0.05) or the R² is lower than 0.9, the key parameters of the model are dynamically adjusted, including the growth rate parameter r, the inflection point time t0, and the environmental factor weighting coefficients, until the model prediction accuracy meets the requirements. Furthermore, to further improve the adaptability of the model, a multivariate regression analysis method is introduced, incorporating environmental factors such as temperature, humidity, and light intensity as covariates into the model framework. The parameters are then refitted, and their stability under different environmental conditions is verified. The final stability discrimination parameter set must include at least 15 quantitative indicators, covering core parameters such as the rate of change of leaf vein branch angle, the coefficient of variation of leaf area index, and the standard deviation of growth rate, and ensure that it has good repeatability and consistency in field trials, so as to meet the strict stability requirements of DUS test.

[0026] As an optional embodiment of the present invention, optionally, in step S4, based on the three-dimensional morphological feature set and the stability discriminant parameter set, a genotype-phenotype association network is constructed using genome-phenome joint analysis, and a multimodal association feature vector is generated by fusing chloroplast genome SNP data and lateral vein density variation coefficients using a cosine similarity algorithm, including: S401. Based on the three-dimensional morphological feature set, extract phenotypic features such as the leaf margin serration depth / width ratio and the standard deviation of the leaf vein branch angle. Combine the leaf vein branch angle change rate and leaf area index variation coefficient in the stability discrimination parameter set to construct a phenotypic feature vector set. In step S401, it is necessary to explain in detail that the key parameters in the three-dimensional morphological feature set are first screened and normalized to ensure the comparability of features of different magnitudes. Specifically, phenotypic features such as the leaf margin serration depth / width ratio and the standard deviation of leaf vein branching angle are selected and mapped to the [0,1] interval using the min-max normalization method to eliminate the influence of dimensional differences on subsequent analysis. At the same time, dynamic growth indicators such as the leaf vein branching angle change rate and the coefficient of variation of leaf area index are extracted from the stability discrimination parameter set and processed using the Z-score normalization method to make their mean 0 and standard deviation 1. Next, the normalized phenotypic features and stability indicators are concatenated to form an initial phenotypic feature vector set. To improve the feature expressive power, principal component analysis (PCA) is introduced to reduce the dimensionality of the initial vector set, retaining the principal components with a cumulative contribution rate of more than 95%, thereby generating a compact and information-rich phenotypic feature vector set. This process not only effectively reduces the data dimensionality but also retains the key information of the original data to the greatest extent.

[0027] Based on the constructed phenotypic feature vector set, genomic data is further integrated to enhance the analytical capabilities of the association network. Specific operations include: extracting high-frequency mutation sites from chloroplast genome SNP data, calculating their polymorphism information content (PIC) in the population, and selecting sites with PIC values ​​greater than 0.3 as candidate markers. Subsequently, correlation analysis is performed between these markers and various indicators in the phenotypic feature vector set. The Pearson correlation coefficient is used to assess the strength of the linear relationship, and a significance threshold (e.g., p < 0.01) is set to select statistically significant association pairs. For feature combinations with strong nonlinear relationships, mutual information methods are introduced for supplementary analysis to ensure the comprehensiveness and accuracy of the association network. The final generated phenotypic feature vector set not only includes morphological and dynamic growth parameters but also incorporates genetic marker information.

[0028] S402. Based on chloroplast genome sequencing data, SNP sites are screened using bioinformatics tools to generate SNP data vectors containing genotypic variations. In step S402, it is necessary to explain in detail that firstly, the chloroplast genome is sequenced using high-throughput sequencing technology to obtain high-quality sequencing data. Subsequently, the sequencing data is processed using bioinformatics analysis tools, including sequence alignment, quality control, and variant detection, to screen for high-quality single nucleotide polymorphism (SNP) sites. During the screening process, strict filtering conditions must be set, such as a minimum coverage depth of 10, a minimum allele frequency greater than 0.05, and a deletion rate of less than 20%, to ensure the reliability and representativeness of the SNP sites. Next, the screened SNP sites are annotated to determine their location in the chloroplast genome and their potential functional significance. Simultaneously, the polymorphism information content (PIC) of each SNP site is calculated, and sites with higher PIC values ​​are preferentially retained as candidate markers. The final generated SNP data vector contains genotypic information for each sample at all candidate marker sites, represented in a numerical encoding form for subsequent analysis. This effectively captures the genetic diversity of the chloroplast genome.

[0029] S403. Calculate the similarity between phenotypic feature vectors and SNP data vectors using the cosine similarity algorithm, construct a genotype-phenotype association network, and quantify the feature association strength through network node weights. S404. Based on the aforementioned feature association strength, the lateral vein density variation coefficient and SNP data vector are fused, and a weighted average method is used to generate a multimodal association feature vector containing genotype-phenotype association information.

[0030] In step S404, it is necessary to explain in detail that the lateral vein density variation coefficient is first standardized to ensure that it is fused with the SNP data vector on the same scale. Specifically, the Z-score standardization method is used to transform the lateral vein density variation coefficient into a distribution with a mean of 0 and a standard deviation of 1, so as to eliminate the influence of dimensional differences on subsequent calculations. Next, the standardized lateral vein density variation coefficient is fused with the SNP data vector using a weighted average method. The weight allocation is determined based on the node weights of both in the genotype-phenotype association network. The node weights are calculated through edge connection strength in network analysis and can reflect the importance of different features in the association network. Based on this, the generated multimodal association feature vector not only contains genetic marker information but also integrates morphological and dynamic growth parameters, thus forming a comprehensive feature expression system. To improve the discriminative power of the feature vector, the Local Linear Embedding (LLE) algorithm is further introduced to reduce the dimensionality of the multimodal association feature vector, retaining key nonlinear structural information while reducing data redundancy. The final generated multimodal association feature vector must meet the dual requirements of dimensional compactness and information richness.

[0031] As an optional embodiment of the present invention, optionally, in step S403, the similarity between the phenotypic feature vector and the SNP data vector is calculated using the cosine similarity algorithm to construct the genotype-phenotype association network, including: S4031. Construct an N-dimensional phenotypic feature vector based on the phenotypic feature vector set; construct an M-dimensional SNP data vector based on chloroplast genome sequencing data; In step S4031, it is necessary to explain in detail that firstly, dimensionality analysis is performed on the phenotypic feature vector set to determine the number of features N it contains, and the phenotypic features of each sample are mapped to an N-dimensional space to form a corresponding N-dimensional phenotypic feature vector. Simultaneously, based on the SNP data vector generated from chloroplast genome sequencing data, an M-dimensional SNP data vector for each sample is constructed using the number M of high-quality SNP sites selected. To ensure the accuracy of subsequent calculations, both types of vectors need to be preprocessed, including missing value imputation and standardization. For the phenotypic feature vector, the mean imputation method is used to handle missing data, and the Z-score standardization method is used to transform it into a distribution with a mean of 0 and a standard deviation of 1. For the SNP data vector, it is directly normalized according to its numerical encoding form, unifying its numerical range to the [0,1] interval. After the above processing, each sample corresponds to a set of standardized N-dimensional phenotypic feature vectors and M-dimensional SNP data vectors.

[0032] S4032. Based on the N-dimensional phenotypic feature vector and the M-dimensional SNP data vector, the cosine value of the angle between the two vectors is calculated using the cosine similarity formula to generate a similarity matrix containing similarity values. In step S4032, the similarity between the phenotypic feature vector and the SNP data vector is calculated using the cosine similarity algorithm. Specifically, cosine similarity measures the degree of similarity by calculating the cosine of the angle between two vectors, with the formula: cosθ=(A·B) / (||A||·||B||), where A and B represent the phenotypic feature vector and the SNP data vector, respectively, A·B is the dot product of the two vectors, and ||A|| and ||B|| are the magnitudes of the two vectors, respectively. In the actual calculation process, the similarity of the N-dimensional phenotypic feature vector and the M-dimensional SNP data vector for each sample needs to be evaluated one by one to generate a similarity matrix, which reflects the association strength between phenotypic features and genetic markers.

[0033] S4033. Construct a genotype-phenotype association network based on the similarity matrix, and use the PageRank algorithm to quantify the importance of nodes.

[0034] In step S4033, it should be noted that the similarity matrix is ​​first processed into a network, with each phenotypic feature and SNP data vector serving as a node in the network, and the similarity value serving as the weight of the edges between nodes. By setting a similarity threshold (e.g., 0.6), edges with low association strength are filtered out, thus constructing a sparse but information-dense genotype-phenotype association network. To quantify the importance of each node in the network, the PageRank algorithm is introduced for analysis. This algorithm iteratively calculates the score of each node based on the node's connection relationships and edge weights, reflecting its influence in the entire network. Specifically, the node score depends not only on the number of nodes directly connected to it but also on the scores of these neighboring nodes themselves, thus enabling a comprehensive assessment of the node's importance in the global network. Furthermore, to further enhance the interpretability of the analysis results, the node scores are normalized, and importance levels are divided according to the score distribution, such as high, medium, and low. The final generated genotype-phenotype association network can intuitively display key nodes and their association patterns.

[0035] As an optional embodiment of the present invention, optionally, in step S5, based on the multimodal associated feature vector, a hybrid architecture model based on Transformer+CNN is used for feature alignment and classification, and the classification results of leaf margin entire / serrated / deeply lobed and the statistical significance value of leaf vein density difference are generated using a dynamic hyperparameter tuning module, including: S501. Based on the multimodal association feature vector, the Transformer encoder is used to extract the serialized feature association pattern, and the CNN convolutional layer is used to extract local spatial features. Cross-modal feature dimension alignment is performed through an adaptive pooling layer to generate an aligned feature vector. In step S501, it should be noted that the multimodal associated feature vector is first input into the Transformer encoder, which utilizes its self-attention mechanism to capture the global dependencies and serialization patterns between features. Specifically, the feature vector is divided into blocks by a multi-head self-attention module, the attention weight of each subspace is calculated, and the results are concatenated and linearly transformed to generate a context-aware feature representation. Simultaneously, positional encoding is introduced to preserve the order information of the features, ensuring the integrity of the serialization pattern. Next, the feature representation output by the Transformer is passed to a CNN convolutional layer, utilizing its local receptive field characteristics to extract local features in the spatial dimension. The size and stride of the convolutional kernel are optimized to adapt to the feature representation requirements at different scales. Subsequently, an adaptive pooling layer aligns the feature dimensions of the two modalities, using a combination of max pooling and average pooling to dynamically adjust the size of the feature maps, maintaining consistency during cross-modal fusion. The final generated aligned feature vector integrates the global serialization pattern and local spatial features, achieving seamless integration between modalities.

[0036] S502. Based on the aligned feature vector, a hybrid classification model is constructed. The Transformer branch processes global sequence dependencies, and the CNN branch processes local texture and morphological features. The dual-path features are fused through an attention mechanism to generate a fused feature representation. In step S502, it should be noted that the aligned feature vectors are first input into the two branches of the hybrid classification model: the Transformer branch and the CNN branch. In the Transformer branch, a multi-layer self-attention mechanism is used to further explore the global dependencies between features, while residual connections and layer normalization techniques are used to alleviate the gradient vanishing problem and ensure the stability of the deep network. For the CNN branch, multi-scale convolutional kernels are used to extract local texture and morphological features. The kernel sizes are designed in three formats: 3x3, 5x5, and 7x7 to capture detailed information at different resolutions. Next, in the dual-path feature fusion stage, a channel attention mechanism is introduced to weight the feature maps output by the two branches. Specifically, by calculating the importance score of each channel, its weight allocation is dynamically adjusted to highlight key features and suppress redundant information. The final generated fused feature representation not only includes the global sequence pattern but also integrates local texture and morphological details.

[0037] S503. Based on the fused feature representation, a classifier module is constructed, and the softmax function is used to output the classification probability distribution of whole / serrated / deeply lobed leaf margins, generating preliminary classification results; In step S503, it should be noted that the fused feature representation is first input into the classifier module, which consists of a fully connected layer and a softmax function. The fully connected layer performs a linear transformation on the fused features, mapping them to a dimensional space with the same number of target categories, while introducing non-linear expressive power through an activation function. Subsequently, the softmax function is used to normalize the output after the linear transformation, generating a probability distribution for leaf margin morphology categories. Specifically, the softmax function converts the score of each category into a probability value through exponential operations, ensuring that the sum of the probabilities of all categories is 1. In this process, for the three morphological categories of entire, serrated, and deeply lobed leaf margins, their corresponding probability values ​​are calculated, and the category corresponding to the maximum probability is selected as the preliminary classification result. To further improve the accuracy of classification, a confidence evaluation mechanism is introduced based on the output probability distribution. The entropy value of the probability distribution is calculated to measure the degree of certainty of the classification result. If the entropy value is lower than a preset threshold, the classification result is considered to have high confidence; otherwise, it needs to be combined with other auxiliary information for comprehensive judgment. The final generated classification result not only includes category labels but also corresponding probability values ​​and confidence scores.

[0038] S504. Based on the preliminary classification results and field test data, the statistical significance of the difference in leaf vein density is quantified using a t-test, the classification accuracy is evaluated using a confusion matrix, and a discrimination result set containing the classification results and statistical significance values ​​is generated.

[0039] In step S504, it should be noted that the preliminary classification results are first compared with the field trial data to extract relevant indicators of leaf vein density. A t-test is used to statistically analyze the differences in leaf vein density among different leaf margin morphology categories, calculating the p-value to assess the significance of the differences. If the p-value is less than a preset threshold (e.g., 0.05), the difference is considered statistically significant; otherwise, it is considered not significant. Based on this, the performance of the classification model is comprehensively evaluated using a confusion matrix, calculating key indicators such as accuracy, recall, and F1 score to ensure the reliability of the classification results. The final generated discrimination result set includes the classification label for each sample, the corresponding probability value, the confidence score, and the statistical significance value of the leaf vein density difference.

[0040] As an optional embodiment of the present invention, optionally, in step S6, based on the classification results and statistical significance values, the specificity, consistency, and stability discrimination logic is executed through a three-dimensional verification system to generate a comprehensive discrimination conclusion for the DUS test, including: S601. Based on the classification results and statistical significance values, construct the input layer of the three-dimensional verification system, integrate the three-dimensional morphological feature set, stability discrimination parameter set and genotype-phenotype association network, and generate a verification data pool containing multi-dimensional features. In step S601, it should be noted that the classification results and statistical significance values ​​are first processed in a structured manner, transforming them into a form suitable for the input of the three-dimensional validation system. Specifically, the classification label, probability value, confidence score, and statistical significance value of leaf vein density difference for each sample are mapped to a unified feature space to form an initial input vector. Next, the three-dimensional morphological feature set is integrated, including information such as the overall outline of the leaf, edge details, and texture distribution, and the dimensional differences between different features are eliminated through standardization. At the same time, a stability discrimination parameter set is introduced, covering environmental variables, dynamic changes in the growth cycle, and consistency indicators of repeated measurements in the field trial data, to ensure that the validation system can comprehensively reflect the performance characteristics of the samples under different conditions. In addition, the genotype-phenotype association network is used as an auxiliary input layer, and its node weights and edge connection strength information are used to supplement the influence of genetic background on phenotypic expression. After the above processing, the generated validation data pool contains multi-dimensional feature information, realizing seamless fusion of cross-modal data.

[0041] S602. Based on the verification data pool, execute specific discrimination logic, use difference significance analysis to identify the unique morphological combinations of leaf margin entire / serrated / deeply lobed and the significant difference characteristics of leaf vein density, and generate specific discrimination index; In step S602, it should be noted that the multi-dimensional feature information in the validation data pool is first subjected to hierarchical analysis to extract key feature combinations related to leaf margin morphology and vein density. Difference significance analysis methods, such as ANOVA or the Kruskal-Wallis test, are used to evaluate the distribution differences in morphological features among different leaf margin morphology categories. Specifically, for the three morphology categories of entire, serrated, and deeply lobed leaf margins, the mean and variance of indicators such as leaf contour curvature, edge complexity, and texture gradient are calculated respectively, and unique morphological combinations with significant discriminative power are identified by combining statistical test results. Simultaneously, the distribution characteristics of vein density are analyzed locally and globally, and spatial statistical methods are used to detect its variation patterns in different regions, further screening out feature variables that can reflect significant differences in vein density. Based on this, the generated specificity discrimination index not only covers unique combination patterns at the morphological level but also integrates the spatial distribution characteristics of vein density.

[0042] S603. Based on the specificity discrimination index, execute the consistency discrimination logic, verify the phenotypic consistency of the same genotype under different environments by comparing and verifying the data of X years of field trials, and verify the data distribution consistency by combining the chi-square test, and generate consistency discrimination index. In step S603, it should be noted that the specificity discriminant index is first analyzed to extract key characteristic variables that reflect phenotypic consistency. By comparing and analyzing X years of field trial data, the performance stability of the same genotype under different environmental conditions is assessed. Specifically, for each sample's morphological characteristics and leaf vein density distribution, its coefficient of variation over many years of trials is calculated, and its dynamic changes are detected using time series analysis. Simultaneously, the chi-square test is introduced to verify the consistency of the data distribution. By comparing the deviation between observed and expected values, the phenotypic consistency level of the samples under different environments is quantified. If the chi-square test result shows no significant difference in data distribution, the sample is considered to have high consistency; otherwise, further analysis of the source of variation is required. The final consistency discriminant index includes the stability assessment results of phenotypic characteristics and integrates supporting information from statistical tests.

[0043] S604. Based on the consistency discrimination index, execute the stability discrimination logic, combine environmental adaptability analysis and chloroplast genome haplotype verification, quantify the stability performance of genotype at different growth stages, and generate stability discrimination index; In step S604, it should be noted that, firstly, the consistency discriminant index is analyzed in depth to extract core characteristic variables related to genotype stability. Then, combined with environmental adaptability analysis methods, the sample's responsiveness to environmental changes at different growth stages is assessed. Specifically, a multivariate analysis of variance (MANOVA) model is used to quantify the interaction strength between environmental variables and phenotypic traits, and to identify trait combinations that are sensitive to or stable in response to environmental changes. Simultaneously, chloroplast genome haplotype verification technology is introduced, and the genetic stability of the genotype is further confirmed through comparative analysis of chloroplast DNA sequences. During this process, Bayesian inference methods are used to calculate the posterior probability distribution of different haplotypes to assess their adaptive performance under different environmental conditions. The final stability discriminant index integrates the results of environmental adaptability analysis and genetic background information, comprehensively reflecting the stability characteristics of the genotype at different growth stages and its potential adaptation mechanisms.

[0044] S605. Based on the specificity discrimination index, consistency discrimination index, and stability discrimination index, a comprehensive discrimination conclusion for DUS test is generated through the comprehensive discrimination module of the three-dimensional verification system.

[0045] As an optional embodiment of the present invention, optionally, in step S605, based on the specificity discrimination index, consistency discrimination index, and stability discrimination index, the comprehensive discrimination module of the three-dimensional verification system generates a comprehensive discrimination conclusion for the DUS test, including: S6051. Construct a three-dimensional index fusion matrix based on the specificity discrimination index, consistency discrimination index, and stability discrimination index, and calculate the comprehensive discrimination score using the weighted average method; The expression for calculating the comprehensive discrimination score is: ; in, This represents the overall discrimination score, reflecting the variety's comprehensive performance in meeting the DUS testing standards. The value range is [0,1]. , and These represent weighting coefficients, corresponding to the contributions of specificity, consistency, and stability, respectively, satisfying the following conditions: + + =1. Based on the UPOV guidelines and Red Flower DUS testing experience, the recommended weight is... =0.4 (specificity is the core discriminant dimension) =0.3 (consistency is key to field performance) =0.3 (stability is guaranteed by genetic expression). The specificity index value represents the significance of differences in morphological combinations of entire / serrated / deeply lobed leaf margins (e.g., the p-value of ANOVA analysis, after standardization and mapping to the [0,1] interval, p < 0.01). =1, P>0.05 =0, intermediate value linear interpolation). The consistency criterion value is determined jointly by the environmental coefficient of variation (CV) and the p-value of the χ² test. A CV < 10% contributes 0.5 points, and a p-value > 0.05 contributes 0.5 points, for a total of [value missing]. ∈[0,1], The stability discrimination index value is represented by the morphological variation coefficient under temperature and humidity fluctuations (e.g., <15%) and the haplotype consistency of the chloroplast genome (1 point is assigned when the matching degree is >95%), and after standardization, it is mapped to the [0,1] interval.

[0046] In step S6051, it should be noted that the specificity, consistency, and stability discriminant indicators are normalized to eliminate dimensional differences between different indicators and ensure their comparability. Specifically, for the specificity discriminant indicator, standardization is achieved by mapping the P-value of the ANOVA analysis to the [0,1] interval, where a smaller P-value corresponds to higher specificity. For the consistency discriminant indicator, a joint score is generated by combining the environmental coefficient of variation and the chi-square test results to ensure that it can comprehensively reflect the stability of the phenotype under different environmental conditions. For the stability discriminant indicator, the morphological coefficient of variation and the chloroplast genome haplotype matching degree are considered comprehensively, and a final stability score is generated through weighted calculation. Subsequently, a three-dimensional indicator fusion matrix is ​​constructed based on the normalized indicator values, and a weighted average method is used to calculate the comprehensive discriminant score. In this process, the selection of weight coefficients is optimized based on the UPOV guidelines and the experience of safflower DUS testing to ensure that the contribution of each dimension matches its importance in variety discrimination. The final comprehensive discriminant score can quantify the overall performance of the variety in meeting the DUS testing standards.

[0047] S6052. Based on the comprehensive discrimination score, and combined with the international DUS testing standard, set a threshold range to determine whether the requirements for variety specificity, consistency and stability are met, and generate a standard compliance verification result. In step S6052, it should be noted that a threshold is set for the overall discrimination score, and minimum requirements for specificity, consistency, and stability are established based on the international DUS testing standards. Specifically, if the overall discrimination score is greater than or equal to 0.8, the variety is deemed to meet the high standard requirements of the DUS test; if the score is between 0.6 and 0.8, further supplementary evaluation based on individual indicators is required; if the score is below 0.6, the variety is deemed not to meet the basic requirements of the DUS test. Based on this, a standard compliance verification result is generated by comparing and analyzing the performance of each individual indicator. For example, for varieties that do not meet the specificity standard but exhibit excellent consistency and stability, it may be recommended to adjust the weighting coefficients or optimize the testing conditions to improve their overall performance. The final output standard compliance verification result includes the overall discrimination score, individual indicator evaluation conclusions, and a clear judgment on whether the DUS test requirements are met.

[0048] S6053. Based on the verification results, generate a judgment conclusion document containing classification labels, statistical significance values, and judgment logic. In step S6053, it is important to explain that the verification results are structured and organized, mapping the classification labels, statistical significance values, and discrimination logic to a unified document framework. Specifically, the classification labels must clearly indicate the leaf margin morphology category (e.g., entire, serrated, or deeply lobed), along with corresponding probability values ​​and confidence scores to enhance the interpretability of the results. Statistical significance values ​​are presented in the form of p-values ​​or entropy values, combined with the results of difference significance analysis to further illustrate the reliability of the classification. Simultaneously, the discrimination logic must record in detail the key calculation processes and decision rules in each step, such as the ANOVA analysis results in the specificity discrimination index, the chi-square test conclusions in the consistency discrimination index, and the environmental adaptability assessment data in the stability discrimination index. Based on this, the generated discrimination conclusion document not only contains the final classification results but also provides a complete reasoning chain and data support, ensuring its high transparency and traceability.

[0049] S6054. Utilize expert assistance to manually review and confirm the judgment conclusion documents, and finally form a comprehensive judgment conclusion for the DUS test.

[0050] In step S6054, it should be noted that an auxiliary expert conducts a detailed review of the judgment conclusion document, focusing on checking the accuracy of the classification labels, the reasonableness of the statistical significance values, and the completeness of the judgment logic. During the review process, the expert needs to combine their domain knowledge and experience to verify the key calculation processes and decision rules in the document one by one to ensure that they meet the requirements of the international DUS testing standards. If any anomalies or uncertainties are found, the expert will verify and correct them by reviewing the original data, re-analyzing key indicators, or discussing with other experts. In addition, the expert also needs to conduct a logical review of the reasoning chain in the document to ensure that each step of the derivation has sufficient data support and scientific basis. After manual review, the expert will make a final confirmation of the judgment conclusion and sign the review opinion and conclusion in the document. The final comprehensive judgment conclusion of the DUS test not only includes the machine-generated classification results and statistical analysis data, but also integrates the expert's professional judgment and quality control, thereby further enhancing the authority and credibility of the conclusion.

[0051] Example 2 A DUS test system for leaf trait discrimination in safflower includes: processor; Memory used to store processor-executable instructions; The processor is configured to implement the safflower DUS test leaf morphology discrimination method when executing executable instructions.

[0052] It should be noted that the computer device includes a processor, a memory, and may also include one or more of a multimedia component, an input / output (I / O) interface, and a communication component.

[0053] The processor controls the overall operation of the computer device to complete all or part of the steps in the above-mentioned method for determining the leaf characteristics of safflower DUS test.

[0054] Memory is used to store various types of data to support the operation of the computer device. This data may include, for example, instructions for any application or method used to operate on the computer device, as well as application-related data. Memory can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0055] The multimedia component may include a screen and an audio component, wherein the screen may be, for example, a touch screen, and the audio component is used to output and / or input audio signals; for example, the audio component may include a microphone for receiving external audio signals, the received audio signals may be further stored in memory or transmitted via a communication component; the audio component may also include at least one speaker for outputting audio signals.

[0056] I / O interfaces provide interfaces between the processor and other interface modules, such as keyboards, mice, buttons, etc.; these buttons can be virtual buttons or physical buttons.

[0057] The communication component is used for wired or wireless communication between the computer device and other devices; wireless communication, such as Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G, 4G or 5G, or one or more combinations thereof, and the corresponding communication component may include: Wi-Fi module, Bluetooth module, NFC module, mobile communication module.

[0058] As a preferred embodiment, the computer device may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to execute the above-described method for determining the leaf morphology of red safflower DUS test.

[0059] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A method for identifying leaf traits in safflower using the DUS test, characterized in that, The method includes: Collect image data of the first and sixth leaves of the safflower; Based on the image data of the first and sixth leaves, a cross-scale three-dimensional reconstruction model is constructed to generate a three-dimensional morphological feature set including the leaf margin serration depth / width ratio and the standard deviation of the leaf vein branch angle. Based on the aforementioned three-dimensional morphological feature set, dynamic growth data from the seedling stage to maturity of safflower were collected using a time-series three-dimensional scanning system, and a stability discrimination parameter set was generated using a dynamic model of leaf vein branch angle change rate and leaf area index. Based on the three-dimensional morphological feature set and stability discrimination parameter set, a genotype-phenotype association network was constructed using genome-phenome joint analysis, and a multimodal association feature vector was generated by fusing chloroplast genome SNP data and lateral vein density variation coefficients through a cosine similarity algorithm. Based on the multimodal associated feature vectors, a hybrid architecture model based on Transformer+CNN is used for feature alignment and classification. The dynamic hyperparameter tuning module is used to generate classification results of leaf margin entire / serrated / deeply lobed and statistical significance values ​​of leaf vein density differences. Based on the classification results and statistical significance values, the specificity, consistency and stability discrimination logic is executed through the three-dimensional verification system to generate a comprehensive discrimination conclusion for the DUS test.

2. The method for determining leaf traits in safflower using the DUS test as described in claim 1, characterized in that, The generation of a three-dimensional morphological feature set includes: Based on the image data of the first and sixth blades, multi-view optical slices and three-dimensional point cloud data were acquired using a combination of micro-CT and laser confocal scanning microscopy to generate an original three-dimensional dataset containing the blade surface contour and leaf vein network structure. Based on the original 3D dataset, a 3D reconstruction model is generated by combining a 3D reconstruction algorithm based on voxel color assignment with a sparse reconstruction algorithm based on calibration constraints, and optimizing the matching degree between leaf texture and contour through the SFNet feature descriptor network. Based on the three-dimensional reconstruction model, the leaf edge serrated region is segmented using the watershed algorithm and the serration depth / width ratio is calculated. The straight vein features are detected by Hough transform and the standard deviation of the main vein width and the branch angle of the lateral veins are calculated. The leaf edge contour curve is fitted by the least squares method in combination with the three-dimensional point cloud data to generate quantified three-dimensional morphological feature parameters. Based on the aforementioned three-dimensional morphological feature parameters, the reconstruction accuracy is verified by comparing with manually measured values ​​and the reconstruction parameters are dynamically adjusted to generate a three-dimensional morphological feature set that meets the quantification requirements of DUS testing.

3. The method for determining leaf traits in safflower using the DUS test as described in claim 1, characterized in that, The generated stability discrimination parameter set includes: Based on the three-dimensional morphological feature set, a time-series three-dimensional scanner equipped with a ring light source and a polarization filter is used to collect three-dimensional morphological data of leaves from seedling stage to maturity stage at n-day intervals, while simultaneously recording environmental temperature, humidity and light intensity parameters, and generating an original growth dataset containing dynamic changes in leaf morphology and environmental correlation data. Based on the original growth dataset, the leaf vein branch angle change rate model and the leaf area index dynamic model are combined to calculate the leaf vein branch angle change rate, the leaf area index coefficient of variation and the growth rate standard deviation, and generate intermediate stability quantification parameters. Based on the aforementioned intermediate stability quantification parameters, the consistency of the X-year field trial data was verified. Combined with chloroplast genome haplotype analysis to exclude environmental interference, a preliminary stability discrimination parameter set containing statistically significant indicators was generated. Based on the preliminary stability discrimination parameter set, the leaf area index dynamic model is verified and optimized by comparing with manually measured values, and finally a stability discrimination parameter set that meets the DUS stability test requirements is generated.

4. The method for determining leaf traits in safflower using the DUS test as described in claim 1, characterized in that, Generating multimodal association feature vectors includes: Based on the three-dimensional morphological feature set, phenotypic features such as the leaf margin serration depth / width ratio and the standard deviation of the leaf vein branch angle are extracted. Combined with the leaf vein branch angle change rate and leaf area index variation coefficient in the stability discrimination parameter set, a phenotypic feature vector set is constructed. Based on chloroplast genome sequencing data, SNP sites are screened using bioinformatics tools to generate SNP data vectors containing genotypic variations; The similarity between phenotypic feature vectors and SNP data vectors is calculated using the cosine similarity algorithm, and a genotype-phenotype association network is constructed. The strength of feature association is quantified by the weights of network nodes. Based on the aforementioned feature association strength, the lateral vein density variation coefficient and SNP data vector are fused, and a weighted average method is used to generate a multimodal association feature vector containing genotype-phenotype association information.

5. The method for determining leaf morphology using the DUS test for safflower as described in claim 4, characterized in that, Constructing a genotype-phenotype association network includes: An N-dimensional phenotypic feature vector is constructed based on the phenotypic feature vector set; an M-dimensional SNP data vector is constructed based on chloroplast genome sequencing data; Based on the N-dimensional phenotypic feature vector and the M-dimensional SNP data vector, the cosine value of the angle between the two vectors is calculated using the cosine similarity formula to generate a similarity matrix containing similarity values. A genotype-phenotype association network was constructed based on the similarity matrix, and the PageRank algorithm was used to quantify the importance of nodes.

6. The method for determining leaf traits in safflower using the DUS test as described in claim 1, characterized in that, The statistical significance values ​​for the classification results of leaf margins (entire / serrated / deeply lobed) and the differences in leaf vein density include: Based on the multimodal association feature vector, the Transformer encoder is used to extract the serialized feature association pattern, and the CNN convolutional layer is used to extract local spatial features. The cross-modal feature dimension alignment is performed through the adaptive pooling layer to generate the aligned feature vector. A hybrid classification model is constructed based on the aligned feature vectors. The Transformer branch handles global sequence dependencies, and the CNN branch handles local texture and morphological features. The dual-path features are fused through an attention mechanism to generate a fused feature representation. A classifier module is constructed based on the fused feature representation, and the softmax function is used to output the classification probability distribution of whole / serrated / deeply lobed leaf margins, generating preliminary classification results; Based on the preliminary classification results and field trial data, the statistical significance of the differences in leaf vein density was quantified using a t-test, and the classification accuracy was evaluated using a confusion matrix to generate a set of discrimination results containing the classification results and statistical significance values.

7. The method for determining leaf traits in safflower using the DUS test as described in claim 1, characterized in that, The comprehensive judgment conclusions generated by the DUS test include: Based on the classification results and statistical significance values, a three-dimensional verification system input layer is constructed, integrating a three-dimensional morphological feature set, a stability discrimination parameter set, and a genotype-phenotype association network to generate a verification data pool containing multi-dimensional features. Based on the verification data pool, specific discrimination logic is executed, and the unique morphological combinations of leaf margins (entire / serrated / deeply lobed) and significant differences in leaf vein density are identified using difference significance analysis, thereby generating specific discrimination indicators. Based on the specific discriminant index, the consistency discriminant logic is executed. The phenotypic consistency of the same genotype under different environments is verified by comparing field trial data from X years. The chi-square test is used to verify the consistency of data distribution, and a consistency discriminant index is generated. Based on the aforementioned consistency discrimination index, stability discrimination logic is executed. Combined with environmental adaptability analysis and chloroplast genome haplotype verification, the stability performance of genotypes at different growth stages is quantified, and stability discrimination index is generated. Based on the specificity discrimination index, consistency discrimination index, and stability discrimination index, the comprehensive discrimination module of the three-dimensional verification system generates a comprehensive discrimination conclusion for the DUS test.

8. The method for determining leaf traits in safflower using the DUS test as described in claim 7, characterized in that, The comprehensive judgment conclusions generated by the DUS test include: A three-dimensional index fusion matrix is ​​constructed based on the specificity discrimination index, consistency discrimination index, and stability discrimination index, and the comprehensive discrimination score is calculated by the weighted average method. Based on the comprehensive discrimination score, and combined with the threshold range set by the international DUS testing standard, it is determined whether the requirements for variety specificity, uniformity and stability are met, and a standard compliance verification result is generated. Based on the verification results, a judgment conclusion document is generated, which includes classification labels, statistical significance values, and judgment logic. The judgment conclusion documents are manually reviewed and the conclusions are confirmed by the auxiliary experts, and finally a comprehensive judgment conclusion of DUS test is formed.

9. The method for determining leaf morphology using the DUS test for safflower as described in claim 8, characterized in that, The expression for calculating the comprehensive discrimination score is: in, This indicates the overall judgment score. , and Indicates the weighting coefficient. This indicates the specificity discrimination index value. This represents the consistency discrimination index value. This represents the stability discrimination index value.

10. A device for judging the leaf morphology of safflower using the DUS test, characterized in that, The system includes: processor; Memory used to store processor-executable instructions; The processor is configured to implement the safflower DUS test leaf morphology discrimination method according to any one of claims 1 to 9 when executing the executable instructions.

Images