A high-precision phenotype parameter automatic extraction method and system for rice growth stages

Abstract

This invention provides a high-precision automatic extraction method and system for phenotypic parameters of rice growth stages, belonging to the field of agricultural information technology. The invention identifies the rice growth stage by acquiring and preprocessing observational data containing canopy height information, and then matching phenotypic parameter definition rules from a rule base based on the growth stage. Plant height is determined by reference plane and height distribution statistics during the tillering or jointing stage. During the heading stage, panicle height and plant height are calculated separately by distinguishing between panicle and leaf tip. During the grain-filling stage, plant height is geometrically corrected based on lodging posture. The phenotypic parameter result set is then aggregated and output according to the growth stage, achieving unified semantics and stable calculation of plant height and panicle height across different growth stages, improving the accuracy of phenotypic parameters and comparability across growth stages.

Description

Technical Field

[0001] This invention relates to the field of agricultural information technology, and in particular to a method and system for automatic extraction of high-precision phenotypic parameters for rice growth stages. Background Technology

[0002] For a long time, rice phenotypic parameters have been primarily obtained through manual measurement or specialized equipment. A typical example is CN103791846B, which discloses a rice plant architecture measuring instrument and a method for measuring plant architecture parameters, applicable to the measurement of plant architecture and related parameters. With the application of 3D reconstruction and sensing technologies, schemes have emerged that reconstruct the 3D morphological structure of rice populations based on images or point clouds and extract structural trait parameters accordingly. For instance, CN106056665B discloses a method for the digital and visual reconstruction of the 3D morphological structure of rice populations, providing a 3D data foundation for subsequent calculations of phenotypic parameters such as plant height and canopy structure. At the field scale, the combination of UAV imagery and algorithms has been used to acquire phenotypic information such as coverage and canopy status, serving as an auxiliary basis for growth assessment and breeding selection.

[0003] Current research is evolving from single-phase extraction to continuous monitoring throughout the entire growth period. Utilizing UAV time-series imagery to automatically identify rice phenological or growth stage stages and generate staged maps has become an important direction. For example, Drones 2023 (Lu X et al.) proposed a rice phenological stage mapping method based on UAV imagery and deep learning. Simultaneously, research on remote sensing estimation of key phenotypic parameters, such as leaf area index (LAI), based on UAV imagery throughout the entire growth period is increasing. For instance, PlantMethods 2021 (Gong Y et al.) presented a remote sensing estimation framework for rice LAI covering the entire growth period and discussed modeling differences and stage-specific performance under different varietal conditions. Regarding organ-level phenotypes, panicle 3D reconstruction and trait extraction technologies are rapidly developing. The panicle is treated as an independent organ for 3D modeling and trait calculation. For example, PlantPhenomics 2024 (Yang X et al.) proposed PanicleNeRF for field panicle 3D reconstruction and trait extraction.

[0004] Existing automatic extraction methods generally fix phenotypic parameters of the same name to a single geometric meaning. For example, plant height is directly taken as the highest point in the point cloud or the maximum value of the canopy height model, without incorporating changes in organ composition and observable changes caused by variations in the growth period into the parameter definition. This leads to systematic biases across growth periods: during the tillering or jointing stages, the canopy is sparse, and background and outliers are more likely to dominate the highest point; during the heading stage, the panicle and leaf tip are mixed, and the object corresponding to the highest point is unstable; during the grain-filling stage, lodging causes tilting, resulting in an underestimation of height that is difficult to correct uniformly. While staged mapping methods based on UAV time-series imagery can output results for different growth periods, these results are usually not further used to constrain the mathematical definitions and calculation rules of phenotypic parameters such as plant height, resulting in a disconnect between staged information and parameter calculation. Furthermore, the sensitivity of parameter estimation throughout the entire growth period varies at different stages, especially in high-coverage stages where information saturation or amplified stage-specific biases are prone to occur, making it difficult to guarantee consistent accuracy across the entire cycle. The three-dimensional reconstruction method of the ear has proven that the ear can be extracted in three dimensions with high precision as an independent organ object. However, organ-level extraction often focuses on the ear's own traits and still lacks a consistent definition mechanism for the same parameters throughout the entire growth period, making it difficult to guarantee the comparability and high-precision stable output of phenotypic parameters across the growth period. Summary of the Invention

[0005] To overcome the shortcomings of existing technologies, the purpose of this invention is to provide a high-precision automatic extraction method and system for phenotypic parameters of rice growth stages. Through a phenotypic parameter definition mechanism based on growth stage conditions, the invention achieves unified semantics and stable calculation of plant height and panicle top height at different growth stages, significantly improving the accuracy of phenotypic parameters and their comparability across growth stages.

[0006] To achieve the above objectives, the present invention provides the following solution:

[0007] A method for high-precision automatic extraction of phenotypic parameters for rice growth stages includes:

[0008] Acquire observation data from the target rice sample area to form an observation dataset to be processed. The observation data includes height observation information used to characterize canopy height.

[0009] Preprocessing is performed on the observation dataset to be processed to obtain a preprocessed observation dataset;

[0010] Based on the preprocessed observation dataset, the growth period of the target rice was identified, and the growth period identification result was obtained.

[0011] Based on the reproductive period identification results, a set of phenotypic parameter definition rules matching the reproductive period identification results is determined from a preset reproductive period conditional parameter definition rule base;

[0012] When the growth period identification result indicates the tillering stage or the jointing stage, the reference surface of the ground or water surface is estimated based on the preprocessed observation dataset. The canopy height distribution is generated using the reference surface as the height reference, and the plant height is determined by the statistical upper quantile height of the canopy height distribution, thus obtaining the plant height data of the tillering stage or the jointing stage.

[0013] When the growth period identification result indicates the heading stage, the candidate point set of the panicle and the leaf tip point set are extracted based on the preprocessed observation dataset. The panicle top height is determined based on the candidate point set of the panicle and the plant height is determined based on the leaf tip point set, so as to obtain the plant height data and panicle top height data at the heading stage.

[0014] When the growth period identification result indicates the grain filling period, lodging detection is performed based on the preprocessed observation dataset to obtain lodging posture parameters, and the plant height calculated with the reference surface as the height reference is geometrically corrected based on the lodging posture parameters to obtain the plant height data during the grain filling period.

[0015] Based on the set of phenotypic parameter definition rules, the plant height data at the tillering or jointing stage, the plant height data at the heading stage, the panicle top height data at the heading stage, and the plant height data at the grain-filling stage are selected and collected according to the growth period identification results, and the phenotypic parameter result set is output; the phenotypic parameter result set includes at least the plant height and the panicle top height.

[0016] Preferably, preprocessing is performed on the observation dataset to be processed to obtain a preprocessed observation dataset, including:

[0017] Anomalies inconsistent with altitude observation information in the dataset to be processed are removed to obtain a quality-controlled observation dataset.

[0018] Spatial registration processing is performed on the quality-controlled observation dataset to place it in a unified spatial coordinate system, resulting in a registered observation dataset.

[0019] The registered observation dataset is denoised while retaining height observation information to obtain the preprocessed observation dataset.

[0020] Preferably, the growth stage of the target rice is identified based on the preprocessed observation dataset to obtain the growth stage identification result, including:

[0021] Based on the preprocessed observation dataset, a set of reproductive period features is extracted to characterize the reproductive period; the set of reproductive period features includes at least canopy height distribution features;

[0022] The set of reproductive period features is input into the reproductive period discrimination model, and the discrimination result of the candidate reproductive period set is output.

[0023] The reproductive period identification result is determined based on the discrimination result; the reproductive period identification result is one of the tillering stage, jointing stage, heading stage and grain filling stage.

[0024] Preferably, based on the reproductive period identification result, a set of phenotypic parameter definition rules matching the reproductive period identification result is determined from a preset reproductive period conditional parameter definition rule base, including:

[0025] In the defined rule base for the conditional parameters of the growth period, plant height definition rule entries are preset for the tillering stage and the jointing stage. The plant height definition rule entries declare the calculation relationship between the reference plane, the canopy height distribution and the height of the statistical upper quantile.

[0026] In the growth period conditional parameter definition rule base, plant height definition rule entries and ear top height definition rule entries are preset for the heading period. In the plant height definition rule entries and ear top height definition rule entries, the extraction requirements of ear candidate point set and leaf tip point set and the determination relationship between plant height and ear top height are declared.

[0027] In the rule base for defining conditional parameters during the growth period, a rule entry for geometric correction of plant height during the grain filling period is pre-set. In the rule entry for geometric correction of plant height, the calculation relationship between the lodging posture parameter and the geometric correction is declared.

[0028] Based on the reproductive period identification results, rule entries that match the reproductive period identification results are retrieved and output to form the phenotypic parameter definition rule set.

[0029] Preferably, estimating the reference surface of the ground or water surface based on the preprocessed observation dataset includes:

[0030] Based on the preprocessed observation dataset, a set of reference candidate points for the ground or water surface is extracted; the set of reference candidate points is a set of points whose height observation information satisfies a preset flatness condition.

[0031] Based on the aforementioned set of candidate reference points, a plane fitting process is performed to obtain the reference surface parameters;

[0032] The reference plane is determined based on the reference plane parameters, and the reference plane is used as the height reference for height calculation.

[0033] Preferably, the canopy height distribution is generated using the reference plane as a height reference, and the plant height is determined using the statistical upper quantile height of the canopy height distribution, including:

[0034] Using the reference plane as the height reference, the height difference between each canopy observation point in the preprocessed observation dataset and the reference plane is calculated to form a height difference sequence;

[0035] The canopy height distribution is generated based on the height difference sequence;

[0036] Perform quantile statistical processing on the height difference sequence, and take the height corresponding to the upper quantile of the height difference sequence as the statistical upper quantile height;

[0037] The plant height is obtained by using the height of the statistical upper quantile as the plant height, and then obtaining the plant height data at the tillering stage or jointing stage.

[0038] Preferably, the process of extracting candidate point sets for the panicle and leaf tip points based on the preprocessed observation dataset includes:

[0039] A candidate object feature set is constructed based on the preprocessed observation dataset; the candidate object feature set includes height features and morphological features.

[0040] Based on the candidate object feature set, object category discrimination processing is performed on the preprocessed observation dataset to obtain candidate categories for the ear and leaf tip;

[0041] The observation points belonging to the candidate category of the panicle are aggregated to form the candidate point set of the panicle, and the observation points belonging to the candidate category of the leaf tip are aggregated to form the candidate point set of the leaf tip.

[0042] Preferably, lodging detection is performed based on the preprocessed observation dataset to obtain lodging posture parameters, and the plant height calculated using the reference plane as the height reference is geometrically corrected based on the lodging posture parameters, including:

[0043] Based on the preprocessed observation dataset, lodging areas are identified, and lodging detection results are output.

[0044] The collapse posture parameters are calculated based on the collapse detection results; the collapse posture parameters include tilt angle parameters and tilt direction parameters.

[0045] The plant height is calculated using the aforementioned reference plane as the height reference.

[0046] Based on the tilt angle parameter, the plant height is calculated using geometric correction processing to obtain the plant height data during the grain-filling period.

[0047] Preferably, based on the set of phenotypic parameter definitions, the plant height data at the tillering or jointing stage, the plant height data at the heading stage, the panicle top height data at the heading stage, and the plant height data at the grain-filling stage are selected and collected according to the growth period identification results, and the phenotypic parameter result set is output, including:

[0048] When the result of the growth period identification is the tillering stage or the jointing stage, the plant height data of the tillering stage or the jointing stage is selected and associated with the result of the growth period identification to form a phenotypic parameter entry;

[0049] When the growth period identification result is the heading period, the plant height data and the ear top height data at the heading period are selected and associated with the growth period identification result to form phenotypic parameter entries.

[0050] When the result of the fertility period identification is the grain-filling stage, the plant height data of the grain-filling stage is selected and associated with the fertility period identification result to form a phenotypic parameter entry;

[0051] The generated phenotypic parameter entries are aggregated to obtain the phenotypic parameter result set, which is then output.

[0052] A high-precision automatic phenotypic parameter extraction system for rice growth stages includes:

[0053] The observation data acquisition unit is used to acquire observation data of the target rice sample area and form an observation dataset to be processed. The observation data includes height observation information used to characterize the canopy height.

[0054] An observation data preprocessing unit is used to perform preprocessing on the observation dataset to be processed to obtain a preprocessed observation dataset.

[0055] The growth period identification unit is used to identify the growth period of the target rice based on the preprocessed observation dataset and obtain the growth period identification result.

[0056] The parameter definition rule matching unit is used to determine, based on the reproductive period identification result, a set of phenotypic parameter definition rules that match the reproductive period identification result from a preset reproductive period conditional parameter definition rule library;

[0057] The tillering and jointing stage plant height calculation unit is used to estimate the ground or water surface reference surface based on the preprocessed observation dataset when the growth period identification result indicates the tillering stage or jointing stage. The unit generates the canopy height distribution using the reference surface as the height reference and determines the plant height using the statistical upper quantile height of the canopy height distribution, thereby obtaining the plant height data at the tillering stage or jointing stage.

[0058] The panicle-leaf separation and height calculation unit at the heading stage is used to extract the panicle candidate point set and leaf tip point set based on the preprocessed observation dataset when the growth period identification result indicates the heading stage, and to determine the panicle top height based on the panicle candidate point set and the plant height based on the leaf tip point set, respectively, to obtain the plant height data and panicle top height data at the heading stage.

[0059] The lodging correction plant height calculation unit during the grain filling period is used to perform lodging detection based on the preprocessed observation dataset to obtain lodging posture parameters when the growth period identification result indicates the grain filling period, and to perform geometric correction on the plant height calculated with the reference surface as the height reference based on the lodging posture parameters to obtain the plant height data during the grain filling period.

[0060] The phenotypic parameter collection and output unit is used to select and collect the plant height data at the tillering stage or jointing stage, the plant height data at the heading stage, the panicle top height data at the heading stage, and the plant height data at the grain filling stage according to the phenotypic parameter definition rule set and the corresponding growth period identification results, and output the phenotypic parameter result set; the phenotypic parameter result set includes at least the plant height and the panicle top height.

[0061] The present invention discloses the following technical effects:

[0062] This invention incorporates rice growth period identification results into the automatic extraction of phenotypic parameters and determines a set of matching growth period conditional parameter definition rules based on these results. This allows phenotypic parameter calculations to no longer rely on a single, fixed geometric definition, but rather to be differentiated according to the growth period stage. This avoids the systematic bias caused by semantic inconsistencies of phenotypic parameters with the same name at different growth periods in existing technologies, and significantly improves the comparability and consistency of phenotypic parameters across growth periods.

[0063] This invention estimates a reference surface on the ground or water surface during the tillering or jointing stage, generates a canopy height distribution based on the reference surface, and further determines the plant height using the statistical upper quantile height of the canopy height distribution. This effectively suppresses height misjudgment caused by local anomalies, background reflections, or noise points, and solves the problem in the prior art where the highest point is used as the definition of plant height, which is easily interfered with in the early growth stage. This improves the stability and robustness of plant height calculation.

[0064] This invention achieves explicit differentiation between panicle traits and plant traits by extracting candidate point sets for panicle and leaf tip points respectively during the heading stage, and determining panicle top height based on the candidate point set for panicle and plant height based on the leaf tip point set for plant height. This avoids the problems of unclear plant height boundaries and panicle-leaf overlap leading to height definition drift in the prior art, so that plant height and panicle top height at the heading stage have clear and stable physical meaning.

[0065] This invention obtains lodging posture parameters by performing lodging detection during the grain-filling stage, and performs geometric correction on the plant height calculated based on the reference plane based on the lodging posture parameters. This effectively compensates for the tilting effect caused by rice lodging, and solves the problem of systematic underestimation of plant height due to plant tilting in the middle and late growth stages of the existing technology, thereby improving the accuracy and practicality of plant height parameters during the grain-filling stage.

[0066] This invention selects and aggregates plant height and panicle height data obtained from different growth stages according to the results of growth stage identification, forming a unified set of phenotypic parameter results. This ensures that the output results have consistent definition logic and calculation basis throughout the entire growth period, avoiding the problems of data incomparability and analysis fragmentation caused by simple splicing of multi-stage parameters in the prior art. It provides a reliable, continuous and high-precision phenotypic data foundation for rice breeding evaluation and field management. Attached Figure Description

[0067] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0068] Figure 1 A flowchart of the method provided in an embodiment of the present invention;

[0069] Figure 2 This is a schematic diagram of a fifty-layer residual network model for reproductive period identification provided in an embodiment of the present invention.

[0070] Figure 3 This is a schematic diagram of the system structure provided in an embodiment of the present invention. Detailed Implementation

[0071] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0072] The purpose of this invention is to provide a high-precision automatic extraction method and system for rice growth stages, which eliminates systematic errors caused by differences in growth stages from the source of parameter definition, so that rice phenotypic parameters have consistency, reliability and engineering practical value throughout the entire growth period.

[0073] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0074] Figure 1 The method flowchart provided in the embodiments of the present invention is as follows: Figure 1 As shown, this invention provides a high-precision automatic extraction method for phenotypic parameters of rice growth stages, including:

[0075] Step 100: Obtain observation data from the target rice sample area to form an observation dataset to be processed; the observation data includes height observation information used to characterize canopy height;

[0076] Step 200: Perform preprocessing on the observation dataset to be processed to obtain the preprocessed observation dataset;

[0077] Step 300: Identify the growth period of the target rice based on the preprocessed observation dataset to obtain the growth period identification results;

[0078] Step 400: Based on the reproductive period identification results, determine the set of phenotypic parameter definition rules that match the reproductive period identification results from the preset reproductive period conditional parameter definition rule library;

[0079] Step 500: When the growth period identification result indicates the tillering stage or the jointing stage, estimate the reference surface of the ground or water surface based on the preprocessed observation dataset, generate the canopy height distribution with the reference surface as the height reference, and determine the plant height using the statistical upper quantile height of the canopy height distribution to obtain the plant height data of the tillering stage or the jointing stage.

[0080] Step 600: When the growth period identification result indicates the heading stage, extract the candidate point set of the panicle and the leaf tip point set based on the preprocessed observation dataset. Determine the panicle top height based on the candidate point set of the panicle and the plant height based on the leaf tip point set, respectively, to obtain the plant height data and panicle top height data at the heading stage.

[0081] Step 700: When the growth period identification result indicates the grain filling period, lodging detection is performed based on the preprocessed observation dataset to obtain lodging posture parameters, and the plant height calculated with the reference plane as the height reference is geometrically corrected based on the lodging posture parameters to obtain the plant height data during the grain filling period.

[0082] Step 800: Based on the set of phenotypic parameter definition rules, select and collect plant height data at the tillering or jointing stage, heading stage, panicle top height at heading stage, and grain-filling stage according to the growth period identification results, and output the phenotypic parameter result set; the phenotypic parameter result set shall at least include plant height and panicle top height.

[0083] Specifically, in this embodiment, obtaining observation data of the target rice sample area in step 100 refers to collecting information reflecting the vertical structure of the rice canopy under the natural growth state of the target rice sample area to form a dataset of observations to be processed for subsequent processing. Specifically, in this embodiment, observation positions are set above or around the target rice sample area to conduct multi-view or single-view observations. The obtained observation data at least includes information reflecting the height difference of the rice canopy relative to the ground or water surface. This type of information is referred to as "height observation information" in this embodiment, meaning data content that can be used to determine the relative height relationship between different observation points in the rice canopy. The height observation information can be reflected as the vertical distance distribution from each observation point to the reference surface. For example, multiple height value points can be formed through continuous sampling, so that no less than 100 effective height observation points can be obtained within the same sample area, thereby reflecting the overall height distribution characteristics of the canopy. To ensure the reliability of subsequent growth stage identification and plant height calculation, this embodiment preferably achieves a vertical resolution of 1 to 5 for the height observation information, enabling the differentiation of adjacent height differences. Simultaneously, the observation coverage within the same area is no less than 80% to avoid distortion of canopy height distribution due to local omissions. After collection, the aforementioned observation data are aggregated according to the sample area range, forming a dataset of observations to be processed, corresponding one-to-one with the target rice sample area. Each observation record is associated with at least one height observation value, used for subsequent generation of canopy height distribution, determination of statistical upper quantile height, and calculation of phenotypic parameters under different growth stage conditions.

[0084] Optionally, in this embodiment, the preprocessing of the observation dataset to be processed in step 200 refers to screening and standardizing abnormal data in the observation dataset to be processed that affect the accuracy of subsequent height calculation and reproductive period identification, while maintaining the validity of the height observation information. Specifically, this embodiment first performs a consistency check on each observation record in the observation dataset to be processed, and determines and removes observation records that are significantly inconsistent with the height observation information as abnormal data. The abnormal data includes data whose height values ​​exceed the reasonable growth range of the sample area or data whose local height mutations exceed a preset threshold. Through the abnormal data removal process, a quality-controlled observation dataset is formed, making the height observation information in the remaining observation data continuous and stable in numerical distribution. Preferably, the proportion of abnormal data is controlled below 5%, thereby reducing the interference of abnormal height values ​​on the construction of canopy height distribution.

[0085] After removing outlier data, this embodiment performs spatial registration on the quality-controlled observation dataset to place all observation records in the dataset under a unified spatial coordinate system, thereby eliminating spatial offsets introduced by different observation locations or times. Subsequently, based on the unified spatial coordinate system, denoising processing is performed on the registered observation dataset to weaken the impact of local random fluctuations on height observation information, while maintaining the overall trend of canopy height changes without smoothing them out. Through the above spatial registration and denoising processing, the preprocessed observation dataset is obtained, in which the height observation information is spatially consistent and numerically smooth. Preferably, the height difference between adjacent observation points is kept within a reasonable variation range of 1 to 5, thus providing a standardized data foundation for subsequent identification of growth period and calculation of plant height based on canopy height distribution.

[0086] In this embodiment, the step 300 of identifying the target rice growth stage based on the preprocessed observation dataset refers to extracting characteristic quantities that can distinguish different growth stage stages from the preprocessed observation dataset and determining the stage accordingly. Specifically, this embodiment constructs a growth stage feature set to characterize the growth stage based on the preprocessed observation dataset. The growth stage feature set includes at least canopy height distribution characteristics, which characterize the distribution pattern and concentration of canopy height within the sample area. Preferably, this embodiment selects no less than 100 effective height observation points within the same sample area, forms a canopy height distribution based on the effective height observation points, and extracts at least three statistical features from the canopy height distribution as the canopy height distribution characteristics. This ensures that the growth stage feature set can reflect the height morphological differences of the canopy from sparse to dense, from increasing height during jointing to significant heading, and from the grain-filling stage affected by lodging or senescence, thereby meeting the distinguishability requirements of different growth stage stages.

[0087] After extracting the reproductive period feature set, this embodiment inputs the reproductive period feature set into a reproductive period discrimination model to output the discrimination result of a candidate reproductive period set. The candidate reproductive period set includes at least the tillering stage, jointing stage, heading stage, and grain-filling stage. This embodiment uses the discrimination result as the basis for determining the reproductive period identification result. Preferably, the candidate reproductive period corresponding to the maximum confidence level is used as the reproductive period identification result. When the maximum confidence level is below 80, the reproductive period identification result is marked as requiring review to avoid misjudgment at the boundary stage, which could lead to deviations in the selection of subsequent phenotypic parameter definition rules. Through the above discrimination and determination process, the reproductive period identification result is obtained. The reproductive period identification result is one of the tillering stage, jointing stage, heading stage, and grain-filling stage, and is used to subsequently determine the matching phenotypic parameter definition rule set from the reproductive period conditional parameter definition rule base.

[0088] Exemplary, such as Figure 3As shown Figure 3 This is a schematic diagram of the structure of a 50-layer growth stage discrimination model of a residual network for growth stage identification in an embodiment of the present invention. In the figure, "Input Image / Height Distribution Map" represents the input data of the model, and its corresponding Chinese meaning is "input image or height distribution map", which is used to characterize the spatial distribution characteristics of the canopy height of the rice sample area generated by the preprocessed observation dataset; "Convolution Layer" represents the convolutional layer, and its corresponding Chinese meaning is "convolutional layer", which is used to perform local feature extraction on the input data and form an initial feature map; "ResidualBlock" represents the residual block, and its corresponding Chinese meaning is "residual module", which is used to extract higher-level abstract features while maintaining the continuity of feature transfer; "Residual Block Stacked 3, 4, 6, 3" represents the staged stacking structure of the residual module, and its corresponding Chinese meaning is "the residual modules are stacked in groups of three, four, six, and three", which is used to form the main feature extraction backbone of the 50 layers of the residual network; "Global Average Pooling" represents global average pooling, and its corresponding Chinese meaning is "global average pooling layer", which is used to compress the high-dimensional spatial feature map into a global feature vector; "Classification Output" represents the classification output, and its corresponding Chinese meaning is "classification output layer", which is used to output the discrimination results of each growth stage category. In the figure, "Tillering Stage", "Jointing Stage", "Heading Stage", and "Grain Filling Stage" respectively correspond to "tillering stage", "jointing stage", "heading stage", and "grain filling stage" in Chinese. During the training process, in this embodiment, the sample areas with known growth stage labels are used as training data, and the input image or height distribution map is input into the 50-layer growth stage discrimination model of the residual network. Through forward feature extraction and backward parameter update, the model gradually learns the difference features of different growth stages in the canopy height distribution pattern. After the training is completed, the model can discriminate the input height distribution map or image data in the inference stage and output the corresponding growth stage category as the growth stage identification result, which is used for the subsequent definition and calculation process of the phenotypic parameters conditional on the growth stage in the present invention.

[0089] Specifically, in this embodiment, step 400, which involves determining the matching set of phenotypic parameter definition rules from a preset fertility period conditional parameter definition rule base based on the fertility period identification result, refers to pre-setting the "computational semantics" of the phenotypic parameters in the form of rule entries. The fertility period identification result triggers the retrieval and assembly of corresponding rule entries, ensuring that phenotypic parameters with the same name have consistent mathematical meaning and comparability across different fertility periods. In this embodiment, the fertility period conditional parameter definition rule base is a searchable data structure, with rule entries as its basic unit. Each rule entry contains at least four types of fields: fertility period identifier, parameter name, input data requirements, and calculation determination method. The fertility period identifier indicates the applicable fertility period for the rule entry; the parameter name indicates the corresponding phenotypic parameter; the input data requirements declare the set of observation points or the type of height observation information required to execute the rule entry; and the calculation determination method declares the determination path and output format from the input data to the phenotypic parameter. Preferably, this embodiment associates each rule entry with at least one executable calculation determination method, allowing the rule entry to be directly used for calculating plant height and ear height in subsequent steps.

[0090] In the defined rule base for conditional parameters of the growth period, this embodiment pre-sets plant height definition rule entries for the tillering and jointing stages. These plant height definition rule entries explicitly declare the calculation relationship between the reference plane, canopy height distribution, and statistical upper quantile height. Specifically, the reference plane is defined as a reference plane used to describe the height benchmark of the ground or water surface. The canopy height distribution is defined as the statistical distribution of the height differences between each canopy observation point within the sample area and the reference plane. The statistical upper quantile height is defined as the height value corresponding to the upper proportion position in the set of height differences. This embodiment further declares the range of the upper proportion in the plant height definition rule entry, preferably from 90 to 99, and provides an example of a upper proportion of 95. In this case, the statistical upper quantile height corresponds to a height threshold where the proportion of observation points higher than or equal to this height value in the set of height differences is 5. By characterizing the representative height of the upper edge of the canopy through the statistical upper quantile height, excessive elevation of plant height by outlier points can be suppressed. The rule entry also states that the determination of plant height is to use the height of the statistical upper quantile as the output value of plant height during the tillering or jointing stages, thereby providing a stable and repeatable method for determining plant height during the tillering and jointing stages.

[0091] In the defined rule base for conditional parameters during the growth period, this embodiment pre-sets plant height and ear tip height definition rule entries for the heading stage, and declares the extraction requirements for the candidate ear point set and leaf tip point set, as well as the relationship between plant height and ear tip height in these rule entries. Specifically, the candidate ear point set is defined as a set of observation points that satisfy the ear morphological and height position characteristics, and the leaf tip point set is defined as a set of observation points that satisfy the leaf tip morphological and height position characteristics. The extraction requirements include obtaining at least 50 candidate ear observation points and at least 50 leaf tip observation points in the same region to ensure that the candidate ear point set and leaf tip point set can represent the upper edge of the corresponding organ's height. The rule entry defining ear top height states that the determination relationship for ear top height is to extract a set of height differences relative to the reference plane from the candidate point set of the ear, and select the upper representative height from the height difference set as the ear top height output. The upper representative height is preferably the height of the upper quantile, and the upper proportion is preferably 95 to 99 to reduce the impact of a single outlier on the ear top height. The rule entry defining plant height states that the determination relationship for plant height is to extract a set of height differences relative to the reference plane from the leaf tip point set, and select the upper representative height from the height difference set as the plant height output at the heading stage. The upper representative height is also preferably the height of the upper quantile. By binding ear top height and plant height to the upper representative height values ​​of different point sets, this embodiment achieves semantic separation between ear traits and plant traits at the heading stage, so that the same parameters at the heading stage have a clear and consistent determination method.

[0092] In the defined rule base for conditional parameters during the growth period, this embodiment specifies a pre-set geometric correction rule entry for plant height during the grain-filling stage. This rule entry declares the calculation relationship between the lodging posture parameter and the geometric correction. Specifically, the lodging posture parameter is defined as a set of parameters describing the plant's tilt state. This set of parameters includes at least a tilt angle parameter and a tilt direction parameter. The tilt angle parameter characterizes the degree of deviation of the plant's main axis from the vertical direction, and the tilt direction parameter characterizes the planar direction in which the tilt occurs. The plant height geometric correction rule entry declares that the geometric correction determination relationship is as follows: the plant height obtained with the reference plane as the height reference is used as the height calculation plant height, and the height calculation plant height is corrected according to the tilt angle parameter, so that the corrected plant height represents the equivalent height of the plant along the main axis direction; preferably, the tilt angle parameter takes a value range of 0 to 60. When the tilt angle parameter is 0, the corrected plant height is consistent with the height calculation plant height. When the tilt angle parameter increases, the corrected plant height increases according to the preset geometric relationship to compensate for the underestimation of vertical height caused by tilting, so that the plant height during the grain filling period still maintains the same semantic reference under lodging conditions as under non-lodging conditions.

[0093] After the above-mentioned rule entries are preset, this embodiment retrieves and outputs rule entries that match the growth period identification result, forming the phenotypic parameter definition rule set. Specifically, when the growth period identification result is the tillering stage or the jointing stage, plant height definition rule entries for the tillering stage or the jointing stage are retrieved and output; when the growth period identification result is the heading stage, plant height definition rule entries and ear tip height definition rule entries for the heading stage are retrieved and output; when the growth period identification result is the grain filling stage, plant height geometric correction rule entries for the grain filling stage are retrieved and output. The retrieved and output rule entries are assembled in the form of a set to form the phenotypic parameter definition rule set, which serves as the calculation basis for determining the reference surface, generating the canopy height distribution, extracting the ear candidate point set and the leaf tip point set, and performing lodging geometric correction in subsequent steps, thereby realizing consistent invocation and stable execution of the growth period conditional parameter definition in the process.

[0094] Specifically, in this embodiment, step 500 is used to determine plant height when the growth period identification result indicates the tillering or jointing stage. This includes determining the reference plane and constructing the canopy height distribution based on the reference plane, as well as extracting the height of the statistical upper quantile. In this embodiment, the "reference plane" is defined as a reference plane used to describe the height benchmark of the ground or water surface, used to eliminate the influence of terrain undulations or water surface disturbances on height calculation. The "canopy observation point" is defined as an observation record point in the preprocessed observation dataset that belongs to the rice canopy area and contains height observation information. To ensure the representativeness of plant height calculation at the tillering or jointing stage, this embodiment preferably obtains no less than 100 canopy observation points in the target sample area, and ensures that the spatial coverage of the canopy observation points within the sample area is no less than 80%, to avoid local omissions that cause the canopy height distribution to shift.

[0095] Regarding the determination of the reference surface, this embodiment first extracts a set of candidate reference points for the ground or water surface based on the preprocessed observation dataset. This set of candidate reference points is a collection of points whose height observation information satisfies a preset flatness condition. In this embodiment, the preset flatness condition is used to constrain the local height variation range of the candidate point set, preferably limiting it to a height variation range of no more than 2 within a preset neighborhood, so that the candidate reference point set can reflect the flat area characteristics of the ground or water surface. Simultaneously, it is preferable that the number of points in the candidate reference point set is not less than 50 to ensure the stability of subsequent fitting results. The candidate reference point set obtained through the above screening is used as input data for reference surface fitting, and is used to establish a unified height reference during the tillering or jointing stages.

[0096] After obtaining the baseline candidate point set, this embodiment performs plane fitting processing based on the baseline candidate point set to obtain baseline surface parameters. These parameters characterize the position and orientation of the baseline surface in a unified spatial coordinate system and are used for subsequent height difference calculations. In this embodiment, the plane fitting process involves performing a minimum residual fitting solution on the baseline candidate point set to minimize the overall deviation between the baseline candidate point set and the fitting plane. Preferably, after fitting, a consistency check is performed on the fitting residuals of the baseline candidate point set. If the concentration of the residuals does not meet a preset threshold, the baseline candidate point set is re-selected and fitted again to avoid a small number of outliers causing offsets to the baseline surface parameters. The baseline surface is determined based on the baseline surface parameters and used as the height reference for height calculation, enabling height observation information from different locations within the sample area to be compared under the same height reference.

[0097] Regarding the determination of canopy height distribution and plant height, this embodiment uses the reference surface as the height reference, calculates the height difference between each canopy observation point in the preprocessed observation dataset and the reference surface, forming a height difference sequence, and generates the canopy height distribution based on the height difference sequence. In this embodiment, the canopy height distribution is used to characterize the statistical distribution of canopy height differences within the sample area. Preferably, the height difference sequence is divided into no fewer than 20 height intervals to form a distribution statistics, thus balancing distribution details and stability. Subsequently, this embodiment performs quantile statistical processing on the height difference sequence, taking the height corresponding to the upper quantile of the height difference sequence as the statistical upper quantile height. The upper quantile is used to characterize the representative height of the upper edge of the canopy, preferably ranging from 90 to 99, and exemplarily taken as 95 to reduce the influence of a very small number of outliers on plant height. Finally, the statistical upper quantile height is used as the plant height to obtain the plant height data at the tillering or jointing stage, ensuring that the plant height at the tillering or jointing stage still has a stable basis for determination even in the presence of noise.

[0098] Furthermore, step 600 of this embodiment is used to differentiate between panicle traits and plant traits when the growth period identification result indicates the heading stage, in order to avoid semantic confusion of parameters caused by unclear boundaries of the canopy height in the panicle. Specifically, this embodiment constructs a candidate object feature set based on the preprocessed observation dataset. The candidate object feature set is used to describe the criteria for distinguishing the organ types of different observation points, and includes at least height features and morphological features. The height features are used to characterize the height position of the observation point relative to the reference plane, preferably including the relative ranking position of the observation point height in the sample area height distribution; the morphological features are used to characterize the morphological attributes of the observation point in local space, preferably including features such as local curvature changes, point density, and spatial extension direction. By combining height features and morphological features, this embodiment enables the candidate object feature set to simultaneously reflect the characteristics of the panicle, which is usually located at the upper edge of the canopy and has a relatively concentrated structure, and the structural characteristics of the leaf tips, which are usually slender and dispersed, thereby providing a sufficient information basis for subsequent object category discrimination.

[0099] After obtaining the candidate object feature set, this embodiment performs object category discrimination processing on the preprocessed observation dataset based on the candidate object feature set to distinguish between ear-shaped candidate categories and leaf-tip candidate categories. In this embodiment, the object category discrimination processing refers to determining the category of the observation point based on the value combination relationship of each feature in the candidate object feature set. Preferably, the ear-shaped candidate category and the leaf-tip candidate category are separable in height distribution range and morphological feature space. After discrimination, the observation points determined to be ear-shaped candidate categories are aggregated to form an ear-shaped candidate point set, and the observation points determined to be leaf-tip candidate categories are aggregated to form a leaf-tip candidate point set. To ensure the stability of subsequent ear-top height and plant height calculations, this embodiment preferably ensures that the ear-shaped candidate point set and the leaf-tip candidate point set each contain no less than 50 valid observation points, and that the spatial distribution of the two types of point sets within the sample area covers the upper edge of the canopy. Through the above method, this embodiment forms mutually independent ear-shaped candidate point sets and leaf-tip candidate point sets during the heading stage, providing a clear data basis for determining ear-top height and plant height respectively.

[0100] Furthermore, in this embodiment, step 700 is used to correct the underestimation of plant height caused by lodging when the growth period identification result indicates the grain-filling stage. Specifically, this embodiment identifies lodging areas based on the preprocessed observation dataset. Lodging areas refer to regions within the sample area where the height distribution pattern, spatial extension direction, or local structural features have significantly changed compared to the upright growth state. By analyzing the height change trend and spatial arrangement characteristics of each observation point in the preprocessed observation dataset, regions with lodging characteristics are determined and lodging detection results are output. These lodging detection results are used to identify the spatial location range affected by lodging in the sample area during the grain-filling stage. To ensure the stability of lodging detection, this embodiment preferably ensures that the continuous coverage area of ​​the lodging area within the sample area is not less than 10% of the total sample area, thereby avoiding misjudgment of local accidental disturbances as lodging.

[0101] After obtaining the lodging detection results, this embodiment calculates lodging posture parameters based on these results. These parameters describe the tilting state of the plant during the grain-filling stage and include at least a tilt angle parameter and a tilt direction parameter. The tilt angle parameter characterizes the degree of deviation of the plant's main axis from the vertical direction, preferably ranging from 0 to 60 degrees. The tilt direction parameter characterizes the plane direction in which the plant tilts, distinguishing the influence of different lodging directions on height measurement. By statistically analyzing the directional changes in the height distribution of observation points within the lodging area, this embodiment obtains lodging posture parameters that represent the overall lodging state, thus providing a quantitative basis for subsequent geometric correction.

[0102] After calculating the lodging posture parameters, this embodiment first uses the reference plane as the height reference and performs height calculations on the observation points belonging to the canopy region in the preprocessed observation dataset to obtain the calculated plant height, which is the plant height value without considering the lodging effect. Subsequently, this embodiment performs geometric correction processing on the calculated plant height based on the tilt angle parameter, so that the corrected plant height can represent the equivalent height of the plant along the main axis. Preferably, when the tilt angle parameter is 0, the corrected plant height is consistent with the calculated plant height; when the tilt angle parameter increases, the corrected plant height increases accordingly to compensate for the underestimation of vertical height caused by lodging. Through the above geometric correction processing, the plant height data during the grain-filling stage is obtained, so that the plant height during the grain-filling stage still has the same physical meaning and comparability as the upright growth state under the condition of lodging.

[0103] Furthermore, in this embodiment, step 800 is used to select and uniformly aggregate the phenotypic parameters obtained under different growth stage conditions according to the growth stage identification results, so as to form a phenotypic parameter result set that can be directly used for subsequent analysis and application. The set of phenotypic parameter definition rules in this embodiment is used to constrain the parameter types and semantics output at different growth stage stages, so that parameters with the same name have consistent meanings when compared across growth stages. In this embodiment, the phenotypic parameter result set is a data set aggregated with sample areas as the basic unit. Preferably, the sample area identifier and growth stage identification results are used as association fields, and plant height and ear height are included as at least one parameter field, thereby ensuring that the result set can reflect the key phenotypic status of the sample area under the corresponding growth stage.

[0104] When the growth period identification result is the tillering stage or the jointing stage, this embodiment selects the plant height data at the tillering stage or the jointing stage according to the phenotypic parameter definition rule set, and associates the plant height data at the tillering stage or the jointing stage with the growth period identification result to form a phenotypic parameter entry. In this embodiment, the phenotypic parameter entry refers to a data record that provides a structured expression of the phenotypic parameter output of a single sample area at a single growth period. The phenotypic parameter entry includes at least a sample area identifier, a growth period identification result, and a plant height field, wherein the plant height field takes the value of the plant height data at the tillering stage or the jointing stage. Preferably, when forming the phenotypic parameter entry, a consistency check is performed on the value range of the plant height field to ensure that the numerical variation of the plant height field is within a reasonable range, thereby avoiding individual abnormal outputs from entering the result set.

[0105] When the growth period identification result is the heading stage, this embodiment selects the plant height data and the ear tip height data at the heading stage according to the phenotypic parameter definition rule set, and associates the plant height data and ear tip height data at the heading stage with the growth period identification result to form a phenotypic parameter entry. This phenotypic parameter entry includes at least a sample area identifier, a growth period identification result, a plant height field, and an ear tip height field, wherein the plant height field takes the value of the plant height data at the heading stage, and the ear tip height field takes the value of the ear tip height data at the heading stage. Preferably, when forming the phenotypic parameter entry, a relative relationship check is performed on the plant height field and the ear tip height field to ensure that the ear tip height field is not less than the plant height field, thereby guaranteeing that the relative relationship between the upper edge height of the ear and the upper edge height of the leaf tip during the heading stage conforms to the height semantics under the point set definition.

[0106] When the growth period identification result is the grain-filling stage, this embodiment selects the plant height data of the grain-filling stage according to the phenotypic parameter definition rule set, and associates the plant height data of the grain-filling stage with the growth period identification result to form a phenotypic parameter entry. The phenotypic parameter entry includes at least a sample area identifier, a growth period identification result, and a plant height field, wherein the plant height field takes the value of the plant height data of the grain-filling stage. Subsequently, this embodiment performs aggregation processing on the formed phenotypic parameter entries, using the sample area identifier as the aggregation object to aggregate multiple phenotypic parameter entries of the same growth period in the same sample area. Preferably, the aggregated phenotypic parameter result set contains at least 3 phenotypic parameter entries corresponding to the growth period, thereby forming a structured result that can reflect the phenotypic changes of the sample area across the growth period. Finally, the phenotypic parameter result set is output, so that the phenotypic parameter result set includes at least plant height and ear top height and corresponds one-to-one with the growth period identification result.

[0107] Corresponding to the above methods, such as Figure 3 As shown, this embodiment also provides a high-precision automatic extraction system for phenotypic parameters of rice growth stages, including:

[0108] The observation data acquisition unit is used to acquire observation data of the target rice sample area and form an observation dataset to be processed. The observation data includes height observation information used to characterize the canopy height.

[0109] An observation data preprocessing unit is used to perform preprocessing on the observation dataset to be processed to obtain a preprocessed observation dataset.

[0110] The growth period identification unit is used to identify the growth period of the target rice based on the preprocessed observation dataset and obtain the growth period identification result.

[0111] The parameter definition rule matching unit is used to determine, based on the reproductive period identification result, a set of phenotypic parameter definition rules that match the reproductive period identification result from a preset reproductive period conditional parameter definition rule library;

[0112] The tillering and jointing stage plant height calculation unit is used to estimate the ground or water surface reference surface based on the preprocessed observation dataset when the growth period identification result indicates the tillering stage or jointing stage. The unit generates the canopy height distribution using the reference surface as the height reference and determines the plant height using the statistical upper quantile height of the canopy height distribution, thereby obtaining the plant height data at the tillering stage or jointing stage.

[0113] The panicle-leaf separation and height calculation unit at the heading stage is used to extract the panicle candidate point set and leaf tip point set based on the preprocessed observation dataset when the growth period identification result indicates the heading stage, and to determine the panicle top height based on the panicle candidate point set and the plant height based on the leaf tip point set, respectively, to obtain the plant height data and panicle top height data at the heading stage.

[0114] The lodging correction plant height calculation unit during the grain filling period is used to perform lodging detection based on the preprocessed observation dataset to obtain lodging posture parameters when the growth period identification result indicates the grain filling period, and to perform geometric correction on the plant height calculated with the reference surface as the height reference based on the lodging posture parameters to obtain the plant height data during the grain filling period.

[0115] The phenotypic parameter collection and output unit is used to select and collect the plant height data at the tillering stage or jointing stage, the plant height data at the heading stage, the panicle top height data at the heading stage, and the plant height data at the grain filling stage according to the phenotypic parameter definition rule set and the corresponding growth period identification results, and output the phenotypic parameter result set; the phenotypic parameter result set includes at least the plant height and the panicle top height.

[0116] The beneficial effects of this invention are as follows:

[0117] (1) This invention introduces the results of the fertility period identification into the automatic extraction process of phenotypic parameters, and drives the selection of the set of rules for defining phenotypic parameters with the results of the fertility period identification. This makes the calculation semantics of plant height and ear height no longer fixed as a single geometric meaning, but switches the calculation determination method according to preset rules under different fertility periods. This solves the problem of the meaning drift of the same parameter across fertility periods in the prior art, and realizes the unified semantics and comparability of parameters across fertility periods. This is one of the core innovations of this invention.

[0118] (2) In the tillering or jointing stage, the present invention uses the ground or water surface as the height reference, and determines the plant height based on the canopy height distribution and the height of the statistical upper quantile. This avoids the defect that the plant height is easily affected by outliers or local noise when the highest point is used as the plant height, so that the plant height output in the early growth stage has stronger stability and repeatability, thereby improving the anti-interference ability and statistical reliability of the plant height parameter.

[0119] (3) In the heading stage, the present invention introduces the separation of the candidate point set of the panicle and the leaf tip point set, and determines the panicle top height based on the candidate point set of the panicle and the plant height based on the leaf tip point set, so that the panicle traits and plant traits can be clearly distinguished at both the data level and the calculation level. This avoids the problem of the panicle and leaf overlap causing uncertainty in the plant height boundary, and makes the plant height and panicle top height at the heading stage have clear calculation objects and stable physical meanings, thereby improving the extraction accuracy of key traits at the heading stage.

[0120] (4) The present invention obtains lodging posture parameters based on lodging detection during the grain filling period, and performs geometric correction on the plant height calculated based on the reference plane as the height reference, thereby making up for the deficiency of underestimation of vertical height caused by lodging in the middle and late stages, so that the plant height during the grain filling period still maintains the semantic reference and comparability with the upright state under lodging conditions, thereby improving the applicability and reliability of the phenotypic parameters in the middle and late stages under real field conditions.

[0121] (5) Based on the set of phenotypic parameter definition rules, this invention selects and collects plant height data and ear top height data of each stage according to the results of the growth period identification to form a unified result set, so that the output not only includes parameter values, but also implicitly contains definition rules that match the growth period, thereby avoiding statistical breaks and interpretation ambiguities caused by simple splicing of multi-stage parameters; Overall, this invention couples growth period information and parameter definition mechanism into the same closed-loop process, reduces systematic errors from the source, and realizes high-precision, stable and comparable automatic extraction of phenotypic parameters throughout the entire growth period.

[0122] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0123] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for high-precision automatic extraction of phenotypic parameters for rice growth stages, characterized in that, include: Acquire observation data from the target rice sample area to form an observation dataset to be processed. The observation data includes height observation information used to characterize canopy height. Preprocessing is performed on the observation dataset to be processed to obtain a preprocessed observation dataset; Based on the preprocessed observation dataset, the growth period of the target rice was identified, and the growth period identification result was obtained. Based on the reproductive period identification results, a set of phenotypic parameter definition rules matching the reproductive period identification results is determined from a preset reproductive period conditional parameter definition rule base; When the growth period identification result indicates the tillering stage or the jointing stage, the reference surface of the ground or water surface is estimated based on the preprocessed observation dataset. The canopy height distribution is generated using the reference surface as the height reference, and the plant height is determined by the statistical upper quantile height of the canopy height distribution, thus obtaining the plant height data of the tillering stage or the jointing stage. When the growth period identification result indicates the heading stage, the candidate point set of the panicle and the leaf tip point set are extracted based on the preprocessed observation dataset. The panicle top height is determined based on the candidate point set of the panicle and the plant height is determined based on the leaf tip point set, so as to obtain the plant height data and panicle top height data at the heading stage. When the growth period identification result indicates the grain filling period, lodging detection is performed based on the preprocessed observation dataset to obtain lodging posture parameters, and the plant height calculated with the reference surface as the height reference is geometrically corrected based on the lodging posture parameters to obtain the plant height data during the grain filling period. Based on the set of phenotypic parameter definition rules, the plant height data at the tillering or jointing stage, the plant height data at the heading stage, the panicle top height data at the heading stage, and the plant height data at the grain-filling stage are selected and collected according to the growth period identification results, and the phenotypic parameter result set is output; the phenotypic parameter result set includes at least the plant height and the panicle top height.

2. The method for high-precision automatic extraction of phenotypic parameters for rice growth stages according to claim 1, characterized in that, Preprocessing is performed on the observation dataset to be processed to obtain a preprocessed observation dataset, including: Anomalies inconsistent with altitude observation information in the dataset to be processed are removed to obtain a quality-controlled observation dataset. Spatial registration processing is performed on the quality-controlled observation dataset to place it in a unified spatial coordinate system, resulting in a registered observation dataset. The registered observation dataset is denoised while retaining height observation information to obtain the preprocessed observation dataset.

3. The method for high-precision automatic extraction of phenotypic parameters for rice growth stages according to claim 1, characterized in that, Based on the preprocessed observation dataset, the growth period of the target rice was identified, and the growth period identification results were obtained, including: Based on the preprocessed observation dataset, a set of reproductive period features is extracted to characterize the reproductive period; the set of reproductive period features includes at least canopy height distribution features; The set of reproductive period features is input into the reproductive period discrimination model, and the discrimination result of the candidate reproductive period set is output. The reproductive period identification result is determined based on the discrimination result; the reproductive period identification result is one of the tillering stage, jointing stage, heading stage and grain filling stage.

4. The method for high-precision automatic extraction of phenotypic parameters for rice growth stages according to claim 1, characterized in that, Based on the reproductive period identification result, a set of phenotypic parameter definition rules matching the reproductive period identification result is determined from a preset reproductive period conditional parameter definition rule base, including: In the defined rule base for the conditional parameters of the growth period, plant height definition rule entries are preset for the tillering stage and the jointing stage. The plant height definition rule entries declare the calculation relationship between the reference plane, the canopy height distribution and the height of the statistical upper quantile. In the growth period conditional parameter definition rule base, plant height definition rule entries and ear top height definition rule entries are preset for the heading period. In the plant height definition rule entries and ear top height definition rule entries, the extraction requirements of ear candidate point set and leaf tip point set and the determination relationship between plant height and ear top height are declared. In the rule base for defining conditional parameters during the growth period, a rule entry for geometric correction of plant height during the grain filling period is pre-set. In the rule entry for geometric correction of plant height, the calculation relationship between the lodging posture parameter and the geometric correction is declared. Based on the reproductive period identification results, rule entries that match the reproductive period identification results are retrieved and output to form the phenotypic parameter definition rule set.

5. The method for high-precision automatic extraction of phenotypic parameters for rice growth stages according to claim 1, characterized in that, Estimating a reference surface for the ground or water surface based on the preprocessed observation dataset includes: Based on the preprocessed observation dataset, a set of reference candidate points for the ground or water surface is extracted; the set of reference candidate points is a set of points whose height observation information satisfies a preset flatness condition. Based on the aforementioned set of candidate reference points, a plane fitting process is performed to obtain the reference surface parameters; The reference plane is determined based on the reference plane parameters, and the reference plane is used as the height reference for height calculation.

6. The method for high-precision automatic extraction of phenotypic parameters for rice growth stages according to claim 1, characterized in that, Generating a canopy height distribution using the aforementioned reference plane as a height reference, and determining plant height using the statistical upper quantile height of the canopy height distribution, including: Using the reference plane as the height reference, the height difference between each canopy observation point in the preprocessed observation dataset and the reference plane is calculated to form a height difference sequence; The canopy height distribution is generated based on the height difference sequence; Perform quantile statistical processing on the height difference sequence, and take the height corresponding to the upper quantile of the height difference sequence as the statistical upper quantile height; The plant height is obtained by using the height of the statistical upper quantile as the plant height, and then obtaining the plant height data at the tillering stage or jointing stage.

7. The method for high-precision automatic extraction of phenotypic parameters for rice growth stages according to claim 1, characterized in that, Based on the preprocessed observation dataset, candidate point sets for the panicle and leaf tip are extracted, including: A candidate object feature set is constructed based on the preprocessed observation dataset; the candidate object feature set includes height features and morphological features. Based on the candidate object feature set, object category discrimination processing is performed on the preprocessed observation dataset to obtain candidate categories for the ear and leaf tip; The observation points belonging to the candidate category of the panicle are aggregated to form the candidate point set of the panicle, and the observation points belonging to the candidate category of the leaf tip are aggregated to form the candidate point set of the leaf tip.

8. The method for high-precision automatic extraction of phenotypic parameters for rice growth stages according to claim 1, characterized in that, Lodging detection is performed based on the preprocessed observation dataset to obtain lodging posture parameters, and the plant height calculated using the reference plane as the height reference is geometrically corrected based on the lodging posture parameters, including: Based on the preprocessed observation dataset, lodging areas are identified, and lodging detection results are output. The collapse posture parameters are calculated based on the collapse detection results; the collapse posture parameters include tilt angle parameters and tilt direction parameters. The plant height is calculated using the aforementioned reference plane as the height reference. Based on the tilt angle parameter, the plant height is calculated using geometric correction processing to obtain the plant height data during the grain-filling period.

9. The method for high-precision automatic extraction of phenotypic parameters for rice growth stages according to claim 1, characterized in that, Based on the set of phenotypic parameter definitions, the plant height data at the tillering or jointing stage, the plant height data at the heading stage, the panicle top height data at the heading stage, and the plant height data at the grain-filling stage are selected and collected according to the growth period identification results. The phenotypic parameter result set is then output, including: When the result of the growth period identification is the tillering stage or the jointing stage, the plant height data of the tillering stage or the jointing stage is selected and associated with the result of the growth period identification to form a phenotypic parameter entry; When the growth period identification result is the heading period, the plant height data and the ear top height data at the heading period are selected and associated with the growth period identification result to form phenotypic parameter entries. When the result of the fertility period identification is the grain-filling stage, the plant height data of the grain-filling stage is selected and associated with the fertility period identification result to form a phenotypic parameter entry; The generated phenotypic parameter entries are aggregated to obtain the phenotypic parameter result set, which is then output.

10. A high-precision automatic extraction system for phenotypic parameters of rice during its growth period, characterized in that, include: The observation data acquisition unit is used to acquire observation data of the target rice sample area and form an observation dataset to be processed. The observation data includes height observation information used to characterize the canopy height. An observation data preprocessing unit is used to perform preprocessing on the observation dataset to be processed to obtain a preprocessed observation dataset. The growth period identification unit is used to identify the growth period of the target rice based on the preprocessed observation dataset and obtain the growth period identification result. The parameter definition rule matching unit is used to determine, based on the reproductive period identification result, a set of phenotypic parameter definition rules that match the reproductive period identification result from a preset reproductive period conditional parameter definition rule library; The tillering and jointing stage plant height calculation unit is used to estimate the ground or water surface reference surface based on the preprocessed observation dataset when the growth period identification result indicates the tillering stage or jointing stage. The unit generates the canopy height distribution using the reference surface as the height reference and determines the plant height using the statistical upper quantile height of the canopy height distribution, thereby obtaining the plant height data at the tillering stage or jointing stage. The panicle-leaf separation and height calculation unit at the heading stage is used to extract the panicle candidate point set and leaf tip point set based on the preprocessed observation dataset when the growth period identification result indicates the heading stage, and to determine the panicle top height based on the panicle candidate point set and the plant height based on the leaf tip point set, respectively, to obtain the plant height data and panicle top height data at the heading stage. The lodging correction plant height calculation unit during the grain filling period is used to perform lodging detection based on the preprocessed observation dataset to obtain lodging posture parameters when the growth period identification result indicates the grain filling period, and to perform geometric correction on the plant height calculated with the reference surface as the height reference based on the lodging posture parameters to obtain the plant height data during the grain filling period. The phenotypic parameter collection and output unit is used to select and collect the plant height data at the tillering stage or jointing stage, the plant height data at the heading stage, the panicle top height data at the heading stage, and the plant height data at the grain filling stage according to the phenotypic parameter definition rule set and the corresponding growth period identification results, and output the phenotypic parameter result set; the phenotypic parameter result set includes at least the plant height and the panicle top height.

Images