A fertile plant discrimination method and system based on bud size and weight

Abstract

The application relates to a fertile plant identification method and system based on bud size and weight, and relates to the technical field of seed production. The method comprises the following steps: collecting the longitudinal diameter, transverse diameter and fresh weight of part of plants selected from the same piece of seed field to construct a multivariate data set and grouping according to the longitudinal diameter; sorting according to the transverse diameter in each group and screening candidate fertile plants in combination with a preset fresh weight lower limit; verifying the candidate plants by the extrusion method to obtain standard samples and re-collecting data to construct a fertile data set; counting the distribution characteristics of the transverse diameter and fresh weight in the fertile data set and outputting the lower limit threshold value of the corresponding longitudinal diameter interval; collecting the data of the remaining plants and comparing and identifying according to the generated threshold value. The application can effectively solve the problem that the existing fertile plant identification technology cannot consider the operation labor intensity and accuracy, and realizes efficient and accurate fertile plant screening.

Description

Technical Field

[0001] This invention belongs to the field of seed production technology, and in particular relates to a method and system for identifying fertile plants based on flower bud size and weight. Background Technology

[0002] Identifying fertile plants is a crucial step in rapeseed breeding. Currently, existing technologies for identifying fertile plants include pulp-squeezing identification, bud morphology identification, and flower observation identification, which are applicable during the small bud stage (from bolting to large bud stage), the large bud stage, and the initial flowering stage, respectively.

[0003] However, while the sap extraction identification technology boasts high accuracy, it requires breeders to repeatedly pinch off flowers and extract sap, resulting in high labor intensity, especially when the cultivated area is large, where the labor intensity increases exponentially. Flower bud morphology identification technology and flower observation identification technology have relatively lower labor intensity, but their accuracy is insufficient. Summary of the Invention

[0004] This invention addresses the problem that existing fertile plant identification technologies cannot balance operational labor intensity and accuracy by providing a method and system for identifying fertile plants based on flower bud size and weight. It constructs a complete closed-loop process of data acquisition, initial screening, verification, threshold generation, and field-wide identification. Through an adaptive threshold mechanism and hierarchical processing strategy, it achieves systematic technical advantages, solving the problem of existing fertile plant identification technologies failing to balance operational labor intensity and accuracy. This provides a new solution for fertile plant identification in the seed production field, and can form a series of fixed, dynamic data standards for identifying fertile plants under different varieties, planting times, regions, and agricultural models.

[0005] The technical solution adopted in this invention is: A method for identifying fertile plants based on flower bud size and weight includes the following steps: Step S1: Select some plants in the same growth cycle from the same field, and collect the longitudinal diameter, transverse diameter, and fresh weight of at least one flower bud on the outermost ring of the main flower stalk of each selected plant to construct a multivariate dataset; each multivariate data in the multivariate dataset includes plant number, longitudinal diameter, transverse diameter, and fresh weight; the growth cycle refers to a specific developmental stage of the plant between the bolting stage and the large bud stage. Step S2: Divide the longitudinal diameter data in the multivariate dataset into longitudinal diameter grouping intervals, and assign the corresponding plant number, transverse diameter and fresh weight to the corresponding sample set with the same longitudinal diameter. Step S3: In each of the sample sets with the same longitudinal diameter, sort the transverse diameter data from largest to smallest, select the fresh weight data corresponding to the top-ranked transverse diameter, and further verify whether it meets the preset lower limit of fresh weight under the corresponding longitudinal diameter interval. Record the plants with the corresponding plant numbers whose fresh weight data passes the verification as candidate fertile plants. Step S4: Destructive sampling of the candidate fertile plants is performed using the pulping method to identify fertility. The candidate fertile plants confirmed as fertile through pulping are used as standard samples. Referring to steps S1 and S2, the longitudinal diameter, transverse diameter, and fresh weight of at least one flower bud in the outermost ring of the main flower stalk of each standard sample are re-collected to construct a multivariate fertility dataset and a fertile sample set with the same longitudinal diameter. The multivariate fertility dataset includes longitudinal diameter, transverse diameter, and fresh weight. Step S5: In each set of fertile samples with the same longitudinal diameter, statistically analyze the distribution characteristics of transverse diameter and fresh weight, and output the lower limit threshold of transverse diameter and the lower limit threshold of fresh weight corresponding to different longitudinal diameter grouping intervals. Step S6, referring to step S1, collect the longitudinal diameter, transverse diameter and fresh weight of at least one flower bud on the outermost ring of the main flower stalk of each remaining plant in the field. Based on the longitudinal diameter data, select the corresponding longitudinal diameter grouping interval and the corresponding lower limit threshold of transverse diameter and lower limit threshold of fresh weight for comparison. If the transverse diameter is greater than or equal to the corresponding lower limit threshold of transverse diameter and the fresh weight is greater than or equal to the corresponding lower limit threshold of fresh weight, then the plant is identified as a fertile plant.

[0006] A fertile plant identification system based on flower bud size and weight, for implementing the fertile plant identification method based on flower bud size and weight as described herein, comprising: The data acquisition module is used to select some plants in the same growth cycle from the same field, and collect the longitudinal diameter, transverse diameter and fresh weight of at least one flower bud on the outermost ring of the main flower stalk of each selected plant to construct a multivariate dataset; each multivariate data in the multivariate dataset includes plant number, longitudinal diameter, transverse diameter and fresh weight. The grouping processing module is used to divide the longitudinal diameter data in the multivariate dataset into longitudinal diameter grouping intervals, and assign the corresponding plant number, transverse diameter and fresh weight to the corresponding sample set with the same longitudinal diameter. The initial screening and discrimination module is used to sort the samples in each longitudinal diameter set according to the transverse diameter data from largest to smallest, select the fresh weight data corresponding to the top-ranked transverse diameter, and further verify whether it meets the preset lower limit of fresh weight under the corresponding longitudinal diameter interval. Plants with the corresponding plant number whose fresh weight data passes the verification are recorded as candidate fertile plants. The fertility verification module is used to perform destructive sampling on the candidate fertile plants using the pulping method to identify their fertility. The candidate fertile plants confirmed as fertile by pulping are used as standard samples. The longitudinal diameter, transverse diameter, and fresh weight of at least one flower bud in the outermost ring of the main flower stalk of each standard sample are re-collected to construct a multivariate fertility dataset and a fertile sample set with the same longitudinal diameter. The multivariate fertility dataset includes longitudinal diameter, transverse diameter, and fresh weight. The threshold generation module is used to statistically analyze the distribution characteristics of transverse diameter and fresh weight in each set of fertile samples with the same longitudinal diameter, and output the lower limit threshold of transverse diameter and the lower limit threshold of fresh weight corresponding to different longitudinal diameter grouping intervals. The whole-field discrimination module is used to collect the longitudinal diameter, transverse diameter, and fresh weight of at least one flower bud on the outermost ring of the main flower stalk of each remaining plant in the field. Based on the longitudinal diameter data, the corresponding longitudinal diameter grouping interval, as well as the corresponding transverse diameter lower limit threshold and fresh weight lower limit threshold, are selected for comparison. If the transverse diameter is greater than or equal to the corresponding transverse diameter lower limit threshold and the fresh weight is greater than or equal to the corresponding fresh weight lower limit threshold, the plant is judged as a fertile plant.

[0007] The beneficial effects of this invention are: This application provides a method and system for identifying fertile plants based on bud size and weight. The method involves selecting plants from the same field at the same growth stage, collecting the longitudinal diameter, transverse diameter, and fresh weight of at least one bud from the outermost ring of the main flower stalk of each plant to construct a multivariate dataset. The longitudinal diameter data is then divided into longitudinal diameter grouping intervals to form a sample set of the same longitudinal diameter. Based on this, the transverse diameter data are sorted from largest to smallest, and the fresh weight data corresponding to the highest-ranking transverse diameters is selected for a preset fresh weight lower limit verification to obtain candidate fertile plants. Finally, the fertility of the candidate fertile plants is determined by destructive sampling using a sap extraction method. Fertile plants were considered as standard samples. Subsequently, the longitudinal diameter, transverse diameter, and fresh weight of the flower buds of the standard samples were re-collected to construct a multivariate fertile dataset and a fertile sample set with the same longitudinal diameter. The distribution characteristics of transverse diameter and fresh weight were statistically analyzed to output the lower limit thresholds of transverse diameter and fresh weight corresponding to different longitudinal diameter grouping intervals. Finally, the flower bud data of the remaining plants in the field were collected, and the corresponding longitudinal diameter grouping interval, as well as the corresponding lower limit thresholds of transverse diameter and fresh weight, were selected based on the longitudinal diameter data for comparison. If the transverse diameter is greater than or equal to the corresponding lower limit threshold and the fresh weight is greater than or equal to the corresponding lower limit threshold, the plant is identified as a fertile plant. By first establishing a preliminary screening model using a small sample size, and then optimizing the discrimination threshold through real fertility verification, a shift from experience-based judgment to data-driven analysis is achieved. This allows most plants to be identified without destructive testing, effectively solving the problem that existing fertile plant identification technologies cannot balance labor intensity and accuracy. It avoids the high labor intensity of traditional pulping methods or the low accuracy of morphological identification methods, improving the scientific rigor, stability, and universality of the discrimination process. This ensures reliable fertile plant identification results can be obtained under different varieties and environmental conditions, achieving an optimal balance between accuracy and efficiency. Furthermore, this invention allows for the creation of a series of fixed, dynamic data standards for identifying fertile plants under different varieties, planting times, regions, and agricultural models. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 A flowchart of the fertile plant identification method based on bud size and weight provided in Embodiment 1 of this application.

[0010] Figure 2 This is a schematic diagram of a rapeseed inflorescence.

[0011] Figure 3 This is a diagram of a flower bud. Detailed Implementation

[0012] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," "counterclockwise," "axial," "radial," and "circumferential" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0013] The following disclosure provides many different embodiments or examples for implementing different structures of the present invention. To simplify the disclosure of the present invention, the components and arrangements of specific examples are described below. Of course, these are merely examples and are not intended to limit the present invention.

[0014] The embodiments of the invention will now be described in detail with reference to the accompanying drawings.

[0015] Example 1

[0016] Figure 1 A flowchart illustrating a method for identifying fertile plants based on bud size and weight, provided in this application embodiment. This method includes the following steps: Step S1: Select some plants in the same growth cycle from the same field, and collect the longitudinal diameter, transverse diameter and fresh weight of at least one flower bud on the outermost ring of the main flower stalk of each selected plant to construct a multivariate dataset; each multivariate data in the multivariate dataset includes plant number, longitudinal diameter (flower bud height, accurate to 0.01mm), transverse diameter (the size of the circumference of the flower bud at its maximum, accurate to 0.01mm) and fresh weight (accurate to 0.001g).

[0017] The growth cycle can refer to a specific developmental stage of the plant, from bolting to budding, and then to the large bud stage. For example, if the plant is in the large bud stage, the bud morphology is stable and the buds are not yet open, making it suitable for non-destructive measurement. The outermost ring of the main flower stalk can refer to the earliest developing ring of buds on the main inflorescence. The buds in this area have good uniformity in size and are representative. Figure 2The flower buds circled in red in the diagram are shown. The collection process involved measuring the longitudinal length and transverse width of the buds using measuring tools (such as calipers) and weighing them fresh using a precision electronic balance. The multivariate dataset serves as the foundational database for subsequent analysis. Each multivariate data point is indexed by a unique plant number to ensure a one-to-one correspondence between data and plants. For example, taking the fertility identification of rapeseed plants as an example, for rapeseed plant numbered P001, the third bud from the outermost ring of the main flower stalk during the budding stage (e.g., ...) was collected. Figure 3 As shown in the figure, the longitudinal diameter was measured to be 6.75 mm, the transverse diameter to be 3.72 mm, and the fresh weight to be 0.207 g, forming a complete multivariate data record. This step aims to establish an original sample library, describing the bud state through multidimensional physical characteristics, thus avoiding the high labor intensity problem caused by direct destructive testing in traditional methods.

[0018] It should be noted that when selecting plants from the same field that are in the same growth cycle, breeders can make a rough judgment on the plant height of the selected plants to exclude those that are significantly lower (e.g., 10cm lower than the roughly estimated average plant height) or significantly higher (e.g., 10cm higher than the roughly estimated average plant height). This eliminates differences in the plant population and can prevent obvious outliers in the initial data collection from affecting the initial judgment.

[0019] Step S2: Divide the longitudinal diameter data in the multivariate dataset into longitudinal diameter grouping intervals, and assign the corresponding plant number, transverse diameter and fresh weight to the corresponding sample set with the same longitudinal diameter.

[0020] The longitudinal diameter grouping intervals are continuous intervals divided according to the numerical range of the longitudinal diameter of the flower buds. These intervals are used to eliminate size deviations caused by minor differences in the developmental stages of the flower buds. The division can be done using equal width or equal frequency division. For example, a longitudinal diameter of 6.70mm-6.79mm can be divided into the first interval, and 6.80mm-6.89mm into the second interval. The same longitudinal diameter sample set is a subset that aggregates data with similar longitudinal diameter characteristics. Its purpose is to analyze the distribution patterns of transverse diameter and fresh weight while controlling for the longitudinal diameter variable, thereby eliminating the interference of longitudinal diameter differences on the discrimination results. For example, if a flower bud has a longitudinal diameter of 7.2mm and falls into the 7.20mm-7.29mm interval, its corresponding plant number, transverse diameter of 3.03mm, and fresh weight of 0.163g are included in the same longitudinal diameter sample set for that interval. This grouping process enables refined comparison at the same developmental level, making the subsequent threshold settings more targeted and scientific, and effectively improving the stability of the discrimination.

[0021] Step S3: In each sample set with the same longitudinal diameter, sort the transverse diameter data from largest to smallest, select the fresh weight data corresponding to the top-ranked transverse diameter, and further verify whether it meets the preset lower limit of fresh weight under the corresponding longitudinal diameter interval. Record the plants with the corresponding plant numbers whose fresh weight data passes the verification as candidate fertile plants.

[0022] The "ranking higher" refers to the portion of data with larger transverse diameter values ​​within the same longitudinal diameter sample set, typically selecting the top 10% to 30%, as buds with larger transverse diameters are more likely to develop well and have fertility potential. The preset fresh weight lower limit is a minimum weight standard for fertile buds within this longitudinal diameter range, set based on experience or preliminary statistics. It is used to eliminate falsely large buds that are large but shriveled (i.e., hollow or aborted). The verification process compares the fresh weight of the selected large transverse diameter buds with this lower limit value; only plants that simultaneously meet the criteria of large transverse diameter and sufficient fresh weight are marked as candidate fertile plants. For example, in the 7.30mm-7.39mm longitudinal diameter group, the top 20% of flower buds with the largest transverse diameter are selected. If a flower bud has a transverse diameter of 3.11mm but a fresh weight of only 0.080g (below the preset lower limit of 0.100g), it is discarded; while another flower bud with a transverse diameter of 4.30mm and a fresh weight of 0.148g is verified and recorded as a candidate fertile plant. Through the dual screening mechanism of transverse diameter sorting and fresh weight verification, high-probability fertile individuals can be quickly identified from a large number of samples, significantly reducing the number of samples requiring destructive verification and lowering the workload.

[0023] Step S4: Destructive sampling of candidate fertile plants is performed using the pulping method to identify fertility. Candidate fertile plants identified as fertile through pulping are used as standard samples. Referring to steps S1 and S2, the longitudinal diameter, transverse diameter, and fresh weight of at least one flower bud in the outermost ring of the main flower stalk of each standard sample are collected again to construct a multivariate fertility dataset and a fertile sample set with the same longitudinal diameter. The multivariate fertility dataset includes longitudinal diameter, transverse diameter, and fresh weight.

[0024] Among these methods, the sap extraction method is the existing gold standard for fertility testing. It determines fertility by squeezing flower buds to observe whether pollen sap leaks out. While accurate, it is destructive and labor-intensive. In this step, this method is only performed on candidate fertile plants that have undergone rigorous initial screening, ensuring minimal destructive sampling. Standard samples refer to the set of plants confirmed as truly fertile by the sap extraction method, representing the distribution of physical characteristics of truly fertile plants under the current farming conditions. Re-collection involves re-measuring the precise size and weight of these confirmed standard samples to construct a pure multivariate fertility dataset. Subsequently, the samples are grouped according to their longitudinal diameter to form fertile sample sets of the same longitudinal diameter. For example, after sap extraction verification, plants numbered P005 and P012 are confirmed as truly fertile, and their flower bud data are remeasured and categorized into the corresponding longitudinal diameter interval sample sets. The core of this step is to use a small number of high-precision verification results to correct and calibrate the discrimination model, transforming the empirical initial screening into a data-driven standard generation process, providing a reliable source of truth for subsequent accurate threshold calculations.

[0025] Step S5: In each fertile sample set with the same longitudinal diameter, statistically analyze the distribution characteristics of transverse diameter and fresh weight, and output the lower limit threshold of transverse diameter and the lower limit threshold of fresh weight corresponding to different longitudinal diameter grouping intervals.

[0026] The distribution characteristics refer to the statistical regularities of the transverse diameter and fresh weight data in the standard sample set, including but not limited to the mean, median, and specific percentiles (such as the 10th percentile). The lower limit thresholds for transverse diameter and fresh weight are critical values ​​calculated based on these distribution characteristics, used to define the minimum size and weight standards for fertile plants. Specifically, in statistical analysis, the 10th percentile of the transverse diameter data in a fertile sample set with the same longitudinal diameter can be used as the lower limit threshold for that group, and the 10th percentile of the fresh weight data can be used as the lower limit threshold for fresh weight. This ensures coverage of the vast majority of truly fertile plants while excluding abnormally small values. For example, in a fertile sample set within a certain longitudinal diameter range, if statistics show that 90% of the fertile flower buds have a transverse diameter greater than 3.60 mm and a fresh weight greater than 0.200 g, then 3.60 mm and 0.200 g are set as the discrimination thresholds for that range, respectively. Thresholds generated in this way can adapt to changes in flower bud characteristics under different varieties and growth environments, solving the problem of fixed thresholds failing under different conditions and achieving dynamic optimization of the discrimination criteria.

[0027] Step S6, referring to step S1, collect the longitudinal diameter, transverse diameter and fresh weight of at least one flower bud on the outermost ring of the main flower stalk of each remaining plant in the field. Based on the longitudinal diameter data, select the corresponding longitudinal diameter grouping interval and the corresponding lower limit threshold of transverse diameter and lower limit threshold of fresh weight for comparison. If the transverse diameter is greater than or equal to the corresponding lower limit threshold of transverse diameter and the fresh weight is greater than or equal to the corresponding lower limit threshold of fresh weight, then the plant is identified as a fertile plant.

[0028] The remaining plants can refer to all plants to be tested in the field excluding the initial sampling and candidate verification plants, or all plants to be tested excluding the candidate verification plants. The discrimination process involves matching and comparing the measured data with the adaptive thresholds generated above: first, the interval to which the plant belongs is determined based on the measured longitudinal diameter, and then the lower limit threshold of the transverse diameter and fresh weight of that interval is used for dual judgment. Only when the measured transverse diameter and fresh weight are both greater than or equal to the corresponding thresholds is the plant finally judged as fertile; if the measured transverse diameter or fresh weight is less than the corresponding threshold, it is judged as infertile. For example, for the remaining plant numbered P100, its longitudinal diameter is measured to be 7.26 mm (belonging to the 7.20 mm-7.29 mm group), its transverse diameter is 3.69 mm, and its fresh weight is 0.216 g; if the threshold for this group is a transverse diameter of 3.60 mm and a fresh weight of 0.200 g, since 3.69 > 3.6 and 0.216 > 0.200, P100 is judged as a fertile plant. For example, the remaining plant, numbered P120, has a longitudinal diameter of 7.30 mm (belonging to the 7.30 mm-7.39 mm group), a transverse diameter of 3.11 mm, and a fresh weight of 0.148 g. If the threshold for this group is a transverse diameter of 3.61 mm and a fresh weight of 0.210 g, then since 3.11 < 3.61 and 0.148 < 0.210, P120 is determined to be an infertile plant. This step utilizes the precise model established in the previous steps to efficiently screen all plants in the field without any further destructive operations. This ensures high accuracy in the determination while significantly reducing the labor intensity in large-scale breeding work, achieving a perfect balance between efficiency and precision.

[0029] In this embodiment, a complete closed-loop system for identifying fertile plants is constructed through the synergistic effect of the aforementioned technical means. First, a preliminary screening mechanism rapidly narrows down the candidate pool using non-destructive size and weight data, effectively filtering out most obviously infertile plants and reducing the workload of subsequent high-cost verification. Second, a small amount of destructive pulping verification is introduced as a truth value correction, transforming the preliminary screening results into highly reliable standard samples, overcoming the low accuracy of relying solely on morphological characteristics. Based on this, adaptive discrimination thresholds are statistically generated using standard samples, allowing the discrimination criteria to dynamically adjust with germplasm resources and environmental conditions, avoiding the poor universality of traditional fixed standards. Finally, these dynamic thresholds are applied to batch-discriminate the remaining plants in the field, achieving a transition from point-based verification to broad-based application. The entire process is progressive, with each step providing data support or a screening basis for the next, and each subsequent step verifying or optimizing the results of the previous step. Ultimately, while ensuring high accuracy, the number of destructive samplings is minimized, solving the technical problem of existing technologies failing to balance labor intensity and accuracy. Meanwhile, the method in the embodiments of this application can form a series of fixed dynamic data standard sets for identifying fertile plant data under different varieties, different planting times, different regions, and different agricultural models.

[0030] It should be noted that in this embodiment, the total time from collecting flower buds in step S1 to constructing a multivariate dataset to identifying the fertility of the remaining plants in the field in step S6 is controlled to be completed within 1 to 3 days as much as possible, so as to avoid significant changes in the longitudinal diameter, transverse diameter and fresh weight of flower buds due to the rapid development of plants, which would affect the identification results.

[0031] This application's embodiments introduce a data quality control mechanism during the data acquisition phase or the standard sample construction phase. Specifically: First, in step S1, the longitudinal diameter, transverse diameter, and fresh weight of at least one flower bud on the outermost ring of the main flower stalk of each selected plant are collected. Then, outlier statistical analysis is performed on the longitudinal diameter, transverse diameter, and fresh weight to remove abnormal data.

[0032] Outlier analysis refers to the process of identifying and removing extreme values ​​in a dataset that deviate from the overall distribution characteristics, using statistical principles. Specifically, this analysis is performed independently for the three indicators—vertical diameter, horizontal diameter, and fresh weight—obtained in step S1. Outlier analysis can be performed by calculating the mean and standard deviation of the dataset, identifying data exceeding the mean plus or minus three times the standard deviation (3σ principle) as outliers; or by using the interquartile range (IQR) method, calculating the first quartile (Q1) and the third quartile (Q3), marking data points less than Q1 - 1.5 × IQR or greater than Q3 + 1.5 × IQR as outliers and removing them. For example, in the horizontal diameter data of a sample set, if most data are concentrated between 3mm and 4mm, but some data are recorded as 6mm due to measurement error, the IQR method shows that the 6mm data point is far beyond the normal fluctuation range, and it is automatically identified as an outlier and deleted from the multivariate dataset. This outlier statistical analysis can eliminate extreme data interference caused by human error, instrument malfunction, or individual plant disease, ensuring that the division of longitudinal diameter grouping intervals in subsequent step S2 and the initial screening of candidate fertile plants in step S3 are based on real and reliable samples, thereby improving the robustness of the discrimination model.

[0033] Secondly, in step S4, after re-collecting the longitudinal diameter, transverse diameter, and fresh weight of at least one flower bud on the outermost ring of the main flower stalk of each standard sample, outlier statistical analysis was performed on the longitudinal diameter, transverse diameter, and fresh weight to remove abnormal data.

[0034] This step involves a secondary cleaning of the standard sample data from plants confirmed as fertile through the pulping method. In step S4, although candidate fertile plants have undergone preliminary screening and destructive verification, the re-collection of their bud size and weight to construct a multivariate fertile dataset may still be affected by environmental factors or secondary measurement errors. Therefore, the aforementioned outlier statistical analysis is also required on the re-collected longitudinal diameter, transverse diameter, and fresh weight data. Specifically, a distribution test is performed on the data within the fertile sample set of the same longitudinal diameter to remove outliers that, although they passed the pulping verification, have morphological indicators that significantly deviate from the normal fertile plant characteristic range for that variety. For example, if a plant confirmed as fertile through pulping has an abnormally high re-measured fresh weight due to soil adhesion or insufficient drying, exceeding the fresh weight distribution range of other standard samples in the same group, this data will be judged as abnormal and removed, and will not participate in the calculation of the lower limit thresholds for transverse diameter and fresh weight in subsequent step S5. This mechanism ensures the high purity of the fertile dataset used as the gold standard, prevents individual abnormal high-weight data from skewing the threshold, and makes the final discrimination criteria more accurately reflect the true morphological characteristics of fertile plants in the field.

[0035] This application constructs a dual data quality defense line through the synergistic effect of the aforementioned technical means. First, outlier removal is implemented in the initial data collection stage (step S1), purifying the multivariate dataset used for initial screening from the source and avoiding misselection or omission of candidate fertile plants due to dirty data. Second, outlier removal is implemented again in the standard sample construction stage (step S4), ensuring the accuracy of the ground truth dataset used to train the discrimination threshold. The combination of these two methods not only solves the technical problem of outliers in the original measurement data potentially interfering with the accuracy of the discrimination, but also improves the anti-interference capability of the entire discrimination method. Even if the data collected under complex field conditions has certain fluctuations, statistical methods can automatically filter noise, ensuring that the output lower limit thresholds for transverse diameter and fresh weight are highly representative and stable. This makes the discrimination results of the remaining plants in step S6 more reliable, effectively balancing labor intensity and discrimination accuracy.

[0036] It should be noted that in this embodiment, outlier statistical analysis can be performed on the collected data only when executing step S1 or step S4 to remove abnormal data.

[0037] The process of dividing the longitudinal axis grouping intervals in this embodiment of the application has also been optimized, as follows: First, the plant variety information of the test field is obtained. This plant variety information refers to characteristic data used to identify specific genetic lines of rapeseed or other cruciferous crops in the test field. This information is obtained by reading field management records, scanning field QR code labels, receiving manual input from users, or analyzing plant morphological characteristics using image recognition technology. The role of plant variety information is to serve as a key index for subsequently determining data grouping strategies, ensuring that grouping parameters match the biological characteristics of the current crop. For example, when the test field is planted with the Huayouza 62 variety, the system obtains the variety name by scanning the electronic sign at the field entrance. By accurately obtaining plant variety information, independent data processing benchmarks can be established for crops with different genetic backgrounds, avoiding data distribution distortion caused by differences in flower bud development patterns between varieties.

[0038] Then, based on the plant variety information, the corresponding longitudinal diameter grouping interval division parameters are automatically matched. These parameters include the grouping interval width and / or the number of groups; different plant varieties correspond to different division parameters. The longitudinal diameter grouping interval division parameters are control variables used to guide step S2 in discretizing the continuously changing longitudinal diameter data into several intervals. The grouping interval width can refer to the length of the numerical range covered by each longitudinal diameter group, such as 0.10 mm or 1.00 mm; the number of groups can refer to how many independent intervals the total longitudinal diameter data is divided into. These parameters are automatically matched based on a pre-constructed variety-parameter mapping database. This database stores the typical distribution statistics of flower bud longitudinal diameter for different rapeseed varieties under the same growth cycle. For varieties with differences in flower bud size, a wider grouping interval width or a larger number of groups is automatically matched to capture their subtle developmental gradients; while for varieties with high uniformity in flower bud size, a narrower interval width is matched.

[0039] For example, the system first receives the input variety code, searches its internal database, and if a match is found for variety A, it calls its corresponding interval width W1 and group number N1; if a match is found for variety B, it calls its interval width W2 and group number N2. For instance, for the large-bud type variety Youyan 202, the system automatically matches a group interval width of 1.00 mm and a group number of 8; while for the small-bud type variety Fengyou 737, the system automatically matches a group interval width of 0.60 mm and a group number of 10. Through this dynamic matching mechanism, the flower buds within the generated sample set with the same longitudinal diameter have higher consistency in biological development stages, thereby improving the accuracy of subsequent fertile plant identification based on transverse diameter and fresh weight.

[0040] This application achieves adaptive adjustment of data acquisition and processing strategies by acquiring plant variety information from the test field and automatically matching corresponding longitudinal diameter grouping interval parameters. Based on this, and utilizing the characteristic that different plant varieties correspond to different partitioning parameters, the system can optimize the data slice granularity for specific varieties' physiological characteristics. This variety-adaptive grouping method effectively solves the problem that fixed grouping parameters are difficult to adapt to the differences in bud development characteristics among different rapeseed varieties, making samples within the same longitudinal diameter sample set more comparable, avoiding discrimination bias caused by cross-variety mixing, and thus supporting accurate discrimination in multi-variety mixed planting scenarios, demonstrating the method's intelligence and adaptability.

[0041] It should be noted that the variety-parameter mapping relationship library in this embodiment can be derived by summarizing and generalizing the archived data (such as plant variety, planting location, growth cycle, collection time, longitudinal diameter, transverse diameter, fresh weight, fertility, etc.) after long-term implementation of the method in this embodiment. It is a conventional operating method that can be directly used by those skilled in the art.

[0042] This embodiment of the application also includes optimization processing for selecting the top-ranked transverse diameter ratios in step S3. Specifically, the following methods can be used: Method 1: In step S3, directly select the fresh weight data corresponding to the top 10%~30% of the transverse diameters. Specifically, within each sample set with the same longitudinal diameter, sort the plants according to the transverse diameter of their flower buds from largest to smallest, and directly extract the plants in the top 10% to 30% of the sorted sequence as the initial screening targets, and extract the corresponding fresh weight data of these plants for subsequent verification. This fixed proportion range is an empirical value derived from a large amount of experimental data, aiming to balance screening coverage and computational efficiency. For example, if a sample set with the same longitudinal diameter contains 100 plants, then select the top 10 to 30 plants with the largest transverse diameters (i.e., the top 10%~30%) to enter the next round of fresh weight verification. By setting this fixed proportion range, under most conventional growth conditions, it is possible to quickly identify potential fertile plants with larger flower bud sizes, avoiding the tedious pulping verification of all data, thereby reducing labor intensity.

[0043] Method 2: In step S3, the proportion of the top-ranked transverse diameters is dynamically determined based on the dispersion of the transverse diameter data within the same longitudinal diameter sample set. Specifically, the standard deviation or interquartile range of the transverse diameter data within the same longitudinal diameter sample set is calculated. The larger the standard deviation or interquartile range, the smaller the proportion of the top-ranked transverse diameters selected; the smaller the standard deviation or interquartile range, the larger the proportion of the top-ranked transverse diameters selected. The proportion ranges from 5% to 40%.

[0044] Method 2 provides an adaptive dynamic screening mechanism to address the poor adaptability of a fixed ratio under different data distributions. Specifically, the system first calculates the dispersion index of all transverse diameter data within the current longitudinal diameter sample set. This index can be the standard deviation reflecting the data fluctuation range or the interquartile range reflecting the distribution range of the middle 50% of the data. Subsequently, a negative correlation mapping relationship is established between the dispersion index and the selection ratio: when the calculated standard deviation or interquartile range is large, it indicates that the flower bud size of the batch of plants varies and the data is scattered. In this case, the selection ratio should be reduced (e.g., close to 5%) to strictly screen out a very small number of individuals with the best characteristics and prevent misselection due to data noise. Conversely, when the standard deviation or interquartile range is small, it indicates that the plants have high uniformity of growth and the data is concentrated. In this case, the selection ratio can be increased (e.g., close to 40%) to retain more potential fertile plants and avoid omissions. For example, when the standard deviation of the transverse diameter of a sample set is 1.00 mm (high dispersion), the system automatically lowers the selection ratio to 8%; while when the standard deviation of the transverse diameter of another sample set is only 0.20 mm (low dispersion), the system raises the selection ratio to 35%. Through this dynamic adjustment strategy, the initial screening mechanism can adapt to the variations in flower bud size under different varieties and environments, maximizing the coverage of truly fertile plants while ensuring the quality of candidate plants.

[0045] Compared to Method 1, Method 2 introduces a dynamic ratio adjustment mechanism based on the dispersion of data distribution, forming a close synergy with the aforementioned longitudinal diameter grouping and transverse diameter sorting steps. Specifically, longitudinal diameter grouping ensures the consistency of the comparison benchmark, while the dynamic ratio mechanism further refines the screening rigor under the same benchmark. By calculating the standard deviation or interquartile range to quantify the dispersion characteristics of the data, the system can intelligently perceive the uniformity of the current sample set and then adjust the leniency of the selection ratio accordingly. This data-driven feedback adjustment not only overcomes the shortcomings of the fixed ratio method, which is prone to introducing noise when the data fluctuates greatly or causing missed selections when the data is uniform, but also ensures the robustness of the screening results by combining it with a preset value range. Ultimately, this mechanism makes the acquisition of candidate fertile plants more scientific and flexible, effectively improving the accuracy of subsequent pulping verification and achieving the optimal balance between accuracy and efficiency in identifying fertile plants in complex field environments.

[0046] In this embodiment of the application, a credibility weighting mechanism is introduced in step S5 to optimize the threshold generation process, as follows: First, a confidence weight is assigned to each standard sample. The confidence weight is determined based on the standard sample's ranking position in diameter and / or the difference between its fresh weight and a preset lower limit for fresh weight in step S3: the higher the ranking in diameter and the greater the remaining fresh weight, the higher the confidence weight. The confidence weight can be a numerical indicator used to quantify the contribution of each standard sample in subsequent statistical calculations. This weight is generated based on the performance data of the standard samples in the initial screening stage (i.e., step S3). Specifically, the weight is determined based on two dimensions: one is the sample's ranking position in diameter within the set of samples with the same vertical diameter; the other is the difference between its fresh weight and the preset lower limit for fresh weight in the corresponding vertical diameter interval (i.e., remaining fresh weight). A higher ranking in diameter indicates a more refined size characteristic of the flower bud, and a higher probability that it represents a typical fertile plant; a greater remaining fresh weight indicates more sufficient nutrient accumulation in the flower bud, and a more stable biological basis for its fertility. Therefore, the system maps the performance of these two dimensions to weight values, so that samples with better performance receive higher weight coefficients. For example, if a standard sample ranks in the top 5% in terms of transverse diameter within the same longitudinal diameter group and its fresh weight exceeds the preset lower limit by 0.050g, it can be assigned a weight value of 1.5; while another sample ranks in the top 20% and its fresh weight only exceeds the lower limit by 0.010g, it can be assigned a weight value of 1.1. This dynamic assignment method based on ranking and remaining fresh weight ensures that high-quality standard samples dominate the threshold calculation, thereby avoiding interference from low-quality or marginal samples on the discrimination criteria.

[0047] Secondly, the distribution characteristics of the transverse diameter and fresh weight are calculated using a weighted average or weighted percentile method, and the lower limits for transverse diameter and fresh weight are output accordingly. The weighted average or weighted percentile method refers to incorporating the determined confidence weights as coefficients into the algorithm logic during statistical calculations. This method corrects the shortcomings of traditional arithmetic mean or simple percentile calculations that treat weights equally, making the final output distribution characteristics closer to the central tendency of high-confidence samples. Specifically, when calculating the lower limit for transverse diameter, the system no longer directly takes the average of the transverse diameter data of all standard samples, but instead calculates the sum of the products of each sample's transverse diameter data and its corresponding weight, and then divides by the total weight; or after sorting, the position of the Pth percentile is determined based on the cumulative weight rather than simply the number of samples. Similarly, the calculation of the lower limit for fresh weight follows the same weighted logic. For example, when calculating the lower limit of the transverse diameter within a certain longitudinal diameter interval, if the transverse diameters of high-weighted samples are generally large, the weighted calculated lower limit threshold will shift upwards accordingly, thus setting a more stringent discrimination standard. Conversely, if high-weighted samples are concentrated in a specific interval, the threshold will converge towards that interval. The resulting lower limit thresholds for transverse diameter and fresh weight can effectively reflect the typical morphological and weight characteristics of high-quality fertile plants, improving the accuracy and robustness of the discrimination model when screening remaining plants in the seed field.

[0048] This application achieves deep synergy of technical methods by assigning confidence weights to each standard sample and generating discrimination thresholds using a weighted statistical approach. Specifically, the high weights determined based on transverse diameter ranking and remaining fresh weight directly affect the aforementioned weighted average or weighted percentile calculation process. This synergy allows standard samples exhibiting superior size and weight characteristics in the initial screening stage to have a greater say in the final threshold formation process. Through this merit-based evaluation logic, the system can automatically suppress threshold deviations caused by individual low-quality standard samples, ensuring that the output lower limit thresholds for transverse diameter and fresh weight better represent the core characteristics of truly fertile plants. This weighting mechanism not only optimizes single-dimensional statistical results but also enhances the adaptability of the discrimination criteria to germplasm resources under different growth environments through multi-dimensional comprehensive consideration, thereby further improving the overall accuracy of fertile plant discrimination while maintaining low labor intensity.

[0049] In this embodiment, the specific calculation methods for the lower limit threshold of transverse diameter and the lower limit threshold of fresh weight generated in step S5 have been optimized. Specifically, as follows: Regarding the lower limit threshold for transverse diameter, in step S5, the lower limit threshold for transverse diameter corresponding to the same longitudinal diameter grouping interval is the average, median, or the transverse diameter data corresponding to the G percentile after sorting the transverse diameter data from largest to smallest, where G ranges from 5 to 30. The lower limit threshold for transverse diameter can refer to the critical value used to determine whether the transverse diameter of the flower buds of the plant under test meets the standard within a set of fertile samples of the same longitudinal diameter. This threshold is obtained by statistically analyzing the transverse diameter data of all standard samples within the same longitudinal diameter grouping interval. Specifically, there are three optional strategies for determining the lower limit threshold for transverse diameter: the first is to calculate the arithmetic mean of the transverse diameter data of all standard samples within the interval, suitable for scenarios where the data distribution is relatively symmetrical and there are no obvious extreme values; the second is to select the median of the transverse diameter data of all standard samples within the interval, i.e., the value located in the middle after sorting, which can effectively resist the interference of extremely large or small values ​​and is suitable for situations where the data has a slightly skewed distribution; the third is to select the transverse diameter data corresponding to the G percentile after sorting the transverse diameter data from largest to smallest, where G ranges from 5 to 30. For example, when G is 10, it means that the transverse diameter data within the fertile sample set with the same longitudinal diameter are arranged from largest to smallest, and the transverse diameter value at the 10th percentile is taken as the lower threshold. This means that 10% of the standard samples are allowed to have a transverse diameter smaller than this threshold, thus ensuring that the vast majority of fertile plants are selected while appropriately accommodating individual differences. Through this diversified threshold generation mechanism, the most suitable statistic can be flexibly selected based on the distribution characteristics of flower bud size data in actual planting fields (such as normal distribution, left-skewed or right-skewed distribution). This avoids the failure of the discrimination standard or excessive stringency caused by using the average value alone when the data fluctuates greatly, thereby improving the adaptability of the discrimination method to different varieties and growing environments.

[0050] Regarding the lower limit threshold for fresh weight, in step S5, the lower limit threshold for fresh weight corresponding to the same longitudinal diameter grouping interval is the average, median, or the fresh weight data corresponding to the H percentile after sorting the fresh weight data from largest to smallest, where H ranges from 5 to 30. The lower limit threshold for fresh weight can refer to the critical value used to determine whether the fresh weight of the flower buds of the plant under test meets the standard within the fertile sample set of the same longitudinal diameter. This threshold is statistically derived based on the fresh weight data of standard samples within the same longitudinal diameter grouping interval. Similar to the lower limit threshold for transverse diameter, the determination of the lower limit threshold for fresh weight also supports multiple statistical strategies: it can be the average of the fresh weight data of all standard samples within the interval, reflecting the overall average level; it can be the median to eliminate the influence of individual abnormally light or abnormally heavy flower buds on the standard; or it can be the H percentile after sorting the fresh weight data from largest to smallest, where H also ranges from 5 to 30. For example, if the fresh weight data of standard samples within a certain longitudinal diameter grouping interval exhibits a long-tailed distribution (i.e., there are a small number of extremely light fertile buds), directly using the average value as the lower limit might lead to a large number of normal fertile plants being misjudged as infertile due to their fresh weight being slightly lower than the average. In this case, if the 20th percentile of H=20 is used as the lower limit threshold for fresh weight, it can more accurately cover the true fertility range of the variety under specific growth conditions. By decoupling or synchronizing the threshold generation rules for transverse diameter and fresh weight, the system can independently handle the data distribution characteristics of both bud size and weight. The two are used together to construct a rigorous yet flexible multidimensional discrimination benchmark, effectively solving the problem that fixed thresholds in existing technologies cannot adapt to the natural variation of biological traits, ensuring high accuracy and high recall in the identification of fertile plants under complex field conditions.

[0051] This application achieves adaptive optimization of the discrimination criteria by providing multiple statistical measures, such as the mean, median, and G / H percentile (e.g., G / H = 5–30), as options for generating the lower limit thresholds for transverse diameter and fresh weight. Based on this, by utilizing the characteristics of the median or lower percentiles (e.g., the 5th to 30th percentiles), threshold shifts caused by anomalies in individual standard sample data (e.g., measurement errors or extreme physiological conditions) can be effectively avoided, enhancing the model's robustness to skewed distribution data. Simultaneously, the adjustable range of G / H values ​​(5–30) provides breeders with a means to adjust the stringency of the discrimination: when pursuing high-purity seed production, a smaller G / H value (e.g., 5%–10%) can be selected to raise the threshold, while when pursuing high yields or facing germplasm resources with significant variability, a larger G / H value (e.g., 20%–30%) can be selected to prevent missed detections. This flexible threshold setting mechanism, combined with the above-mentioned refined processing based on longitudinal diameter grouping, not only reduces the risk of misjudgment caused by the limitations of a single statistical indicator, but also enables the entire discrimination system to dynamically match the bud morphology and weight distribution patterns of different rapeseed varieties. Thus, while ensuring the accuracy of fertile plant identification, it minimizes labor intensity and improves seed production efficiency.

[0052] The embodiments of this application also update the threshold through multiple rounds of iteration until convergence, specifically as follows: First, the lower limit thresholds for transverse diameter and fresh weight, output in step S5, are used as initial preset values. These initial preset values ​​refer to the benchmark criteria obtained from the first round of standard sample statistics before iterative optimization. These initial preset values ​​originate from the statistical analysis results of the standard samples verified as fertile by the pulping method, specifically including the lower limit thresholds for transverse diameter and fresh weight calculated for each longitudinal diameter group interval. Setting the first output thresholds as initial preset values ​​aims to provide a reliable starting point for subsequent iterative corrections, avoiding slow convergence or directional deviations caused by random initialization. For example, if in the first round of statistics, the lower limit threshold for transverse diameter for a certain longitudinal diameter interval is 3.20 mm and the lower limit threshold for fresh weight is 0.220 g, then this set of data is directly used as the input benchmark for the first round of iteration. By establishing this initial state, it is ensured that the iterative process is based on verified real biometric data, thereby guaranteeing the accuracy of subsequent correction directions.

[0053] Then, the following steps are repeated at least two rounds: based on the current round's lower limit threshold for transverse diameter and lower limit threshold for fresh weight, step S3 is re-executed to obtain a new round of candidate fertile plants; steps S4 and S5 are executed on the new round of candidate fertile plants, and the updated lower limit thresholds for transverse diameter and fresh weight are output. This step constructs a closed-loop feedback mechanism, which continuously approaches the optimal discrimination criteria through multiple iterations.

[0054] Specifically, in each iteration, step S3 is first executed backtrackingly using the previous round's output thresholds for the current transverse diameter and fresh weight as screening criteria. This means the system no longer uses fixed preset values ​​but dynamically adjusts the screening thresholds, selecting plants that meet the latest threshold requirements from the original multi-data set as candidates for fertile plants in the new round. Due to the change in thresholds, the set of candidates for fertile plants in the new round is usually purer than the previous round, eliminating more suspected individuals at the critical value. Subsequently, the destructive sampling (squeezing method) verification in step S4 is strictly performed on these newly selected candidates for fertile plants, removing sterile plants and retaining confirmed fertile plants as the standard samples for the new round. This step is crucial, as it ensures that the data source entering the next round of statistics is always a high-purity true positive sample. Next, step S5 is executed based on this updated batch of standard samples to re-statistically analyze the distribution characteristics of their transverse diameter and fresh weight, thereby calculating the updated thresholds for transverse diameter and fresh weight.

[0055] For example, in the first iteration, the initial sample might contain a small amount of marginal data, leading to a low threshold (e.g., a lower limit of 3.20 mm for transverse diameter). After re-screening and validation, the second round of standard sample set removes some weakly fertile individuals, shifting the statistical distribution center to the right, and the calculated new threshold might increase to 3.40 mm. This cyclical process of screening-validation-statistics-updating ensures that the discrimination criteria gradually converge to the true fertility characteristics boundary of the germplasm resources as the number of iterations increases. By repeating the process at least two times, the random errors that may exist in the first round of sampling can be effectively eliminated, improving the robustness of the threshold model.

[0056] Finally, when the changes in the lower limit thresholds for transverse diameter and fresh weight output in two consecutive iterations are both less than the preset convergence threshold, the iteration stops, and the lower limit thresholds for transverse diameter and fresh weight output in the last iteration are used as the discrimination criteria for step S6. The preset convergence threshold is a criterion used to determine whether the model has reached a stable state; its value can be set according to actual accuracy requirements, for example, 0.05mm (for transverse diameter) and 0.020g (for fresh weight). The core logic of this step is to monitor the trend of threshold changes during iteration. When the threshold difference (i.e., the change amplitude) calculated in two consecutive iterations is simultaneously lower than the aforementioned preset convergence threshold, it indicates that the discrimination criteria have stabilized, and the gain from continuing iteration is negligible; at this point, the loop can be terminated.

[0057] Specifically, the system compares the difference between the lower limit threshold of transverse diameter and the lower limit threshold of fresh weight output from round N and round N-1. If both are within the allowable error range, convergence is considered achieved. At this point, the threshold output from round N (the last round) is locked and used as the authoritative standard for the final field discrimination in step S6. This dynamic stopping mechanism avoids overcomputation or underfitting problems that may be caused by a fixed number of iterations. For example, if the change in the transverse diameter threshold is only 0.01 mm and the change in the fresh weight threshold is only 0.005 g in rounds 3 and 4, both less than the preset values, the iteration stops and the threshold of round 4 is used for subsequent large-scale field screening. This ensures that the discrimination standard finally applied to production practice is fully optimized and highly stable, maximizing the balance between the false negative rate and the false negative rate.

[0058] This application achieves adaptive evolution of the discrimination threshold by introducing a multi-round iterative optimization mechanism. By using the output of the previous round as the input condition for the next round and continuously purifying the training sample set using the pulp squeezing method to verify the results, the lower limit thresholds for transverse diameter and fresh weight can gradually converge from the initial rough estimates to the precise optimal solutions. This feedback loop not only effectively overcomes the statistical bias that may be caused by environmental fluctuations or individual differences in a single sampling, but also enables the discrimination model to have self-correction capabilities. As the iteration process progresses, the screening criteria become increasingly closer to the true fertility characteristics of the variety under specific growth cycles, thereby improving the accuracy and reliability of subsequent whole-field discrimination steps and solving the technical problem that traditional fixed threshold methods are difficult to adapt to complex field environments.

[0059] This application embodiment also includes a spatial consistency correction step performed after identifying the plant as a fertile plant in step S6, specifically as follows: First, the location coordinates of each rapeseed plant in the field are obtained. These location coordinates refer to the geographic location information of each rapeseed plant on a two-dimensional plane, used to characterize the spatial relative relationship between plants. These coordinates are obtained through GPS positioning equipment, RTK high-precision positioning instruments, or feature point matching technology based on UAV aerial imagery. Specifically, during the data acquisition phase, in addition to collecting the longitudinal diameter, transverse diameter, and fresh weight of the flower buds, the latitude and longitude coordinates or relative coordinates (X, Y) of each plant relative to the field reference point are simultaneously recorded. For example, a handheld RTK receiver can be used to perform point measurements on each plant to obtain planar coordinate data with centimeter-level accuracy; alternatively, the center point pixel coordinates of each plant can be identified using UAV orthophoto images, and converted into actual physical coordinates based on the spatial resolution of the image. Obtaining these location coordinates provides fundamental data support for subsequent spatial neighborhood statistics, enabling the judgment process to consider not only the morphological indicators of individual flower buds but also the distribution of surrounding plants.

[0060] Secondly, taking each plant identified as fertile as the center, the number or proportion of other plants identified as fertile within its preset neighborhood radius is counted. If the number or proportion is lower than the preset neighborhood density threshold, the plant is removed from the fertile plant identification results and marked as a plant to be confirmed. The preset neighborhood radius is a spatial range set according to the row spacing and plant spacing of rapeseed planting, used to define the concept of proximity. This preset neighborhood radius can be set according to the actual planting density of the field, usually set to 1.5 to 2 times the row spacing, or fixed at 30cm to 50cm. The preset neighborhood density threshold is the critical value for determining whether a fertile plant is isolated. Specifically, for the target plant initially identified as fertile in step S6, the system draws a circle with its position coordinates as the center and the preset neighborhood radius, counts the number of other plants already identified as fertile falling within this circular area, and calculates the ratio of this number to the total number of plants in the area. If the number of fertile plants obtained from the statistics is less than 3, or the proportion of fertile plants is less than 20% (i.e., the preset neighborhood density threshold), then the target plant is considered to be spatially isolated, which does not conform to the biological rule that fertile plants are usually distributed in patches or rows. This is most likely a misjudgment caused by measurement error or local environmental anomalies. In this case, the plant is removed from the list of fertile plants and remarked as a plant to be confirmed, so that it can be manually verified or verified using a more rigorous pulping method.

[0061] Finally, taking each plant not identified as fertile as the center, the number or proportion of plants identified as fertile within its preset neighborhood radius is counted. If the number or proportion is higher than the preset neighborhood density threshold, the plant is marked as a candidate for supplementary judgment, and supplementary judgment is performed according to the following preset rules: the longitudinal diameter, transverse diameter, and fresh weight of the plant's flower buds are re-compared using a threshold that is relaxed compared to the lower limit thresholds for transverse diameter and fresh weight used in step S6; if the transverse diameter is greater than or equal to the relaxed lower limit threshold for transverse diameter and the fresh weight is greater than or equal to the relaxed lower limit threshold for fresh weight, the plant is supplementary to be classified as fertile; the relaxation range is 5% to 10% of the original threshold. Candidate supplementary judgment plants can refer to plants that were initially identified as infertile, but whose surrounding environment is highly suspected to be a fertile area. The relaxed threshold is a flexible standard set to correct for missed judgments caused by minor individual differences or measurement fluctuations. Specifically, for the target plant initially identified as non-fertile in step S6, the number or proportion of fertile plants within a preset neighborhood radius is counted, using its location coordinates as the center. If this proportion is higher than a preset neighborhood density threshold (e.g., more than 80% of the surrounding plants are fertile), then based on the ecological assumption of similar clustering, the target plant is considered to be highly likely to be fertile, and its failure to pass the initial screening may be due to the flower bud measurement value being slightly lower than the rigid threshold. At this time, the plant is marked as a candidate for supplementary judgment, and the supplementary judgment procedure is initiated. During the supplementary judgment process, the original lower limit thresholds for transverse diameter and fresh weight in step S6 are no longer used; instead, they are lowered by 5% to 10% respectively as new relaxed thresholds. For example, if the original lower limit threshold for transverse diameter is 4.00 mm and the original lower limit threshold for fresh weight is 0.150 g, then the relaxed lower limit threshold for transverse diameter can be set from 3.80 mm (reduced by 5%) to 3.60 mm (reduced by 10%), and the relaxed lower limit threshold for fresh weight can be set from 0.143 g to 0.135 g. If the measured transverse diameter and fresh weight of the flower buds of the candidate plant meet both of these relaxed conditions, then it will be finally approved as a fertile plant.

[0062] This application introduces a spatial consistency correction mechanism, combining the morphological indicators of individual flower buds with the spatial distribution characteristics of the population. By obtaining the location coordinates of the plants, a verification logic based on geographical proximity is constructed. On the one hand, for plants initially identified as fertile but spatially isolated, a preset neighborhood density threshold is used for screening to eliminate outliers that do not conform to aggregation patterns, thereby eliminating false alarms caused by random noise. On the other hand, for plants initially identified as infertile but located in high-concentration fertile areas, remedial discrimination is performed by dynamically relaxing the lower limits of transverse diameter and fresh weight (with a relaxation range of 5%–10%), effectively capturing missed alarms caused by measurement errors or minor individual variations. This dual strategy of removing isolated cases and filling in missed cases fully utilizes the biological characteristic that fertility traits in crop cultivation often have spatial autocorrelation, making the final discrimination results more reasonable and continuous in spatial distribution, and improving the overall accuracy and reliability of fertile plant identification in large-scale seed production fields.

[0063] It should be noted that, in this embodiment, when performing spatial consistency correction, the number or proportion of other plants identified as fertile plants within a preset neighborhood radius can be counted, with each plant identified as fertile as the center, or the number or proportion of plants identified as fertile plants within a preset neighborhood radius can be counted, with each plant not identified as fertile as the center. The specific choice can be made according to the needs.

[0064] Example 2 Based on the same inventive concept, this embodiment also provides a fertile plant identification system that can be used to implement the fertile plant identification method based on flower bud size and weight as in Embodiment 1, comprising: The data acquisition module is used to select some plants in the same growth cycle from the same field, and collect the longitudinal diameter, transverse diameter and fresh weight of at least one flower bud on the outermost ring of the main flower stalk of each selected plant to construct a multivariate dataset; each multivariate data in the multivariate dataset includes plant number, longitudinal diameter, transverse diameter and fresh weight. The grouping processing module is used to divide the longitudinal diameter data in the multivariate dataset into longitudinal diameter grouping intervals, and assign the corresponding plant number, transverse diameter and fresh weight to the corresponding sample set with the same longitudinal diameter. The initial screening and discrimination module is used to sort the samples in each longitudinal diameter set according to the transverse diameter data from largest to smallest, select the fresh weight data corresponding to the top-ranked transverse diameter, and further verify whether it meets the preset lower limit of fresh weight under the corresponding longitudinal diameter interval. Plants with the corresponding plant number whose fresh weight data passes the verification are recorded as candidate fertile plants. The fertility verification module is used to perform destructive sampling on the candidate fertile plants using the pulping method to identify their fertility. The candidate fertile plants confirmed as fertile by pulping are used as standard samples. The longitudinal diameter, transverse diameter, and fresh weight of at least one flower bud in the outermost ring of the main flower stalk of each standard sample are re-collected to construct a multivariate fertility dataset and a fertile sample set with the same longitudinal diameter. The multivariate fertility dataset includes longitudinal diameter, transverse diameter, and fresh weight. The threshold generation module is used to statistically analyze the distribution characteristics of transverse diameter and fresh weight in each set of fertile samples with the same longitudinal diameter, and output the lower limit threshold of transverse diameter and the lower limit threshold of fresh weight corresponding to different longitudinal diameter grouping intervals. The whole-field discrimination module is used to collect the longitudinal diameter, transverse diameter, and fresh weight of at least one flower bud on the outermost ring of the main flower stalk of each remaining plant in the field. Based on the longitudinal diameter data, the corresponding longitudinal diameter grouping interval, as well as the corresponding transverse diameter lower limit threshold and fresh weight lower limit threshold, are selected for comparison. If the transverse diameter is greater than or equal to the corresponding transverse diameter lower limit threshold and the fresh weight is greater than or equal to the corresponding fresh weight lower limit threshold, the plant is judged as a fertile plant.

Claims

1. A method for identifying fertile plants based on flower bud size and weight, characterized in that, Includes the following steps: Step S1: Select some plants in the same growth cycle from the same field, and collect the longitudinal diameter, transverse diameter, and fresh weight of at least one flower bud on the outermost ring of the main flower stalk of each selected plant to construct a multivariate dataset; each multivariate data in the multivariate dataset includes plant number, longitudinal diameter, transverse diameter, and fresh weight; the growth cycle refers to a specific developmental stage of the plant between the bolting stage and the large bud stage. Step S2: Divide the longitudinal diameter data in the multivariate dataset into longitudinal diameter grouping intervals, and assign the corresponding plant number, transverse diameter and fresh weight to the corresponding sample set with the same longitudinal diameter. Step S3: In each of the sample sets with the same longitudinal diameter, sort the transverse diameter data from largest to smallest, select the fresh weight data corresponding to the top-ranked transverse diameter, and further verify whether it meets the preset lower limit of fresh weight under the corresponding longitudinal diameter interval. Record the plants with the corresponding plant numbers whose fresh weight data passes the verification as candidate fertile plants. Step S4: Destructive sampling of the candidate fertile plants is performed using the pulping method to identify fertility. The candidate fertile plants that are identified as fertile through pulping are used as standard samples. Referring to steps S1 and S2, the longitudinal diameter, transverse diameter, and fresh weight of at least one flower bud on the outermost ring of the main flower stalk of each standard sample are re-collected to construct a multivariate fertile dataset and a fertile sample set with the same longitudinal diameter; the multivariate fertile dataset includes the longitudinal diameter, transverse diameter, and fresh weight; Step S5: In each set of fertile samples with the same longitudinal diameter, statistically analyze the distribution characteristics of transverse diameter and fresh weight, and output the lower limit threshold of transverse diameter and the lower limit threshold of fresh weight corresponding to different longitudinal diameter grouping intervals. Step S6, referring to step S1, collect the longitudinal diameter, transverse diameter and fresh weight of at least one flower bud on the outermost ring of the main flower stalk of each remaining plant in the field, and select the corresponding longitudinal diameter grouping interval and the corresponding transverse diameter lower limit threshold and fresh weight lower limit threshold for comparison based on the longitudinal diameter data. If the transverse diameter is greater than or equal to the corresponding lower limit threshold and the fresh weight is greater than or equal to the corresponding lower limit threshold, then the plant is identified as a fertile plant.

2. The method for determining fertile plants based on bud size and weight according to claim 1, characterized in that, In step S1, after collecting the longitudinal diameter, transverse diameter, and fresh weight of at least one flower bud on the outermost ring of the main flower stalk of each plant, outlier statistical analysis is performed on the longitudinal diameter, transverse diameter, and fresh weight to remove abnormal data. And / or, in step S4, after re-collecting the longitudinal diameter, transverse diameter, and fresh weight of at least one flower bud in the outermost ring of the main flower stalk of each standard sample, outlier statistical analysis is performed on the longitudinal diameter, transverse diameter, and fresh weight to remove abnormal data.

3. The method for determining fertile plants based on bud size and weight according to claim 1 or 2, characterized in that, Step S2, which involves dividing the longitudinal axis data in the multivariate dataset into longitudinal axis grouping intervals, includes: Obtain plant variety information from the field to be tested; Based on the plant variety information, the corresponding longitudinal diameter grouping interval division parameters are automatically matched; the division parameters include the grouping interval width and / or the number of groups; different plant varieties correspond to different division parameters.

4. The method for determining fertile plants based on bud size and weight according to claim 1 or 2, characterized in that, In step S3, the fresh weight data corresponding to the top 10% to 30% of the transverse diameters are selected. And / or, in step S3, the proportion of the top-ranked transverse diameters is dynamically determined based on the degree of dispersion of the transverse diameter data within the same longitudinal diameter sample set, specifically as follows: Calculate the standard deviation or interquartile range of the transverse diameter data within the same longitudinal diameter sample set; The larger the standard deviation or interquartile range, the smaller the proportion of the horizontal diameter with the highest ranking is selected; the smaller the standard deviation or interquartile range, the larger the proportion of the horizontal diameter with the highest ranking is selected. The value of the ratio ranges from 5% to 40%.

5. The method for determining fertile plants based on bud size and weight according to claim 1 or 2, characterized in that, In step S5, before statistically analyzing the distribution characteristics of transverse diameter and fresh weight, a confidence weight is assigned to each of the standard samples. The confidence weight is determined based on the transverse diameter ranking position of the standard sample in step S3 and / or the difference between its fresh weight and the preset lower limit of fresh weight: the higher the transverse diameter ranking and the larger the remaining fresh weight, the higher the confidence weight. The distribution characteristics of transverse diameter and fresh weight are calculated using weighted average or weighted percentile methods during statistical analysis, and the lower limit thresholds of transverse diameter and fresh weight are output accordingly.

6. The method for determining fertile plants based on bud size and weight according to claim 1 or 2, characterized in that, In step S5, the lower limit threshold of the transverse diameter corresponding to the same longitudinal diameter grouping interval is the average value, median value, or the transverse diameter data corresponding to the G percentile after sorting the transverse diameter data from largest to smallest, where the value of G ranges from 5 to 30. And / or, in step S5, the lower limit threshold of fresh weight corresponding to the same longitudinal diameter grouping interval is the average value, median value of the fresh weight data, or the fresh weight data corresponding to the H percentile after the fresh weight data is sorted from largest to smallest, and the value of H ranges from 5 to 30.

7. The method for determining fertile plants based on bud size and weight according to claim 1 or 2, characterized in that, Before step S6 is executed, the lower limit threshold for transverse diameter and the lower limit threshold for fresh weight output in step S5 are used as initial preset values, and the following steps are repeated at least two rounds: Based on the current round's lower limit threshold for transverse diameter and lower limit threshold for fresh weight, repeat step S3 to obtain a new round of candidate fertile plants; Perform steps S4 and S5 on the new round of candidate fertile plants, and output the updated lower limit threshold for transverse diameter and lower limit threshold for fresh weight. When the change range of the lower limit threshold of transverse diameter and the lower limit threshold of fresh weight output in two consecutive rounds of iteration is less than the preset convergence threshold, the iteration stops, and the lower limit threshold of transverse diameter and the lower limit threshold of fresh weight output in the last round are used as the discrimination criteria for step S6.

8. The method for determining fertile plants based on bud size and weight according to claim 1 or 2, characterized in that, In step S6, after identifying the plant as a fertile plant, spatial consistency correction is also performed: Obtain the location coordinates of each plant in the field; Centered on each plant identified as fertile, count the number or proportion of other plants identified as fertile within a preset neighborhood radius. If the number or proportion is lower than the preset neighborhood density threshold, the plant is removed from the fertile plant identification results and marked as a plant to be confirmed. And / or, taking each plant that was not identified as a fertile plant as the center, count the number or proportion of plants identified as fertile plants within its preset neighborhood radius; If the number or proportion exceeds a preset neighborhood density threshold, the plant is marked as a candidate for supplementary evaluation and is evaluated according to the following preset rules: The longitudinal diameter, transverse diameter, and fresh weight of the flower buds of the plant are re-compared using thresholds that are relaxed compared to the lower limit thresholds for transverse diameter and fresh weight used in step S6. If the transverse diameter is greater than or equal to the relaxed lower limit threshold for transverse diameter and the fresh weight is greater than or equal to the relaxed lower limit threshold for fresh weight, the plant is reclassified as a fertile plant. The relaxation range is 5% to 10% of the original threshold.

9. A fertile plant identification system based on flower bud size and weight, for implementing the fertile plant identification method based on flower bud size and weight as described in any one of claims 1 to 8, characterized in that, include: The data acquisition module is used to select some plants in the same growth cycle from the same field, and collect the longitudinal diameter, transverse diameter and fresh weight of at least one flower bud on the outermost ring of the main flower stalk of each selected plant to construct a multivariate dataset; each multivariate data in the multivariate dataset includes plant number, longitudinal diameter, transverse diameter and fresh weight. The grouping processing module is used to divide the longitudinal diameter data in the multivariate dataset into longitudinal diameter grouping intervals, and assign the corresponding plant number, transverse diameter and fresh weight to the corresponding sample set with the same longitudinal diameter. The initial screening and discrimination module is used to sort the samples in each longitudinal diameter set according to the transverse diameter data from largest to smallest, select the fresh weight data corresponding to the top-ranked transverse diameter, and further verify whether it meets the preset lower limit of fresh weight under the corresponding longitudinal diameter interval. Plants with the corresponding plant number whose fresh weight data passes the verification are recorded as candidate fertile plants. The fertility verification module is used to perform destructive sampling on the candidate fertile plants using the pulping method to identify fertility, and the candidate fertile plants that are confirmed to be fertile by pulping identification are used as standard samples. The longitudinal diameter, transverse diameter, and fresh weight of at least one flower bud in the outermost ring of the main flower stalk of each standard sample were re-collected to construct a multivariate fertility dataset and a fertile sample set with the same longitudinal diameter; the multivariate fertility dataset includes longitudinal diameter, transverse diameter, and fresh weight; The threshold generation module is used to statistically analyze the distribution characteristics of transverse diameter and fresh weight in each set of fertile samples with the same longitudinal diameter, and output the lower limit threshold of transverse diameter and the lower limit threshold of fresh weight corresponding to different longitudinal diameter grouping intervals. The whole-field discrimination module is used to collect the longitudinal diameter, transverse diameter, and fresh weight of at least one flower bud on the outermost ring of the main flower stalk of each remaining plant in the field. Based on the longitudinal diameter data, the corresponding longitudinal diameter grouping interval, as well as the corresponding transverse diameter lower limit threshold and fresh weight lower limit threshold, are selected for comparison. If the transverse diameter is greater than or equal to the corresponding transverse diameter lower limit threshold and the fresh weight is greater than or equal to the corresponding fresh weight lower limit threshold, the plant is judged as a fertile plant.

Images