Seedling grading methods, computer equipment and readable storage media

Abstract

The application discloses a seedling grading method, a computer device and a readable storage medium, and belongs to the field of agricultural informatization. The method comprises the following steps: determining a time sequence corresponding to a target seedling; the time sequence is obtained by performing relevant processing on multi-modal data; the multi-modal data is obtained by analyzing parameters after collecting the parameters of the target seedling; based on the time sequence, a growth prediction model is called to generate a target index value of a key growth index of the target seedling; and the target seedling is graded based on the target index value. The method can improve the survival rate of seedlings, thereby reducing the agricultural planting cost.

Description

Technical Field

[0001] This application belongs to the field of agricultural informatization, and specifically relates to a seedling grading method, computer equipment, and readable storage medium. Background Technology

[0002] Currently, in various seedling-related scenarios such as greenhouse seedling cultivation, breeding factories, and smart greenhouses, seedlings are usually graded in order to ensure uniform and robust seedlings.

[0003] However, even with seedling grading, the survival rate of seedlings remains low, increasing agricultural planting costs. Summary of the Invention

[0004] The purpose of this application is to provide a seedling grading method that can solve the problem of how to reduce agricultural planting costs.

[0005] To solve the above-mentioned technical problems, this application is implemented as follows:

[0006] In a first aspect, embodiments of this application provide a seedling grading method, the method comprising:

[0007] The time series corresponding to the target seedling is determined; the time series is obtained by performing correlation processing on multimodal data; the multimodal data is obtained by collecting parameters from the target seedling and then analyzing those parameters.

[0008] Based on the time series, the growth prediction model is invoked to generate target index values ​​for the key growth indicators of the target seedling.

[0009] The target seedlings are graded based on the target index values.

[0010] In a second aspect, embodiments of this application provide a computer device including a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the steps of the method described in the first aspect.

[0011] Thirdly, embodiments of this application provide a computer-readable storage medium on which a program or instructions are stored, which, when executed by a processor, implement the steps of the method described in the first aspect.

[0012] Fourthly, embodiments of this application also provide a computer program product, including a computer program that, when executed by a processor, implements the steps of the method described in the first aspect.

[0013] This application proposes a seedling grading method, which includes determining the time series corresponding to the target seedling; obtaining the time series through correlation processing of multimodal data; obtaining the multimodal data by collecting and analyzing parameters of the target seedling; based on the time series, calling a growth prediction model to generate target index values ​​for key growth indicators of the target seedling; and grading the target seedling based on the target index values. In other words, this method, by accurately grading seedlings, avoids misjudging seedlings with poor growth as those with good growth, thereby ensuring uniform and robust seedlings, improving seedling survival rate, and reducing agricultural planting costs. Attached Figure Description

[0014] Figure 1 This is a flowchart illustrating a seedling grading method provided in some embodiments of this application;

[0015] Figure 2 This is a schematic diagram of the cross-modal attention mechanism feature fusion process provided in some embodiments of this application;

[0016] Figure 3 These are internal structural diagrams of a computer device provided in some embodiments of this application. Detailed Implementation

[0017] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0018] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0019] The seedling grading method provided in this application will be described in detail below with reference to the accompanying drawings, through specific embodiments and application scenarios.

[0020] In one exemplary embodiment, such as Figure 1As shown, a seedling grading method is provided. This method can grade each seedling in a batch individually. In this embodiment, the seedling to be graded is used as the target seedling for explanation. The method includes steps 102-106. Wherein:

[0021] Step 102: Determine the time series corresponding to the target seedling; the time series is obtained by performing correlation processing on multimodal data; the multimodal data is obtained by collecting parameters from the target seedling and then analyzing those parameters.

[0022] When collecting parameters of the target seedlings, the acquisition equipment includes at least a 3D reconstruction camera and a multispectral imaging device. This multispectral imaging device includes, but is not limited to, industrial charge-coupled device (CCD) / complementary metal-oxide-semiconductor (CMOS) cameras or commercial multispectral cameras. The multispectral imaging device is equipped with filters for multiple wavelengths, the center wavelengths of which include, but are not limited to, 450nm, 550nm, 650nm, 700nm, 710nm, 800nm, 860nm, 900nm, and 1240nm, corresponding to red light, near-infrared light, and short-wave infrared light.

[0023] The parameters acquired by the 3D reconstruction camera are 3D point cloud data, from which 3D phenotypic feature data can be extracted. The parameters acquired by the multispectral imaging device are the reflected light of the target area (leaf area) of the target seedling at each wavelength, thereby calculating the average reflectance and then calculating the multispectral feature data. That is, multimodal data can include at least one of the 3D phenotypic feature data and at least one of the multispectral feature data.

[0024] It can be understood that the process of obtaining multimodal data through parameter analysis is a data preprocessing process. This preprocessing process includes 3D point cloud data filtering (e.g., using mean filtering or Gaussian filtering to remove noise), point cloud voxel downsampling, spatiotemporal registration of multispectral images and 3D point clouds, and extraction of multimodal data. For spatiotemporal registration, it can be achieved based on ORB (Oriented Fast and Rotated BRIEF) feature point matching combined with the PnP algorithm, or it can be achieved through the Iterative Closest Point (ICP) algorithm, thereby ensuring feature space alignment for the same seedling species.

[0025] The three-dimensional phenotypic data include, but are not limited to, plant height (H), canopy volume (V), projected leaf area (PA) or leaf area index (LAI), and stem diameter; the multispectral data include, but are not limited to, relative chlorophyll content (Soil and Plant Analyzer Development, SPAD), normalized difference water index (NDWI), and normalized difference vegetation index (NDVI).

[0026] The resolution during downsampling can be set to 1 cm³.

[0027] Furthermore, after extracting the three-dimensional phenotypic feature data and multispectral feature data, it is possible to further calculate the feature changes between adjacent time points (e.g., plant height increment) for each feature data. Leaf area increase ).

[0028] Here, a time series is a sequence obtained by arranging multimodal data in chronological order. For example, if the multimodal data is 14 days of historical multimodal data, the time series is a sequence of multimodal data arranged chronologically on a daily basis; or if the multimodal data is 7 days of historical multimodal data, the time series is a sequence of multimodal data arranged chronologically on a daily basis. Of course, the unit can also be hours, etc., and this embodiment does not make a specific limitation.

[0029] In one embodiment, determining the time series corresponding to the target seedling includes: standardizing the three-dimensional phenotypic feature data and the multispectral feature data to obtain standardized features; performing dimensionality reduction processing on the standardized features to obtain low-dimensional features; and constructing the time series corresponding to the target seedling based on the low-dimensional features and their temporal order.

[0030] Standardization can be achieved using Z-score standardization; dimensionality reduction can be achieved by using principal component analysis (PCA) or autoencoder neural networks to compress features, reducing the dimensionality of standardized features to, for example, 5-6 dimensions, thereby removing redundant information.

[0031] In addition, in recent years, some studies have attempted to introduce remote sensing indicators such as the normalized differential moisture index to assess the moisture status of seedlings. However, such indicators are essentially static spectral ratios, which do not take into account the physiological differences in the response of seedlings to stress at different growth stages, nor do they establish a coupling relationship with the morphological development process, resulting in low signal-to-noise ratio and weak discrimination in the seedling stage.

[0032] Based on this, in one embodiment, the multimodal data further includes a growth-photosynthesis decoupling index; the multispectral feature data includes at least a relative chlorophyll content and a normalized differential moisture index; the growth-photosynthesis decoupling index is calculated using a growth vigor index and a spectral response index; the growth vigor index is calculated using the relative chlorophyll content, and the spectral response index is calculated using the relative chlorophyll content and / or the normalized differential moisture index.

[0033] Among them, the growth-photosynthesis decoupling index is used to characterize the dynamic imbalance between morphological growth and photosynthetic response of seedlings under stress. This index breaks through the static characterization mode of traditional vegetation indices, and achieves high-sensitivity identification of early stress by monitoring the consistency of changes in the two physiological processes.

[0034] The role of this growth-photosynthesis decoupling index can be understood in the following context:

[0035] For example, if seedlings are in a state of water and nutrient deficiency, their leaves will stop growing, but photosynthesis has not yet decreased. At this time, the seedlings will show slow growth but no obvious abnormalities. This phenomenon is an early stage of seedling disease. It is understandable that if only a single indicator, such as the normalized differential moisture index, is used to assess the condition of seedlings, it will be impossible to detect the disconnect between growth and photosynthesis.

[0036] The growth-photosynthesis decoupling index proposed in this embodiment is specifically designed to measure the degree of this decoupling. This index reflects whether the pace of change between growth and photosynthesis is consistent, rather than focusing on the value of a particular indicator. That is, if the pace is found to be out of sync, it can provide an early warning of potential problems with the seedlings, and can detect seedling problems much earlier than through other means, such as yellowing or drying of leaves.

[0037] The Green Vegetation Index (GVI) can be calculated using a weighted combination method, and its calculation formula is shown in formula (1):

[0038] (1)

[0039] in, Seedling height (unit: cm) reflects longitudinal growth capacity; The stem volume (unit: cm³) reflects the amount of growth in three-dimensional space. Leaf area (unit: cm²), reflecting photosynthetic capacity; The leaf chlorophyll content index reflects photosynthetic efficiency; , and These are the standardized reference values ​​for the corresponding indicators (determined based on the standard values ​​for the variety or the average of historical data). , , and These are the weighting coefficients for each indicator, and the sum of these weighting coefficients is 1.

[0040] In one embodiment, the growth-photosynthesis decoupling index specifically characterizes the dynamic imbalance between morphological growth rate and photosynthetic activity response. In this embodiment, the growth-photosynthesis decoupling index (GPDI) is calculated using the following formula (2):

[0041] (2)

[0042] Among them, the The growth vigor index is the rate of change over time. The spectral response index (SI) is the rate of change over time, used to characterize the dynamics of photosynthetic physiological activity; the... This is a normalization coefficient used to eliminate dimensional differences, and can be set using historical data mean or prior knowledge of the variety; the aforementioned It is positively correlated with the degree to which the target seedling is under potential stress.

[0043] The specific composition of this spectral response index can be selected according to the application scenario and monitoring focus.

[0044] In one specific embodiment, the spectral response index can be directly expressed as the value of the relative chlorophyll content, i.e. This is used to reflect the rate of change in photosynthetic pigment content.

[0045] In another embodiment, the spectral response index can be the value of the normalized differential moisture index, i.e. This is used to reflect the rate of change in the plant's water status.

[0046] In another embodiment, to more comprehensively reflect the photosynthetic physiological state, the spectral response index can be constructed using a linear combination of SPAD and NDWI, for example... , in and These are weighting coefficients predetermined based on historical experience data or specific crop varieties.

[0047] In one embodiment, constructing the time series corresponding to the target seedling based on the low-dimensional features and their temporal order includes: performing feature fusion on the low-dimensional features based on a cross-modal attention mechanism to obtain fused features; and constructing the time series corresponding to the target seedling based on the fused features and their temporal order.

[0048] like Figure 2 As shown, this embodiment introduces a cross-modal attention module, which dynamically learns and fuses the complementarity and correlation between features from different modalities, adaptively weights key information and suppresses redundancy, thereby generating a more discriminative fused feature representation.

[0049] Specifically, the portions corresponding to the three-dimensional phenotypic features and the portions corresponding to the multispectral features in the low-dimensional features are encoded separately using fully connected layers or small convolutional networks. This cross-modal attention mechanism dynamically evaluates the importance of different modal features in different dimensions by calculating the interaction attention weights (Query-Key-Value mechanism) between the three-dimensional phenotypic features and the multispectral features.

[0050] Specifically, this cross-modal attention mechanism can learn the dependencies between different modalities, automatically strengthen features that are highly discriminative for the current hierarchical task (e.g., if a certain growth stage is sensitive to volume, then strengthen volume; if another stage is sensitive to chlorophyll, then strengthen the corresponding chlorophyll index), while suppressing redundant or noisy information, and finally outputting a high-dimensional, unified semantic fusion feature vector (fusion feature) rich in cross-modal complementary information.

[0051] Step 104: Based on the time series, call the growth prediction model to generate the key growth indicators target values ​​for the target seedling.

[0052] In one embodiment, the growth prediction model is a deep learning model that integrates a causal convolutional network and a long short-term memory network; the causal convolutional network is used to extract long-term temporal dependencies, and the long short-term memory network is used to model temporal dynamic characteristics.

[0053] That is, after inputting the time series data into a pre-trained hybrid model of Temporal Convolutional Networks (TCN) and Long Short-Term Memory (LSTM), the model can output the predicted target index values ​​(e.g., plant height) of the key growth indicators of the target seedlings for the next N days. leaf area (etc.). Where N is a positive integer, which can be set as needed.

[0054] In another embodiment, the growth trajectory of the target seedling can be fitted using a Gompertz growth curve model. That is, the residual prediction network is used to model the residual of the individual growth trajectory fitting of the target seedling, thereby predicting the correction values ​​of key growth indicators for the next N days.

[0055] Specifically, the Gompertz growth curve model is used to fit the potential growth trajectory, and a residual prediction module is constructed by combining a temporal convolutional network or a residual neural network (ResNet) to model and compensate for nonlinear disturbances that deviate from the standard model during actual growth, thereby achieving personalized and accurate prediction of key growth indicators of seedlings in the future.

[0056] For growth trajectory modeling, a time series (time points) containing historical multimodal fusion features is first maintained for the target seedling. Then, the Gompertz growth curve model was used to perform nonlinear fitting on the historical time series of key growth indicators. For example, key growth indicators include plant height and leaf area.

[0057] The following uses plant height as the key growth indicator. Explanation:

[0058] The formula for the Gompertz growth curve model can be expressed as formula (3):

[0059] (3)

[0060] in, This represents the asymptotic maximum of growth (growth potential). Indicates the maximum growth rate. It is the turning point.

[0061] The target seedling was obtained through fitting. and This can describe its potential growth trend.

[0062] After obtaining the target values ​​for key growth indicators (such as plant height and leaf area) for the target seedlings over the next N days, an evaluation system needs to be established to classify them. Therefore, we first define these predicted key growth indicators as classification indicators for the final evaluation.

[0063] For each grading indicator, its evaluation criteria need to be pre-defined based on agricultural production standards or expert experience. These criteria should include at least the following:

[0064] Weighting coefficients ( ): indicates the importance of the grading indicator in the comprehensive evaluation.

[0065] Lower threshold ( ): Represents the minimum acceptable value or passing grade for this grading indicator.

[0066] Upper limit threshold ( ): Represents the ideal value or full score for this grading indicator.

[0067] In the residual prediction module, the actual growth process of seedlings often deviates from the ideal trajectory of the Gompertz growth curve model due to environmental fluctuations, pests, and other factors. To capture this nonlinear disturbance, this embodiment introduces a residual prediction network.

[0068] This residual prediction network takes historical fused feature sequences and / or fitted residual sequences as input and learns to predict residual values ​​for future time points (e.g., the next 7 days). The final predicted value of personalized growth indicators. Predicted values ​​from the Gompertz growth curve model The result is obtained by adding the residual value, as shown in formula (4):

[0069] (4)

[0070] Before grading the seedlings, the relationships between the key terms in this application must be clarified:

[0071] Key growth indicators (CGIs) refer to specific biological parameter predictions directly output by growth prediction models, such as predicted plant height and leaf area for the next N days. During grading evaluation, these predicted CGIs are directly used as grading indicators, and their predicted values ​​are the values ​​used in the grading calculations. Therefore, in the following description, grading indicators can be understood as key growth indicators at the evaluation stage.

[0072] Step 106: Grade the target seedlings based on their key growth indicators.

[0073] Specifically, based on the grading indicators and their weights, the comprehensive grading score of the target seedling is calculated, and the final grading level is output according to the preset grading score interval rules.

[0074] The formula for calculating the comprehensive graded score S is as shown in formula (5):

[0075] (5)

[0076] in, For the first Actual or predicted values ​​of each graded indicator; ; This represents the total number of tiered indicators.

[0077] The grading level can be set as needed. For example, when S≥0.8, the grading level is "Excellent"; when 0.6≤S<0.8, the grading level is "Good"; and when S<0.6, the grading level is "Medium or Poor".

[0078] This application proposes a seedling grading method, which includes determining the time series corresponding to the target seedling; obtaining the time series through correlation processing of multimodal data; obtaining the multimodal data by collecting and analyzing parameters of the target seedling; based on the time series, calling a growth prediction model to generate target index values ​​for key growth indicators of the target seedling; and grading the target seedling based on the target index values. In other words, this method, by accurately grading seedlings, avoids misjudging seedlings with poor growth as those with good growth, thereby ensuring uniform and robust seedlings, improving seedling survival rate, and reducing agricultural planting costs.

[0079] In one embodiment, in order to reduce the impact of different regions, seasons and facility environments on the accuracy of seedling grading, this embodiment monitors environmental parameters in real time and dynamically corrects the lower limit threshold of grading indicators based on environmental parameters; according to the corrected lower limit threshold, weight and predicted or measured value, a weighted linear scoring function is used to calculate the comprehensive grading score S; and the grading level is determined according to the comprehensive grading score S and the preset grading score interval rules.

[0080] The environmental parameters include the current ambient temperature and light intensity.

[0081] Dynamic correction refers to the closed-loop correction and optimization of the classification threshold and indicator weights through a dynamic feedback mechanism based on Bayesian optimization, ensuring that the classification standards can be dynamically adjusted according to environmental changes. This dynamic feedback mechanism uses classification accuracy or consistency as its objective function.

[0082] The dynamic feedback mechanism uses the following formula (6) to correct the grading threshold:

[0083] (6)

[0084] in: For the revised first The lower threshold of each graded indicator; The current ambient temperature; This represents the current light intensity. and These are the preset ambient temperature reference value and light intensity reference value, respectively; and These are the temperature sensitivity coefficient and the light sensitivity coefficient, respectively.

[0085] Furthermore, when grading seedlings of new varieties, transfer learning can be used to obtain new growth prediction models adapted to the new varieties, enabling rapid model adaptation. Transfer learning involves freezing the underlying network parameters of a pre-trained growth prediction model and fine-tuning the output layer parameters using only a small number of labeled samples of the new varieties.

[0086] It should be noted that this growth prediction model is a multi-variety pre-trained model, meaning it can predict the growth of seedlings from multiple varieties. Employing self-supervised pre-training combined with domain alignment technology, it requires only 10-20 new variety samples to efficiently complete model fine-tuning and transfer adaptation.

[0087] Among them, the pre-training of the multi-variety pre-trained model can be carried out on a large-scale dataset containing multiple common varieties, using self-supervised learning (such as contrastive learning, mask autoencoder) combined with supervised hierarchical tasks to obtain a powerful basic model (growth prediction model) with general cross-modal fusion and growth pattern understanding capabilities.

[0088] For transfer learning, efficient transfer can be achieved through small-sample fine-tuning and domain alignment.

[0089] Specifically, for small sample fine-tuning, when a new variety is introduced, most of the low-level parameters in the base model (such as most layers of the feature encoder and cross-modal attention module) are frozen, and the layers near the output (such as the last few layers of the cross-modal fusion layer, the adaptation layer of the growth prediction module, and the classification output layer) are unfrozen and fine-tuned.

[0090] For domain alignment, during fine-tuning, a domain alignment loss function (such as Maximum Mean Discrepancy (MMD) or Domain-Adversarial (DA) loss) is introduced. This loss function aims to reduce the distribution difference in the feature space between the small sample data of the new variety and the pre-training data of the base model, thereby improving the generalization of feature representation.

[0091] Using the above strategy, model adaptation can be completed quickly and efficiently with only a small number of seedling samples, significantly reducing the data requirements and annotation costs for new varieties.

[0092] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0093] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 3 As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When executed by the processor, the computer program implements a seedling grading method.

[0094] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0095] This application also provides a computer-readable storage medium storing a program or instructions. When the program or instructions are executed by a processor, they implement the various processes of the above-described seedling grading method embodiments and achieve the same technical effects. To avoid repetition, they will not be described again here.

[0096] This application embodiment also provides a chip, which includes a processor and a communication interface. The communication interface is coupled to the processor. The processor is used to run programs or instructions to implement the various processes of the above-described seedling grading method embodiment and achieve the same technical effect. To avoid repetition, it will not be described again here.

[0097] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the various processes of the above-described seedling grading method embodiments and achieves the same technical effect. To avoid repetition, it will not be described again here.

[0098] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0099] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0100] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method of grading seedlings, characterized in that, The seedling grading method comprises: determining a time sequence corresponding to the target seedling; the time sequence is obtained by relevant processing of multi-modal data; the multi-modal data is obtained by analyzing parameters collected from the target seedling; the multi-modal data comprises three-dimensional phenotype characteristic data, multi-spectral characteristic data and growth-photosynthesis decoupling index; the three-dimensional phenotype characteristic data at least comprises plant height, canopy volume and leaf area; the multi-spectral characteristic data at least comprises relative chlorophyll content and normalized difference water index; the growth-photosynthesis decoupling index is calculated by growth vigor index and spectral response index; the growth-photosynthesis decoupling index GPDI is calculated by the following formula: Wherein, the target seedling growth vigor index GVI is calculated by the following formula: is the time rate of change of the growth vigor index GVI; the target seedling growth vigor index GVI is calculated by the following formula: is the time rate of change of the spectral response index; the target seedling growth vigor index GVI is calculated by the following formula: is a normalization coefficient for eliminating dimensional differences; the target seedling growth vigor index GVI is calculated by the following formula: is positively correlated with the degree of potential stress state of the target seedling. the growth vigor index is obtained by weighting of the relative chlorophyll content, plant height, canopy volume and leaf area; the spectral response index is calculated by the relative chlorophyll content and / or the normalized difference water index; standardizing the multi-modal data to obtain standardized features; performing dimension reduction processing on the standardized features to obtain low-dimensional features; based on the low-dimensional features and their time sequence, a time sequence corresponding to the target seedling is constructed, comprising: based on a cross-modal attention mechanism, the low-dimensional features are fused to obtain fusion features; based on the fusion features and their time sequence, a time sequence corresponding to the target seedling is constructed; based on the time sequence, a growth prediction model is called to generate target index values of key growth indicators of the target seedling; based on the target index values, the target seedling is graded.

2. The seedling grading method as claimed in claim 1, wherein, The growth prediction model is a deep learning model combining causal convolutional network and long short-term memory network; the causal convolutional network part is used to extract long-term sequence dependence, and the long short-term memory network is used to model time sequence dynamic characteristics.

3. The seedling grading method according to claim 2, characterized in that, The grading of the target seedling based on the target index values comprises: real-time monitoring of environmental parameters; dynamic correction of the lower limit threshold of the target index values based on the environmental parameters; based on the corrected lower limit threshold, a preset weight and the target index values, a comprehensive grading score is calculated by a weighted linear scoring function; based on the comprehensive grading score and a preset grading score interval rule, a grading level is determined.

4. The seedling grading method according to claim 3, wherein when grading seedlings of a new variety, the method further comprises: freezing the bottom network parameters of the growth prediction model; obtaining labeled samples of the new variety, and fine-tuning the output layer parameters of the growth prediction model through the labeled samples to obtain a new growth prediction model adapted to the new variety.

5. A computer device, comprising: a processor, a memory and a program or instructions stored on the memory and executable on the processor, which when executed by the processor implements the steps of the seedling grading method according to any one of claims 1-4.

6. A readable storage medium, characterized by, a readable storage medium storing a program or instructions, which when executed by a processor implements the steps of the seedling grading method according to any one of claims 1-4.

Images