Method and device for evaluating seedling strength of plug seedling population based on field growth performance

Abstract

The application provides a plug seedling group strong seedling evaluation method and device based on field growth performance, relates to the field of agricultural information technology, and obtains production data, full-spectrum data of a plug seedling group, and environment data and field growth performance data of seedlings taken from the plug seedling group in different planting environments, so as to train a field growth performance prediction model; plug seedling group strong seedling evaluation is performed based on the field growth performance prediction model, and an evaluation result is determined. The application takes a plug as a unit, considers the consistency and uniformity of plug seedlings, and can meet the evaluation demand of seedling quality in a large-scale seedling raising process; the field growth performance prediction model is trained based on production data, full-spectrum data, environment data and field growth performance data, so as to perform evaluation, can comprehensively evaluate and reflect the internal growth vigor, development potential and adaptability in a complex environment of seedlings, and can also predict the field growth performance of seedlings in different environments.

Description

Technical Field

[0001] This invention relates to the field of agricultural information technology, and in particular to a method and device for evaluating the robustness of plug seedling populations based on field growth performance. Background Technology

[0002] In modern factory-style seedling production, seedling assessment is a key step in ensuring the quality of commercial seedlings and the survival rate of transplanting. Through reasonable seedling control and hardening methods, the stress resistance of seedlings can be effectively improved, the seedling recovery period can be shortened, and the uniformity and consistency of seedlings leaving the nursery can be increased.

[0003] Currently, the assessment of robust seedlings mainly suffers from the following problems: it only focuses on the appearance and morphological indicators of seedlings, such as plant height, stem thickness, and number of leaves, which can only visually identify seedlings that appear robust, and cannot comprehensively assess and reflect the seedlings' intrinsic growth vitality, developmental potential, and adaptability in complex environments; it lacks follow-up feedback on the actual field growth performance of seedlings after transplanting and planting, resulting in a disconnect between seedling quality assessment and actual field production effects, and making it impossible to predict the field growth performance of seedlings in different environments; it only assesses individual seedlings, which cannot reflect important commercial attributes such as the consistency and uniformity of plug seedlings, and cannot meet the needs of seedling quality assessment in large-scale seedling production. Summary of the Invention

[0004] This invention provides a method and apparatus for evaluating the robustness of plug seedling populations based on field growth performance, in order to solve the shortcomings of existing technologies that can only evaluate the appearance of seedlings, lack tracking feedback, and fail to reflect the consistency of plug seedlings.

[0005] This invention provides a method for evaluating the robustness of plug seedling populations based on field growth performance, comprising: Acquire production data, full-spectrum data, and environmental data and field growth performance data of seedlings taken from the plug seedling population in different planting environments; Based on the production data, the production information feature vector of the plug seedling population is determined; based on the full-spectrum data, the spectral feature vector and the spectral inversion feature vector of the plug seedling population are determined; and based on the spectral feature vector and the spectral inversion feature vector, the phenotypic feature vector of the plug seedling population is determined. The production information feature vector, the spectral feature vector, the spectral inversion feature vector, the phenotypic feature vector, and the environmental feature vector corresponding to the environmental data are used as inputs, and the field growth performance vector corresponding to the field growth performance data is used as output to train a field growth performance prediction model. The robustness of the plug seedling population was assessed based on the field growth performance prediction model, and the assessment results were determined.

[0006] As one embodiment, the step of training a field growth performance prediction model by taking the production information feature vector, the spectral feature vector, the spectral inversion feature vector, the phenotypic feature vector, and the environmental feature vector corresponding to the environmental data as input, and taking the field growth performance vector corresponding to the field growth performance data as output, includes: Based on the physical conditions of the growth of the plug seedling population, a physical constraint function is constructed; A multi-task deep learning model is constructed. During the construction process, the physical constraint function is used as the post-processing module or the regularization term of the loss function of the multi-task deep learning model to obtain the initial prediction model. The initial prediction model is trained by taking the production information feature vector, the spectral feature vector, the spectral inversion feature vector, the phenotypic feature vector, and the environmental feature vector as inputs and the field growth performance vector as outputs, to obtain the field growth performance prediction model.

[0007] As one embodiment, the construction of the physical constraint function based on the growth physical conditions of the plug seedling population includes: Based on the matching relationship between seedling age and accumulated temperature of the plug seedling population, the constraint function for seedling age and accumulated temperature is determined. Based on the plant type index of the plug seedling population, the plant type constraint function is determined; Based on the temperature thresholds of the seedlings in different planting environments, an environmental temperature constraint function is determined. Based on the matching relationship between the seedling age and growth stage of the plug seedling population, the constraint function for growth days and developmental status is determined. The physical constraint function is determined based on the combination of the seedling age and accumulated temperature constraint function, the plant type constraint function, the ambient temperature constraint function, and the growth days and developmental state constraint function.

[0008] As one embodiment, the construction of the multi-task deep learning model includes: A feature fusion layer is constructed, which is used to fuse the production information feature vector, the spectral feature vector, the spectral inversion feature vector, the phenotypic feature vector, and the environmental feature vector to obtain fused features; A physical constraint embedding layer is constructed, which is used to use the physical constraint function as a post-processing module or a regularization term of the loss function of the multi-task deep learning model. Based on the number of field growth performance indicators corresponding to the field growth performance data, a multi-task output layer is constructed. Based on the feature fusion layer, the physical constraint embedding layer, and the multi-task output layer, a multi-task deep learning model is determined.

[0009] As an example, when the physical constraint function is used as the regularization term of the loss function of the multi-task deep learning model, the feature fusion layer further includes a channel attention module, and the multi-task deep learning model further includes an adaptive fusion layer, which is used to adjust the weights of the phenotypic feature sub-vector and the environmental feature sub-vector based on the attention mechanism.

[0010] As one embodiment, determining the spectral feature vector and spectral inversion feature vector of the plug seedling population based on the full-spectrum data includes: Based on the full-spectrum data, the spectral feature sub-vectors of the plug seedling population are determined; Based on the full-spectrum data and the pre-constructed spectral inversion model of plug seedlings, the spectral inversion feature vectors of the plug seedling population are determined; The seedling spectral inversion model is obtained by training a preset model based on the full-spectrum pre-training data of the seedling population and the destructive sampling feature data of the seedling population.

[0011] As one embodiment, acquiring production data, full-spectrum data, and environmental data and field growth performance data of seedlings taken from the plug seedling population in different planting environments includes: The production data, full-spectrum data, environmental data, and field growth performance data of the seedlings in different planting environments are acquired using automated data acquisition equipment.

[0012] This invention also provides a device for evaluating the robustness of plug seedling populations based on field growth performance, comprising: The acquisition module is used to acquire production data, full-spectrum data, and environmental data and field growth performance data of seedlings taken from the plug seedling population in different planting environments. The determination module is used to determine the production information feature vector of the plug seedling population based on the production data, determine the spectral feature vector and spectral inversion feature vector of the plug seedling population based on the full-spectrum data, and determine the phenotypic feature vector of the plug seedling population based on the spectral feature vector and the spectral inversion feature vector. The training module is used to take the production information feature vector, the spectral feature vector, the spectral inversion feature vector, the phenotypic feature vector, and the environmental feature vector corresponding to the environmental data as inputs, and take the field growth performance vector corresponding to the field growth performance data as outputs to train a field growth performance prediction model. The evaluation module is used to evaluate the robustness of the plug seedling population based on the field growth performance prediction model and determine the evaluation results.

[0013] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described method for evaluating the robustness of plug seedling populations based on field growth performance.

[0014] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the seedling vigor assessment method for plug seedling populations based on field growth performance as described above.

[0015] This invention provides a method and apparatus for evaluating the robustness of plug seedling populations based on field growth performance. The method acquires production data, full-spectrum data, and environmental data and field growth performance data of seedlings taken from the plug seedling population in different planting environments. Based on the production data, a production information feature vector is determined for the plug seedling population. Based on the full-spectrum data, spectral feature vectors and spectral inversion feature vectors are determined for the plug seedling population. Based on the spectral feature vectors and spectral inversion feature vectors, phenotypic feature vectors are determined for the plug seedling population. The production information feature vectors, spectral feature vectors, spectral inversion feature vectors, phenotypic feature vectors, and environmental feature vectors corresponding to the environmental data are used as inputs, and the field growth performance vectors corresponding to the field growth performance data are used as outputs to train a field growth performance prediction model. The robustness of the plug seedling population is evaluated based on the field growth performance prediction model, and the evaluation result is determined. This invention uses plug trays as units, taking into account the consistency and uniformity of plug tray seedlings, which can meet the needs of seedling quality assessment in large-scale seedling production. Based on the production data, full-spectrum data, environmental data and field growth performance data of plug tray seedling populations, a field growth performance prediction model is trained for evaluation. It can comprehensively assess and reflect the seedling's intrinsic growth vitality, development potential and adaptability in complex environments, and can also predict the field growth performance of seedlings under different environments. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 This is one of the flowcharts of the method for evaluating the robustness of plug seedling populations based on field growth performance provided by the present invention.

[0018] Figure 2 This is the second flowchart of the method for evaluating the robustness of plug seedling populations based on field growth performance provided by this invention.

[0019] Figure 3 This is a schematic diagram of the structure of the seedling vigor assessment device for plug seedling populations based on field growth performance provided by the present invention.

[0020] Figure 4 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0022] Figure 1 This is one of the flowcharts illustrating the method for evaluating the robustness of plug seedling populations based on field growth performance provided by this invention. Figure 2 This is the second flowchart of the method for evaluating the robustness of plug seedling populations based on field growth performance provided by this invention. Figure 1 and Figure 2 As shown, the method includes steps S110-S140.

[0023] Step S110: Obtain production data, full-spectrum data, and environmental data and field growth performance data of seedlings taken from the plug seedling population in different planting environments.

[0024] Optionally, the production data of the plug seedling population includes, but is not limited to, standardized seedling age, accumulated temperature, accumulated temperature percentage, cumulative light, cumulative light index, variety, and growth stage recorded by the nursery.

[0025] Optional, full-spectrum data includes various spectral data from multiple perspectives of plug seedlings.

[0026] Optional, environmental data for different planting environments include environmental data such as temperature, light, and water in the planting environment.

[0027] Optionally, through a planting control experiment, field performance data of seedlings with different phenotypes in different environments can be collected to obtain multi-dimensional indicators such as seedling survival rate, seedling establishment time, growth rate, seedling vigor, developmental potential, adaptability, and plant type.

[0028] Step S120: Based on the production data, determine the production information feature vector of the plug seedling population; based on the full-spectrum data, determine the spectral feature vector and spectral inversion feature vector of the plug seedling population; and based on the spectral feature vector and the spectral inversion feature vector, determine the phenotypic feature vector of the plug seedling population.

[0029] Step S130: The production information feature vector, the spectral feature vector, the spectral inversion feature vector, the phenotypic feature vector, and the environmental feature vector corresponding to the environmental data are used as inputs, and the field growth performance vector corresponding to the field growth performance data is used as output to train a field growth performance prediction model.

[0030] Optionally, the production information feature vector, the spectral feature vector, the spectral inversion feature vector, the phenotypic feature vector, and the environmental feature vector corresponding to the environmental data are merged to obtain the phenotypic feature vector of the plug seedling population. Samples are constructed based on the phenotypic feature vector of the plug seedling population and the field growth performance vector, and a field growth performance prediction model is trained using deep learning or machine learning methods.

[0031] Step S140: Based on the field growth performance prediction model, evaluate the robustness of the plug seedling population and determine the evaluation results.

[0032] Optionally, production data, full-spectrum data, and environmental data of the seedling population to be evaluated are obtained to determine the fusion vector corresponding to the seedling population to be evaluated. The fusion vector corresponding to the seedling population to be evaluated is input into the field growth performance prediction model to obtain the predicted field growth performance of the seedling population to be evaluated output by the field growth performance prediction model. The predicted field growth performance of the seedling population to be evaluated is weighted and summed to obtain the seedling robustness evaluation result of the seedling population to be evaluated.

[0033] It is understood that this invention uses seed trays as units, taking into account the consistency and uniformity of seedlings in trays, which can meet the needs of seedling quality assessment in large-scale seedling production. A field growth performance prediction model is trained based on production data, full-spectrum data, environmental data, and field growth performance data of the seedling population for evaluation. This model can comprehensively assess and reflect the seedling's intrinsic growth vitality, developmental potential, and adaptability in complex environments, and can also predict the field growth performance of seedlings under different environments. It breaks through the limitations of traditional evaluation methods that rely solely on morphological characteristics, and also realizes the transformation from phenotypic evaluation to performance prediction, making the seedling production process more scientific and forward-looking. This invention also provides strong support for achieving precise and customized control of seedlings, helping to improve the quality of commercial seedlings, transplant survival rate, and field adaptability, and promoting the development of agricultural seedling production towards intelligence and data-driven approaches.

[0034] As one embodiment, acquiring production data, full-spectrum data, and environmental data and field growth performance data of seedlings taken from the plug seedling population in different planting environments includes: The production data, full-spectrum data, environmental data, and field growth performance data of the seedlings in different planting environments are acquired using automated data acquisition equipment.

[0035] Optionally, production data can be obtained from IoT monitoring data of intensive seedling farms or from manual record ledgers. Correspondingly, based on the production data, a production information feature sub-vector for the plug seedling population is determined. This includes determining standardized seedling age, accumulated temperature, accumulated temperature percentage, cumulative light intensity, cumulative light intensity index, variety, growth stage, and sowing density based on the standardized seedling age, accumulated temperature, accumulated temperature percentage, cumulative light intensity, cumulative light intensity index, variety, growth stage, and sowing density data included in the generated data. All production information features are then combined and converted into a vector to obtain the production information feature sub-vector.

[0036] The standardized seedling age is expressed as follows: ; in, To standardize seedling age, This represents the actual number of days the seedlings in the plug trays have grown. The standard seedling age for plug seedling populations.

[0037] Accumulated temperature is used to characterize the cumulative daily average temperature during the seedling period, and its expression is as follows: ; in, For accumulated temperature, For the first The average daily temperature (°C).

[0038] The expression for the accumulated temperature ratio is as follows: ; in, The percentage of accumulated temperature. The effective accumulated temperature is calculated using the following formula: ,unit%, Let be the average accumulated temperature on day i. Let the base accumulated temperature be the temperature on day i. The target effective accumulated temperature for plug seedling populations.

[0039] Cumulative light intensity is used to characterize the cumulative value of daily average photosynthetically active radiation during the seedling period, and its expression is as follows: ; in, To accumulate light, For the first The average daily photosynthetically active radiation per day, in μmol / (m²). 2 ·s).

[0040] The expression for the cumulative light intensity index is as follows: ; in, The cumulative light index, The standard is the daily average cumulative photosynthetically active radiation, and t is the cumulative duration of sunlight. It is an instantaneous, effective radiation.

[0041] The varieties of seedlings in plug trays are represented as follows: One-hot coding is used to digitally encode categorical features, for example, tomato seedling = 1, cucumber seedling = 2.

[0042] The reproductive stages of a seedling population in a plug tray are represented as follows: , which are digital unique hot codes, for example, cotyledon stage [1,0,0], true leaf stage [0,1,0], seedling stage [0,0,1].

[0043] Let the sowing density of the plug seedling population be expressed as The categorical characteristics of the acupuncture tray are digitally encoded, for example, according to the specifications of the acupuncture tray, 72 holes are coded as 72, and 36 holes are coded as 36.

[0044] In summary, the production information feature sub-vector of the plug seedling population can be represented as: .

[0045] Optionally, various spectral acquisition devices, such as hyperspectral, multispectral, thermal imaging, chlorophyll fluorescence, depth cameras, and lidar, can be used to collect spectral data from multiple perspectives of the plug seedlings to obtain full-spectrum data. The acquisition area includes the entire canopy of the plug seedlings and a randomly selected number of seedlings, including but not limited to time-series spectral data of visible light, near-infrared, depth, and thermal imaging collected using multiple spectral sensors. The full-spectrum data is in .img / .dat format.

[0046] Optionally, determining the spectral feature vector and spectral inversion feature vector of the plug seedling population based on the full-spectrum data includes: Based on the full-spectrum data, the spectral feature sub-vectors of the plug seedling population are determined; Based on the full-spectrum data and the pre-constructed spectral inversion model of plug seedlings, the spectral inversion feature vectors of the plug seedling population are determined; The seedling spectral inversion model is obtained by training a preset model based on the full-spectrum pre-training data of the seedling population and the destructive sampling feature data of the seedling population.

[0047] Optionally, based on the full-spectrum data S seed Calculate and construct the spectral feature vector X2, which includes, but is not limited to, the mean of the three visible light channels, near-infrared reflectance, visible light contrast, total leaf area, average canopy depth, average canopy temperature, canopy coverage, leaf color depth, simple ratio index, and multidimensional vegetation index.

[0048] Average value of three visible light channels These are the pixel grayscale averages for the red, green, and blue channels of the visible light spectrum, respectively.

[0049] Near-infrared reflectance The expression used to characterize the average reflectance of the near-infrared spectrum (760-2500 nm) is: ,in wavelength The intensity of reflected light at that location.

[0050] Average depth of canopy The distance (cm) from the canopy of the calcaneus to the sensor is the average distance from the depth sensor.

[0051] Canopy mean temperature Average leaf temperature These are the average temperatures (°C) of the canopy and individual leaves in the thermal imaging spectrum.

[0052] Visible light contrast The texture features used to characterize the visible light spectrum are expressed as follows: ,in The pixel grayscale value represents the visible light spectrum. The maximum grayscale value of a pixel in the visible light spectrum. This represents the minimum grayscale value of a pixel in the visible light spectrum. This represents the average grayscale value of the visible light spectrum pixels.

[0053] Total leaf area The expression used to characterize the total leaf area of ​​all seedlings in a single-cell tray is: ,in The average leaf area per plant (cm²) obtained from spectral inversion 2 ), This refers to the number of holes in the acupuncture plate (e.g., 50 holes, 72 holes).

[0054] Canopy coverage The expression is: It is calculated by segmenting the visible light spectral image.

[0055] Leaf color depth The expression is: ,in, This is the green channel value for visible light. The value represents the visible light red channel value. The leaf color depth is essentially the green light / red light ratio; the larger the value, the deeper the color.

[0056] Simple ratio index The expression is: .

[0057] The multidimensional vegetation index (NI) is used to characterize multiple spectral indices related to the spectrum. The index names, formulas, and ecological significance are shown in the table below.

[0058] Table 1. Names, formulas, and ecological significance of multidimensional vegetation indices.

[0059] in, For near-infrared reflectivity, For red light band reflectivity, For green light band reflectivity, For shortwave infrared reflectivity, The temperature of the plant canopy. For air temperature, The reflectivity is 531nm. The reflectivity is 570nm.

[0060] Optionally, the spectral inversion model for plug seedlings can be trained before step S110.

[0061] Optionally, various spectral acquisition devices such as hyperspectral, multispectral, thermal imaging, chlorophyll fluorescence, depth camera, and lidar are used to collect and aggregate various spectral data from multiple perspectives of plug seedlings to obtain full-spectrum pre-training data S. a11 .

[0062] Optionally, destructive sampling data of plug seedlings corresponding one-to-one with spectral data can be collected, including but not limited to data on chlorophyll content, nutrient status, water content, biomass, plant height, and stem diameter collected manually through destructive sampling. Specifically, manual measurement / instrument detection data (structured data) of 10 randomly destructively sampled seedlings in a single plug tray can be collected, with sampling following the principle of uniform distribution (2 seedlings at each of the four corners of the plug tray plus 2 seedlings at the center).

[0063] Based on destructive sampling feature data, a destructive sampling feature vector for plug seedlings is constructed. Each index in the vector is the average value of 10 seedlings. This indicates the content of chlorophyll a. The content of chlorophyll b is indicated by the chlorophyll a and b content (mg / g) detectable by acetone extraction. N, P, and K represent the nitrogen, phosphorus, and potassium contents (%) detected by Kjeldahl method, molybdenum antimony colorimetric method, and flame photometry, respectively. This indicates the fresh weight of the entire seedling. This indicates the dry weight (g) of the entire seedling after blanching at 105℃ and drying at 65℃ to constant weight. H represents the distance from the base of the stem to the highest leaf tip of the seedling, in mm. This indicates the stem diameter at 0.5 cm below the cotyledon node of the seedling, in mm.

[0064] The destructive sampling feature vector of plug seedlings may also include water content. and biomass , biomass With dry weight as the core indicator, .

[0065] Optionally, a seedling phenotypic inversion model based on full-spectrum pre-trained data can be constructed using machine learning or deep learning. p The main phenotypic parameters of the seedlings can be obtained from their spectral data, i.e., p=f p (S a11 During training, the variety and age of the seedlings will be input into the model as pre-parameters to ensure the accuracy of phenotypic inversion.

[0066] Optionally, call the seedling phenotypic inversion model f. p Using full-spectrum data of plug seedlings S seed Calculate and construct the seedling spectral inversion feature vector X3, X3=fp(S seed).

[0067] Optionally, phenotypic feature sub-vectors include seedling vigor index (traditional), plant type characteristics, activity indicators, morphological indicators, 3D phenotype (plant type), uniformity, consistency, and uniformity, which can be represented as follows: .

[0068] Traditional seedling index (Population mean) is calculated using the commonly used agricultural seedling strength index formula. .

[0069] Plant type characteristics The data is digitized, for example, compact = 1, loose = 2, medium = 3, and determined by clustering of 3D point cloud data.

[0070] Activity indicators The expression is: , which is the percentage of chlorophyll dry weight multiplied by cumulative light exposure, characterizes photosynthetic activity.

[0071] morphological indicators The expression is: The ratio of plant height to stem diameter indicates the robustness of the seedling; the smaller the value, the stronger the seedling.

[0072] Phenotype (plant type) Mean 3D canopy volume (cm²) used to characterize plug seedling populations 3 ), calculated from point cloud data from depth sensors, , For the first 3D canopy volume of hole seedlings.

[0073] Unif is used to characterize the coefficient of variation of plant height among all seedlings in a seed tray. SD(H) is the standard deviation of plant height avg(H), and avg(H) is the mean plant height.

[0074] Consistency (Cons) is used to characterize the coefficient of variation of stem diameter of all seedlings in the plug tray. Cons = CV(St) = .

[0075] Uniformity Reg is used to characterize the coefficient of variation of the dry weight of all seedlings in the plug tray, Reg=CV(W dw )= .

[0076] Optional, for After standardization, the phenotypic feature vectors of the plug seedling population are obtained by splicing the vectors together. ; in, After standardization That is, the standardized production information feature sub-vector. X2 is the standardized spectral feature vector. X3 is the standardized spectral inversion eigenvector. After standardization That is, the standardized phenotypic feature sub-vectors. , where is the total feature dimension of the phenotypic feature vector of the plug seedling population. for Feature dimensions, for Feature dimensions, for Feature dimensions, for The feature dimensions.

[0077] Optionally, this invention conducts planting experiments on seedlings with different phenotypes in different environments, tracks the performance of seedlings after planting, and collects environmental data and field growth performance data of seedlings after planting.

[0078] The experimental conditions for seedling transplanting are as follows.

[0079] 1. Experimental group: at least 30 plug seedling populations with different phenotypes, covering different varieties, seedling ages and phenotypic characteristics.

[0080] 2. Environmental gradient: Set 5-8 different planting environment gradients, such as temperature, light, and water (single-factor / multi-factor gradients).

[0081] 3. Number of replicates: Each experimental group was set up with 3 replicates under each environmental gradient, with 1 acupuncture plate per replicate.

[0082] 4. Tracking period: 15 days after transplanting, covering the seedling establishment period and the rapid growth period.

[0083] Obtain environmental data such as temperature, light, and moisture in the seedling planting environment. Collect multi-dimensional environmental data on the seedling planting environment. This represents the overall dimension of environmental characteristics. Specifically, environmental prediction data for the next 7 days of planting can be collected through field IoT monitoring equipment (temperature and humidity sensors, light sensors, and soil moisture sensors). The environmental characteristic sub-vector can be represented as... .

[0084] Accumulated temperature for planting Characterized by the cumulative predicted daily average temperature over 7 days after transplanting. (°C·d).

[0085] diurnal temperature range The daily average maximum temperature - daily average minimum temperature characterizes the establishment environment. (°C).

[0086] Highest temperature during planting , lowest temperature These are the predicted average maximum and minimum temperatures (°C) 7 days after planting.

[0087] Cumulative sunlight after planting Characterized by the predicted cumulative daily average photosynthetically active radiation (μmol / (m)) 7 days after transplanting 2 ·s)·d).

[0088] Light intensity Mean daily photosynthetically active radiation (μmol / (m)) characterizing the planting environment 2 ·s)).

[0089] Light duration The average daily effective light duration (h) characterizes the planting environment.

[0090] Irrigation water volume The planned irrigation water volume (L / m³) is represented by the amount of water to be used 7 days after transplanting. 2 ).

[0091] Soil moisture change The value representing the change in soil volumetric moisture content 7 days after transplanting. (%) For predicted values, This represents the soil moisture content before planting.

[0092] Optionally, the environmental feature subvector may also include the lighting environment features shown in the table below.

[0093] Table 2. Alternative characteristics of lighting environment.

[0094] Where PAR(i) is the average daily photosynthetically active radiation on day i (μmol / (m 2 ·s)); D is the total number of sunshine days, This refers to the actual length of daylight. The optimal length of daylight; This represents the average daily photosynthetically active radiation. This represents the standard deviation of the daily average photosynthetically active radiation.

[0095] Optionally, the environmental feature subvectors may also include the water environmental features shown in the table below.

[0096] Table 3. Alternative characteristics of water environment features.

[0097] in, Actual soil volumetric water content Wilting coefficient / permanent wilting point Field holding capacity; This represents the water stress threshold. This represents the actual water content at the i-th time point; N represents the time step, indicating the time interval between two consecutive soil moisture content observations; t The number of observations is used to calculate the total number of observation data points included in WSI.

[0098] right Perform normalization (Min-Max normalization: After that, the standardized field environment feature vector is obtained: ; in, This represents the total dimension of environmental characteristics.

[0099] Optionally, through a transplanting control experiment, field performance data of different phenotype plug seedlings under different environments are collected, including multi-dimensional indicators such as seedling survival rate, seedling establishment time, growth rate, seedling vigor, developmental potential, adaptability, and plant type. Field growth performance data are obtained through manual measurement / IoT monitoring 15 days after transplanting, and used as a label vector, which can be represented as follows: .

[0100] Survival rate : (Points) is the core indicator.

[0101] Seedling recovery period To quantify the seedling recovery period (d) in reverse, ( ,like but ), which is the core indicator.

[0102] growth rate The average daily increase in plant height (cm / d) is given for 1-15 days after transplanting. (5 is the normalization coefficient), which is the core indicator.

[0103] Seedling Viability The values ​​represent the average chlorophyll content and photosynthetic rate 15 days after transplanting. (point).

[0104] Developmental potential The ratio of flower bud differentiation to branch number 15 days after transplanting is the average. (point).

[0105] Environmental adaptability index For stability parameters based on multi-environment performance: .

[0106] Among them, CV genotype This represents the coefficient of variation for genotypes.

[0107] The quantitative standard for each of the above indicators is a 0-10 point scale, with higher scores indicating better performance.

[0108] Understandably, this invention objectively evaluates and reflects the intrinsic growth vitality, developmental potential, and adaptability to complex environments of seedlings by tracking and monitoring their field growth performance after transplanting. This addresses the shortcomings of existing technologies that primarily rely on seedling morphological characteristics, often only reflecting the seedlings' current appearance. By acquiring field performance data after transplanting, it directly reflects the quality of seedlings in actual production, resolving the disconnect between seedling quality assessment and actual field production results, and predicting seedling performance under different environments. The systematic tracking and analysis of actual field performance after transplanting makes it difficult to establish a causal relationship between seedling performance and the seedling process, thus enhancing the depth and breadth of seedling quality assessment. The assessment of robust seedling populations in plug seedlings primarily utilizes automated data collection methods, enabling efficient, low-error, and real-time data acquisition and analysis, ensuring the accuracy and scientific validity of the assessment results.

[0109] As one embodiment, the step of training a field growth performance prediction model by taking the production information feature vector, the spectral feature vector, the spectral inversion feature vector, the phenotypic feature vector, and the environmental feature vector corresponding to the environmental data as input, and taking the field growth performance vector corresponding to the field growth performance data as output, includes: Based on the physical conditions of the growth of the plug seedling population, a physical constraint function is constructed; A multi-task deep learning model is constructed. During the construction process, the physical constraint function is used as the post-processing module or the regularization term of the loss function of the multi-task deep learning model to obtain the initial prediction model. The initial prediction model is trained by taking the production information feature vector, the spectral feature vector, the spectral inversion feature vector, the phenotypic feature vector, and the environmental feature vector as inputs and the field growth performance vector as outputs, to obtain the field growth performance prediction model.

[0110] Optionally, based on the physical conditions of seedling growth, a physical constraint function can be constructed, providing environmental thresholds, seedling age thresholds, etc. Corresponding extreme values, single-factor constraint functions, and composite-factor constraints can be provided to limit the number of growth days and developmental status, avoiding the production of stunted or overly vigorous seedlings.

[0111] Optionally, the system uses seedling phenotypic data and field planting environment data as input, and the field growth performance of seedlings as output. A machine learning or deep learning model is constructed and trained. Machine learning models include, but are not limited to, multi-task XGBoost / LightGBM / CatBoost algorithms, multi-task random forests, and multi-task elastic networks / ridge regression. Deep learning models include, but are not limited to, multi-task MLPs, multi-task one-dimensional CNNs, and multi-task attention fusion networks. Multilayer perceptrons (MLPs) are preferentially chosen as the base model, and a lightweight architecture is adapted to the low-computing-power scenarios of agricultural production. The performance of seedling populations after planting under different environmental conditions is predicted through seedling phenotypic data, including but not limited to survival rate, seedling establishment time, growth rate, and other multi-dimensional indicators.

[0112] Understandably, this invention integrates production data from plug seedling populations with multi-source phenotypic data such as appearance, spectrum, biomass, and physicochemical values, combined with post-planting field growth performance data, to construct a seedling vigor assessment system based on field growth performance prediction. Machine learning or deep learning algorithms are used to model and analyze the aforementioned multi-dimensional, multi-time-series data, thereby enabling the prediction of seedling field growth performance under different environmental conditions. This not only overcomes the shortcomings of traditional assessment methods in terms of single data sources and insufficient assessment dimensions, but also improves the scientific rigor, systematicity, and accuracy of the assessment.

[0113] As one embodiment, the construction of the physical constraint function based on the growth physical conditions of the plug seedling population includes: Based on the matching relationship between seedling age and accumulated temperature of the plug seedling population, the constraint function for seedling age and accumulated temperature is determined. Based on the plant type index of the plug seedling population, the plant type constraint function is determined; Based on the temperature thresholds of the seedlings in different planting environments, an environmental temperature constraint function is determined. Based on the matching relationship between the seedling age and growth stage of the plug seedling population, the constraint function for growth days and developmental status is determined. The physical constraint function is determined based on the combination of the seedling age and accumulated temperature constraint function, the plant type constraint function, the ambient temperature constraint function, and the growth days and developmental state constraint function.

[0114] Optionally, based on the biological laws of seedling growth and the realities of agricultural production, a set of physical constraint functions can be constructed. This involves embedding constraints into the model's prediction process to correct the model output and avoid predictions that contradict biological common sense (such as incorrect assessments of stunted or etiolated seedlings). The inputs to all constraint functions are the model's original predicted values. The output is the corrected predicted value. Constraint functions are divided into single-factor constraint functions and composite-factor constraint functions.

[0115] Optionally, a seedling age and accumulated temperature constraint function is used to limit the matching between seedling age and accumulated temperature, avoiding misjudgment of seedling age due to insufficient / excessive accumulated temperature. The expression is as follows: ; ; in, For accumulated temperature constraint, (℃·d / day) (℃·d / day) can be adjusted according to the variety, such as for warm-season crops. If the accumulated temperature exceeds the threshold, correct the accumulated temperature value and recalculate the seedling activity index. This leads to the correction of the model's predicted survival rate. . To constrain seedling age, As a penalty coefficient, to ensure that the seedling age is within a reasonable range, Standardized seedling age is the ratio of the actual number of days of growth to the standard seedling age for the variety.

[0116] Optionally, the plant type constraint function is used to limit the range of the plant type index MI = H / St, avoiding incorrect assessments of excessively vegetative seedlings (MI too high) or stunted seedlings (MI too low). The expression for the plant type constraint function is as follows: ; in, , The general values ​​for vegetable seedlings and the values ​​for flower seedlings can be adjusted to... ;like If the threshold is exceeded, the original predicted score of the model will be multiplied by a penalty coefficient (0.3 / 0.5) to reduce the seedling strength level.

[0117] Optionally, an ambient temperature constraint function is used to limit the temperature threshold of the colonization environment to avoid invalid predictions under extreme temperatures, and its expression is as follows: ; in, , General values ​​for crops grown in normal temperatures, and for cold-resistant crops. If the planting environment is an extreme temperature, the model prediction value is set to 0, indicating that the seedlings cannot grow normally in this environment.

[0118] Optionally, the growth days and developmental status constraint function is used to limit the matching between seedling age and growth stage, avoiding seedlings that are too old but in a low growth stage (small and old seedlings) or seedlings that are too young but in a high growth stage (leggy seedlings). The expression is as follows: ; in, Standard seedling age for each growth stage , These represent the start and end seedling ages (in days) of the current growth stage; cotyledon stage. True leaf stage Seedling hardening period If the seedling age does not match the growth stage, the overall score for robust seedlings is adjusted by multiplying by the corresponding penalty coefficient.

[0119] The physical constraint function set is expressed as follows: ; in, Indicates to To synthesize.

[0120] Understandably, this invention encodes plant physiological priors into differentiable constraint functions, thereby improving the interpretability and extrapolation capability of the model.

[0121] As one embodiment, the construction of the multi-task deep learning model includes: A feature fusion layer is constructed, which is used to fuse the production information feature vector, the spectral feature vector, the spectral inversion feature vector, the phenotypic feature vector, and the environmental feature vector to obtain fused features; A physical constraint embedding layer is constructed, which is used to use the physical constraint function as a post-processing module or a regularization term of the loss function of the multi-task deep learning model. Based on the number of field growth performance indicators corresponding to the field growth performance data, a multi-task output layer is constructed. Based on the feature fusion layer, the physical constraint embedding layer, and the multi-task output layer, a multi-task deep learning model is determined.

[0122] Optionally, when the physical constraint function is used as the regularization term of the loss function of the multi-task deep learning model, the feature fusion layer further includes a channel attention module, and the multi-task deep learning model further includes an adaptive fusion layer, which is used to adjust the weights of the phenotypic feature sub-vector and the environmental feature sub-vector based on the attention mechanism.

[0123] Optionally, the phenotypic feature vectors of the plug seedling population can be used as the input feature set, and the field growth performance vectors can be used as the output label set for model training, corresponding one-to-one with the input feature set to form the model training dataset. ,in The total number of samples, This ensures the effectiveness of model training.

[0124] The feature fusion layer is used to achieve cross-dimensional fusion of phenotypic features of plug seedlings and field environmental features, and Directly spliced ​​into fusion features This addresses the information coupling problem of heterogeneous features.

[0125] Preferably, a channel attention module (SE module) is added to the feature fusion layer to calculate the attention weights of each feature. , fusion features are ,in , These are the attention weight parameters.

[0126] Optionally, the input features of the input layer of the multi-task deep learning model are: The input features are encoded using a dual-branch structure.

[0127] The dual-branch structure includes a phenotypic feature encoder and an environmental feature encoder.

[0128] Phenotypic feature encoders are used to process high-dimensional phenotypic data, and their expressions are as follows: ; Where the superscript (l) is the network layer index, This represents the hidden representation of the phenotypic layer l. This is the phenotypic weight matrix. This is the phenotypic bias vector. This represents the hidden phenotypic representation of layer l-1. This represents the batch normalization function. This represents the activation function.

[0129] The expression for the environmental feature encoder is as follows: in, This represents the hidden representation of the environment at layer l. This is the environmental weight matrix. This is the environment bias vector. This represents the hidden representation of the environment at layer l-1.

[0130] Optionally, the feature fusion layer uses an attention mechanism to dynamically adjust the phenotypic-environment feature weights, and the feature encoder uses an attention mechanism. The expression is as follows: ; ; in, For the sigmoid function, For element-wise multiplication, Here is the attention weight matrix. For attention bias vectors, This represents the hidden representation of the third-level phenotype. This is a hidden representation of the third layer of the environment. For element-wise addition, The weights are phenotypic-environmental features.

[0131] Machine learning models use constraint functions as post-processing modules to correct the model's original predictions and output the final corrected predictions; deep learning models use constraint functions as regularization terms in the loss function, embedding them into the model's training process. ,in, Let p represent the p-th constraint function. is the regularization coefficient, enabling joint training based on data-driven and physical constraints.

[0132] Optionally, the loss function is a multi-task weighted loss.

[0133] During training, a Dropout layer (dropout=0.2) and L2 regularization (weight_decay=0.0001) can be added to prevent overfitting.

[0134] Optionally, the multi-task deep learning model outputs a field performance prediction vector. Then, a vector of evaluation results for robust seedling populations in plug trays is generated by weighted summation. .

[0135] Strong seedlings comprehensive score Characterization pairs Core indicator (survival rate) Seedling establishment period growth rate Weighted summation, 0-10 points, the higher the score, the stronger the body.

[0136] Strong seedling grade according to Grading (General Standards for Agricultural Production) Grade 1 robust seedlings Grade II strong seedlings For the third grade of strong seedlings, It is a weak seedling.

[0137] Environmental adaptability The comprehensive performance prediction values ​​(0-10 points) of the seedlings output by the characterization model under the 1st to kth planting environments represent the degree of adaptation of the seedlings to the specific environment.

[0138] Among them, robust seedling grade As shown in the table below.

[0139] Table 4.

[0140] Optionally, the table below is a core technical parameter threshold table, which is suitable for the seedling cultivation scenarios of most vegetable and flower plug seedlings and can be finely adjusted according to the variety.

[0141] Table 5.

[0142] In summary, this invention comprehensively collects production information and seedling spectral data during the seedling cultivation process. Using high-precision phenotypic acquisition equipment, such as automated image acquisition systems, laser scanners, and multispectral cameras, morphological phenotypic data of the seedlings are obtained through spectral inversion, including indicators such as plant height, stem diameter, leaf area, leaf color, number of leaves, and root structure. This not only includes traditionally perceived robust appearance features but also utilizes high-throughput phenotyping technology to obtain more potential phenotypic parameters, providing rich feature inputs for subsequent modeling.

[0143] This invention introduces the concept of dynamic tracking, which involves continuously collecting data on the growth performance of seedlings under different field conditions after transplanting and planting. Key indicators such as growth rate, stress resistance (e.g., drought and disease resistance), survival rate, and yield potential under different environmental conditions are collected. This full life-cycle data collection, from the seedling stage to the field stage, not only fills the gap in traditional seedling assessment lacking field performance feedback but also provides data support for building predictive models.

[0144] In terms of data fusion and modeling, this invention employs multi-source heterogeneous data fusion technology to standardize and extract features from production data, phenotypic data, and environmental data. Preprocessing methods such as data cleaning and normalization ensure data quality and consistency. Subsequently, machine learning or deep learning algorithms are used to model the multi-source data, establishing a mapping relationship between seedling morphological phenotypes and field performance. For example, the model can be trained to predict key indicators such as seedling survival rate, seedling establishment period length, and growth rate changes under different environmental conditions. After model training is complete, a new batch of seedling production data and spectral data can be input to predict their field performance and seedling vigor under different environmental conditions, thereby achieving the intelligent assessment goal of observing seedling growth.

[0145] In practical applications, this invention can be embedded into factory-scale seedling management and decision-making platforms to achieve a closed-loop feedback mechanism for seedling vigor control and field performance prediction during the seedling stage. With the rapid development of intelligent and digital modern agriculture, the seedling vigor prediction model can provide data references for assessing dynamic quality adjustments in seedlings and predicting their field performance. This not only overcomes the problems of low efficiency and lack of follow-up on field performance after transplanting in traditional manual assessment methods, but also applies to the scientific quality grading of seedlings and the selection of commercial seedlings, improving the standardization of commercial seedlings.

[0146] The following describes the seedling vigor assessment device for plug seedling populations based on field growth performance provided by the present invention. The seedling vigor assessment device for plug seedling populations based on field growth performance described below can be referred to in correspondence with the seedling vigor assessment method for plug seedling populations based on field growth performance described above.

[0147] Figure 3 This is a schematic diagram of the structure of the seedling vigor assessment device for plug seedling populations based on field growth performance provided by the present invention, as shown below. Figure 3 As shown, the present invention also provides a device for evaluating the robustness of plug seedling populations based on field growth performance, comprising: The acquisition module 310 is used to acquire production data, full-spectrum data, and environmental data and field growth performance data of seedlings taken from the plug seedling population in different planting environments. The determination module 320 is used to determine the production information feature vector of the plug seedling population based on the production data, determine the spectral feature vector and spectral inversion feature vector of the plug seedling population based on the full-spectrum data, and determine the phenotypic feature vector of the plug seedling population based on the spectral feature vector and the spectral inversion feature vector. Training module 330 is used to take the production information feature vector, the spectral feature vector, the spectral inversion feature vector, the phenotypic feature vector, and the environmental feature vector corresponding to the environmental data as input, and take the field growth performance vector corresponding to the field growth performance data as output to train a field growth performance prediction model. The evaluation module 340 is used to evaluate the robustness of the plug seedling population based on the field growth performance prediction model and determine the evaluation results.

[0148] As one embodiment, the training module 330 is used for: Based on the physical conditions of the growth of the plug seedling population, a physical constraint function is constructed; A multi-task deep learning model is constructed. During the construction process, the physical constraint function is used as the post-processing module or the regularization term of the loss function of the multi-task deep learning model to obtain the initial prediction model. The initial prediction model is trained by taking the production information feature vector, the spectral feature vector, the spectral inversion feature vector, the phenotypic feature vector, and the environmental feature vector as inputs and the field growth performance vector as outputs, to obtain the field growth performance prediction model.

[0149] As one embodiment, the training module 330 is used for: Based on the matching relationship between seedling age and accumulated temperature of the plug seedling population, the constraint function for seedling age and accumulated temperature is determined. Based on the plant type index of the plug seedling population, the plant type constraint function is determined; Based on the temperature thresholds of the seedlings in different planting environments, an environmental temperature constraint function is determined. Based on the matching relationship between the seedling age and growth stage of the plug seedling population, the constraint function for growth days and developmental status is determined. The physical constraint function is determined based on the combination of the seedling age and accumulated temperature constraint function, the plant type constraint function, the ambient temperature constraint function, and the growth days and developmental state constraint function.

[0150] As one embodiment, the training module 330 is used for: A feature fusion layer is constructed, which is used to fuse the production information feature vector, the spectral feature vector, the spectral inversion feature vector, the phenotypic feature vector, and the environmental feature vector to obtain fused features; A physical constraint embedding layer is constructed, which is used to use the physical constraint function as a post-processing module or a regularization term of the loss function of the multi-task deep learning model. Based on the number of field growth performance indicators corresponding to the field growth performance data, a multi-task output layer is constructed. Based on the feature fusion layer, the physical constraint embedding layer, and the multi-task output layer, a multi-task deep learning model is determined.

[0151] As an example, when the physical constraint function is used as the regularization term of the loss function of the multi-task deep learning model, the feature fusion layer further includes a channel attention module, and the multi-task deep learning model further includes an adaptive fusion layer, which is used to adjust the weights of the phenotypic feature sub-vector and the environmental feature sub-vector based on the attention mechanism.

[0152] As one embodiment, the determining module 320 is used for: Based on the full-spectrum data, the spectral feature sub-vectors of the plug seedling population are determined; Based on the full-spectrum data and the pre-constructed spectral inversion model of plug seedlings, the spectral inversion feature vectors of the plug seedling population are determined; The seedling spectral inversion model is obtained by training a preset model based on the full-spectrum pre-training data of the seedling population and the destructive sampling feature data of the seedling population.

[0153] As one embodiment, the acquisition module 310 is used for: The production data, full-spectrum data, environmental data, and field growth performance data of the seedlings taken from the seedling population in different planting environments are obtained using automated data acquisition equipment.

[0154] It should be noted that the seedling vigor assessment device based on field growth performance of plug seedlings provided by this invention has the same technical effects as the seedling vigor assessment method based on field growth performance of plug seedlings, which will not be elaborated further.

[0155] Figure 4 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 4 As shown, the electronic device may include: a processor 410, a communications interface 420, a memory 430, and a communication bus 440, wherein the processor 410, the communications interface 420, and the memory 430 communicate with each other via the communication bus 440. The processor 410 can call logical instructions in the memory 430 to execute a method for evaluating the robustness of plug seedling populations based on field growth performance. This method includes: Acquire production data, full-spectrum data, and environmental data and field growth performance data of seedlings taken from the plug seedling population in different planting environments; Based on the production data, the production information feature vector of the plug seedling population is determined; based on the full-spectrum data, the spectral feature vector and the spectral inversion feature vector of the plug seedling population are determined; and based on the spectral feature vector and the spectral inversion feature vector, the phenotypic feature vector of the plug seedling population is determined. The production information feature vector, the spectral feature vector, the spectral inversion feature vector, the phenotypic feature vector, and the environmental feature vector corresponding to the environmental data are used as inputs, and the field growth performance vector corresponding to the field growth performance data is used as output to train a field growth performance prediction model. The robustness of the plug seedling population was assessed based on the field growth performance prediction model, and the assessment results were determined.

[0156] Furthermore, the logical instructions in the aforementioned memory 430 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0157] On the other hand, the present invention also provides a computer program product, the computer program product comprising a computer program that can be stored on a non-transitory computer-readable storage medium, wherein when the computer program is executed by a processor, the computer is able to execute the above-described methods for evaluating the robustness of plug seedling populations based on field growth performance, the method comprising: Acquire production data, full-spectrum data, and environmental data and field growth performance data of seedlings taken from the plug seedling population in different planting environments; Based on the production data, the production information feature vector of the plug seedling population is determined; based on the full-spectrum data, the spectral feature vector and the spectral inversion feature vector of the plug seedling population are determined; and based on the spectral feature vector and the spectral inversion feature vector, the phenotypic feature vector of the plug seedling population is determined. The production information feature vector, the spectral feature vector, the spectral inversion feature vector, the phenotypic feature vector, and the environmental feature vector corresponding to the environmental data are used as inputs, and the field growth performance vector corresponding to the field growth performance data is used as output to train a field growth performance prediction model. The robustness of the plug seedling population was assessed based on the field growth performance prediction model, and the assessment results were determined.

[0158] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described methods for evaluating the robustness of plug seedling populations based on field growth performance, the method comprising: Acquire production data, full-spectrum data, and environmental data and field growth performance data of seedlings taken from the plug seedling population in different planting environments; Based on the production data, the production information feature vector of the plug seedling population is determined; based on the full-spectrum data, the spectral feature vector and the spectral inversion feature vector of the plug seedling population are determined; and based on the spectral feature vector and the spectral inversion feature vector, the phenotypic feature vector of the plug seedling population is determined. The production information feature vector, the spectral feature vector, the spectral inversion feature vector, the phenotypic feature vector, and the environmental feature vector corresponding to the environmental data are used as inputs, and the field growth performance vector corresponding to the field growth performance data is used as output to train a field growth performance prediction model. The robustness of the plug seedling population was assessed based on the field growth performance prediction model, and the assessment results were determined.

[0159] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0160] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0161] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for evaluating the robustness of plug seedling populations based on field growth performance, characterized in that, include: Production data, full-spectrum data, and environmental data and field growth performance data of seedlings taken from the plug seedling population in different planting environments were obtained. Based on the production data, the production information feature vector of the plug seedling population is determined; based on the full-spectrum data, the spectral feature vector and the spectral inversion feature vector of the plug seedling population are determined; and based on the spectral feature vector and the spectral inversion feature vector, the phenotypic feature vector of the plug seedling population is determined. The production information feature vector, the spectral feature vector, the spectral inversion feature vector, the phenotypic feature vector, and the environmental feature vector corresponding to the environmental data are used as inputs, and the field growth performance vector corresponding to the field growth performance data is used as output to train a field growth performance prediction model. The robustness of the plug seedling population was assessed based on the field growth performance prediction model, and the assessment results were determined.

2. The method for evaluating the robustness of plug seedling populations based on field growth performance according to claim 1, characterized in that, The process involves taking the production information feature vector, the spectral feature vector, the spectral inversion feature vector, the phenotypic feature vector, and the environmental feature vector corresponding to the environmental data as inputs, and taking the field growth performance vector corresponding to the field growth performance data as output to train a field growth performance prediction model, including: Based on the physical conditions of the growth of the plug seedling population, a physical constraint function is constructed; A multi-task deep learning model is constructed. During the construction process, the physical constraint function is used as the post-processing module or the regularization term of the loss function of the multi-task deep learning model to obtain the initial prediction model. The initial prediction model is trained by taking the production information feature vector, the spectral feature vector, the spectral inversion feature vector, the phenotypic feature vector, and the environmental feature vector as inputs and the field growth performance vector as outputs, to obtain the field growth performance prediction model.

3. The method for evaluating the robustness of plug seedling populations based on field growth performance according to claim 2, characterized in that, The physical constraint function constructed based on the growth physical conditions of the plug seedling population includes: Based on the matching relationship between seedling age and accumulated temperature of the aforementioned plug seedling population, the constraint function for seedling age and accumulated temperature is determined. Based on the plant type index of the plug seedling population, the plant type constraint function is determined; Based on the temperature thresholds of the seedlings in different planting environments, an environmental temperature constraint function is determined; Based on the matching relationship between the seedling age and growth stage of the plug seedling population, the constraint function for growth days and developmental status is determined. The physical constraint function is determined based on the combination of the seedling age and accumulated temperature constraint function, the plant type constraint function, the ambient temperature constraint function, and the growth days and developmental state constraint function.

4. The method for evaluating the robustness of plug seedling populations based on field growth performance according to claim 2, characterized in that, The construction of the multi-task deep learning model includes: A feature fusion layer is constructed, which is used to fuse the production information feature vector, the spectral feature vector, the spectral inversion feature vector, the phenotypic feature vector, and the environmental feature vector to obtain fused features; A physical constraint embedding layer is constructed, which is used to use the physical constraint function as a post-processing module or a regularization term of the loss function of the multi-task deep learning model. Based on the number of field growth performance indicators corresponding to the field growth performance data, a multi-task output layer is constructed. Based on the feature fusion layer, the physical constraint embedding layer, and the multi-task output layer, a multi-task deep learning model is determined.

5. The method for evaluating the robustness of plug seedling populations based on field growth performance according to claim 4, characterized in that, When the physical constraint function is used as the regularization term of the loss function of the multi-task deep learning model, the feature fusion layer further includes a channel attention module, and the multi-task deep learning model further includes an adaptive fusion layer, which is used to adjust the weights of the phenotypic feature sub-vector and the environmental feature sub-vector based on the attention mechanism.

6. The method for evaluating the robustness of plug seedling populations based on field growth performance according to claim 1, characterized in that, The determination of the spectral feature vectors and spectral inversion feature vectors of the plug seedling population based on the full-spectrum data includes: Based on the full-spectrum data, the spectral feature sub-vectors of the plug seedling population are determined; Based on the full-spectrum data and the pre-constructed spectral inversion model of plug seedlings, the spectral inversion feature vectors of the plug seedling population are determined; The seedling spectral inversion model is obtained by training a preset model based on the full-spectrum pre-training data of the seedling population and the destructive sampling feature data of the seedling population.

7. The method for evaluating the robustness of plug seedling populations based on field growth performance according to claim 1, characterized in that, The acquisition of production data, full-spectrum data, and environmental data and field growth performance data of seedlings taken from the plug seedling population in different planting environments includes: The production data, full-spectrum data, environmental data, and field growth performance data of the seedlings in different planting environments are acquired using automated data acquisition equipment.

8. A device for evaluating the robustness of plug seedling populations based on field growth performance, characterized in that, include: The acquisition module is used to acquire production data, full-spectrum data, and environmental data and field growth performance data of seedlings taken from the plug seedling population in different planting environments. The determination module is used to determine the production information feature vector of the plug seedling population based on the production data, determine the spectral feature vector and spectral inversion feature vector of the plug seedling population based on the full-spectrum data, and determine the phenotypic feature vector of the plug seedling population based on the spectral feature vector and the spectral inversion feature vector. The training module is used to take the production information feature vector, the spectral feature vector, the spectral inversion feature vector, the phenotypic feature vector, and the environmental feature vector corresponding to the environmental data as inputs, and take the field growth performance vector corresponding to the field growth performance data as outputs to train a field growth performance prediction model. The evaluation module is used to evaluate the robustness of the plug seedling population based on the field growth performance prediction model and determine the evaluation results.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the method for evaluating the robustness of plug seedling populations based on field growth performance as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the method for evaluating the robustness of plug seedling populations based on field growth performance as described in any one of claims 1 to 7.

Images