Method, device and equipment for predicting early growth vigor of rice and medium

Abstract

The application relates to a rice early growth vigor prediction method, device, equipment and medium, and relates to the field of data processing.The method comprises the following steps: generating a single rice canopy height model according to laser radar point cloud data to extract the plant height of the single rice; determining the leaf age and tiller number of the single rice according to the laser radar point cloud data corresponding to the preset number of days after rice transplanting based on a structure feature extraction model; constructing the structure feature data of the single rice according to the plant height, the leaf age and the tiller number; obtaining a first sample data set; inputting the current spectral feature data and the current structure feature data of the single rice of a target variety into an early growth vigor prediction model trained by using the first sample data set to predict the early growth vigor score of the single rice of the target variety at the preset number of days after rice transplanting.The application realizes the quantitative evaluation of the early growth vigor of single rice at the single rice scale in a field environment and breaks the traditional mode of relying on manual judgment.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of data processing, and in particular to a rice early growth vigor prediction method, a corresponding device, an electronic device and a computer readable storage medium. BACKGROUND

[0002] As the staple food source for half of the world's population, rice early growth vigor directly affects photosynthetic efficiency and material accumulation capacity, which is a key trait determining yield, especially in the double-cropping rice planting system, strong early growth vigor varieties can significantly shorten the growth cycle and achieve annual yield breakthrough. Early growth vigor is a complex trait, which is a comprehensive reflection of the change of phenotypic parameters such as greening, tillering and biomass in the time dimension. At present, due to the limitation of methodology, the evaluation of rice early growth vigor mainly relies on the manual judgment of experienced breeders and cultivation scientists, lacking a standardized evaluation system and corresponding quantitative indicators, which is time-consuming and labor-intensive, and it is difficult to make large-scale evaluation in field environment. This seriously restricts the breeding of strong early growth vigor rice varieties and the matching and integration of high-yield and efficient cultivation techniques.

[0003] In recent years, unmanned aerial vehicle remote sensing has become an effective tool for crop phenotype monitoring. The unmanned aerial vehicle platform can carry multiple sensors at a time to invert or directly measure the crop phenotypic traits in the field. The red edge (RE) and near-infrared (NIR) band range provided by the multispectral sensor has been proven to be able to generate a variety of spectral indices related to crop vigor, morphology and density, and chlorophyll content and other phenotypic parameters. However, since optical orthographic images can only provide two-dimensional planar features of crop canopy, the ability to extract phenotypic information in three-dimensional space is limited. Therefore, many researchers have realized the three-dimensional reconstruction of the study area by combining the structure from motion algorithm (SfM) with the multi-view stereo (MVS) algorithm. However, optical remote sensing data is prone to saturation in high biomass or leaf area index (LAI) areas, and digital cameras cannot penetrate the vegetation canopy to obtain canopy internal data. The point cloud model obtained by reconstruction has low density and the collection method is complex and tedious, which is not suitable for large-scale extraction of relatively fine crop three-dimensional features.

[0004] Tillering is one of the core indicators for evaluating the early growth vigor of rice, and its accurate monitoring has long been a technical bottleneck. Current research on tiller monitoring based on sensors can be divided into three categories, including direct measurement based on image analysis, indirect measurement based on crop spatial morphology, and inversion based on vegetation index. The image-based method has strict requirements for image quality, and therefore has low data collection efficiency, which is not conducive to large-scale application in field environment. The measurement method based on crop spatial morphology mainly uses ground-based LiDAR or indoor three-dimensional scanner to obtain high-precision original point cloud data, but the feasibility of this method in field environment still needs to be verified by experiments. Although the multi-spectral index can indirectly reflect the tillering trend of the group, it cannot achieve accurate analysis at the single plant scale. This contradiction between "insufficient group inversion accuracy and low single plant measurement efficiency" has resulted in the fact that tiller number has not been included in the large-scale phenotype group analysis system.

[0005] LiDAR provides a new possibility for breaking this bottleneck. LiDAR, as an active sensing technology, measures the distance from the sensor to the target by emitting short-wavelength laser and recording the laser speed and flight time according to the timer. Since the short-wavelength laser has a certain penetration ability for the vegetation canopy, it makes up for the shortcomings of traditional optical images, which makes LiDAR considered as the most potential technology for quantifying vegetation structure parameters. LiDAR technology has been successfully applied in high-throughput crop phenotype parameter extraction such as plant height, leaf area index and biomass. However, existing researches mainly focus on simple morphological parameters, and there is still a lack of effective point cloud deep learning algorithm for fine structure characteristics such as tillering. The traditional gridding method causes the loss of three-dimensional features, and the segmentation network based on regular voxels is difficult to adapt to the topological heterogeneity of rice tillering. This leads to the fact that the current LiDAR technology has not realized the high-throughput analysis of tiller number.

[0006] The fusion of LiDAR point cloud and spectral image provides information combining 3D structure and spectral attributes of vegetation. Most of the current fusion methods are to rasterize the point cloud and align it with the spectral data space, which preliminarily solves the fusion problem of point cloud and spectral data, but this will cause a large amount of loss of point cloud data features.

[0007] In summary, the multi-spectral index in the existing technology can indirectly reflect the tillering trend of the group, but cannot achieve accurate analysis at the single plant scale, and the traditional gridding method causes the loss of three-dimensional features, and the segmentation network based on regular voxels is difficult to adapt to the topological heterogeneity of rice tillering, which leads to the fact that the current LiDAR technology has not realized the high-throughput analysis of tiller number. The applicant made corresponding exploration to solve this problem. SUMMARY

[0008] The application aims to solve the above problems and provide a rice early growth vigor prediction method, a corresponding device, an electronic device and a computer readable storage medium.

[0009] To achieve the above purposes, the application adopts the following technical solutions:

[0010] A rice early growth vigor prediction method proposed to adapt to one of the purposes of the application, comprising:

[0011] Obtain multispectral image data and laser radar point cloud data corresponding to a preset number of days after rice transplanting of a plurality of single rice plants of a target variety, determine spectral feature data of the single rice plants according to the multispectral image data;

[0012] Generate a canopy height model of the single rice plants according to the laser radar point cloud data to extract the plant height of the single rice plants, determine the leaf age and tiller number of the single rice plants according to the laser radar point cloud data corresponding to the preset number of days after rice transplanting based on a structure feature extraction model trained to a convergence state, and construct structure feature data of the single rice plants according to the plant height, the leaf age and the tiller number;

[0013] Obtain a first sample data set, wherein the first sample data set includes a plurality of first training samples and corresponding first sample labels, the first training samples represent spectral feature time series data constructed by spectral feature data collected at a preset number of days after rice transplanting of single rice plants, and structure feature time series data constructed by structure feature data, and the first sample labels represent early growth vigor scores of the rice;

[0014] Input current spectral feature data and current structure feature data of single rice plants of a target variety into an early growth vigor prediction model trained using the first sample data set to predict early growth vigor scores of single rice plants of the target variety at a preset number of days after rice transplanting, thereby completing the prediction of rice early growth vigor.

[0015] Optionally, the spectral feature data includes a normalized vegetation index, a chlorophyll index, a green normalized vegetation index, an optimized soil-adjusted vegetation index, a normalized red edge difference index and a canopy coverage rate;

[0016] The basic network architecture of the structure feature extraction model is a point cloud deep learning network, wherein the point cloud deep learning network comprises an input layer, an encoder and a decoder, the encoder comprises two T-Net micro networks and two multi-layer perceptrons, the encoder comprises a first T-Net micro network, a first multi-layer perceptron, a second T-Net micro network, a second multi-layer perceptron and a maximum pooling layer connected in sequence; the decoder comprises a third multi-layer perceptron and a fourth multi-layer perceptron connected in sequence;

[0017] The basic network architecture of the early growth vigor prediction model is a long short-term memory network based on a time attention mechanism, wherein the long short-term memory network based on the time attention mechanism comprises an input layer, a double-branch LSTM module, a feature fusion module and a prediction layer connected in sequence, the double-branch LSTM module comprises a first LSTM branch module and a second LSTM branch module, wherein the first LSTM branch module is used for processing spectral feature time series data, the second LSTM branch module is used for processing structure feature time series data, and the first LSTM branch module and the second LSTM branch module each comprise an LSTM processing unit corresponding to a plurality of time steps, and each time step corresponding LSTM processing unit is connected with a time attention mechanism module.

[0018] Optionally, the step of training the structure feature extraction model comprises:

[0019] obtaining a second sample data set, wherein the second sample data set comprises a plurality of second training samples and corresponding second sample labels, the second training samples represent laser radar point cloud data corresponding to a preset number of days after rice transplanting of a single rice plant, and the second sample labels represent leaf age and tiller number of the single rice plant;

[0020] inputting the plurality of second training samples and corresponding second sample labels into a preset structure feature extraction model, generating a 3x3 affine transformation matrix in the first T-Net micro network to perform three-dimensional space alignment on the laser radar point cloud data, and performing dimensionality reduction from 3 dimensions to 64 dimensions in the first multi-layer perceptron to extract local features,

[0021] generating a 64x64 affine transformation matrix in the second T-Net micro network to perform feature space alignment on the 64-dimensional local features, performing dimensionality reduction from 64 dimensions to 1024 dimensions in the second multi-layer perceptron, and aggregating the 1024-dimensional features in the maximum pooling layer to generate a global feature vector;

[0022] After the decoder receives the global feature vector, the 1024-dimensional global feature vector is reduced to 2-dimensional in the third multi-layer perception, and the 2-dimensional global feature vector is fused with the preset number of days after rice transplanting for nonlinear transformation in the fourth multi-layer perception to output the leaf age prediction value and the tiller number prediction value of the single rice plant;

[0023] The above steps are repeated until the structural feature extraction model is trained to a convergent state, so as to determine the structural feature extraction model trained to the convergent state.

[0024] Optionally, the step of inputting the current spectral feature data and the current structural feature data of the single rice plant of the target variety into the early growth vigor prediction model trained by using the first sample data set to predict the early growth vigor score of the single rice plant of the target variety at the preset number of days after rice transplanting comprises:

[0025] The current spectral feature time series data and the current structural feature time series data of the single rice plant of the target variety are input into the input layer of the early growth vigor prediction model.

[0026] The spectral feature time series data is subjected to time dynamic feature extraction by the LSTM processing unit at each time step in the first LSTM branch module, and the structural feature time series data is subjected to time dynamic feature extraction by the LSTM processing unit at each time step in the second LSTM branch module.

[0027] In the time attention mechanism module corresponding to each time step, the spectral feature data or the structural feature data output by the corresponding LSTM processing unit is calculated to obtain a spectral feature weighted vector or a structural feature weighted vector corresponding to the LSTM processing unit at each time step.

[0028] After the spectral feature data at all time steps is weighted and aggregated into a spectral feature global weighted vector by the respective corresponding time attention mechanism module, and the structural feature data at all time steps is weighted and aggregated into a structural feature global weighted vector by the respective corresponding time attention mechanism module.

[0029] In the feature fusion module, the spectral feature global weighted vector and the structural feature global weighted vector are spliced and fused to obtain a global feature vector, and in the prediction layer, the global feature vector is subjected to nonlinear mapping processing by a fully connected layer and a Sigmoid activation function to output the early growth vigor score of the single rice plant of the target variety at the preset number of days after rice transplanting.

[0030] Optionally, the step of determining the early growth vigor score comprises:

[0031] obtaining the number of days required for the target variety of rice to reach the effective tillering number, the maximum value of the number of days required for all varieties of rice to reach the effective tillering number, the minimum value of the number of days required for all varieties of rice to reach the effective tillering number, the number of days required for the target variety of rice to reach the time node at which the leaf age increases at the fastest speed, the maximum value of the number of days required for all varieties of rice to reach the time node at which the leaf age increases at the fastest speed, and the minimum value of the number of days required for all varieties of rice to reach the time node at which the leaf age increases at the fastest speed;

[0032] calculating a first difference value between the number of days required for the target variety of rice to reach the effective tillering number and the minimum value of the number of days required for all varieties of rice to reach the effective tillering number, calculating a second difference value between the maximum value of the number of days required for all varieties of rice to reach the effective tillering number and the minimum value of the number of days required for all varieties of rice to reach the effective tillering number, and calculating a first ratio value between the first difference value and the second difference value;

[0033] calculating a third difference value between the number of days required for the target variety of rice to reach the time node at which the leaf age increases at the fastest speed and the minimum value of the number of days required for all varieties of rice to reach the time node at which the leaf age increases at the fastest speed, calculating a fourth difference value between the maximum value of the number of days required for all varieties of rice to reach the time node at which the leaf age increases at the fastest speed and the minimum value of the number of days required for all varieties of rice to reach the time node at which the leaf age increases at the fastest speed, and calculating a second ratio value between the third difference value and the fourth difference value;

[0034] calculating a first sum value between the first ratio value and the second ratio value, calculating a third ratio value between the first sum value and a first number, and determining the early growth vigor score according to a difference value between the first number and the third ratio value.

[0035] Optionally, the step of generating the canopy height model of the single rice plant according to the laser radar point cloud data comprises:

[0036] obtaining laser radar point cloud data corresponding to a preset number of days after rice transplanting of the target variety of single rice plant;

[0037] generating a digital surface model and a digital elevation model of the single rice plant according to the laser radar point cloud data, and determining the canopy height model of the single rice plant according to a difference value between the digital surface model and the digital elevation model;

[0038] extracting the plant height of the single rice plant according to the canopy height model of the single rice plant.

[0039] Optionally, the step of determining the spectral feature data of the single rice plant according to the multi-spectral image data comprises:

[0040] Acquire the multispectral image data corresponding to the preset number of days after rice transplanting of the single rice plant of the target variety, preprocess the multispectral image data to determine the multispectral reflectance corresponding to the preset number of days after rice transplanting of the single rice plant, wherein the multispectral reflectance includes green waveband reflectance, red waveband reflectance, red edge waveband reflectance, and near-infrared waveband reflectance;

[0041] According to the multispectral reflectance, calculate and determine the spectral feature data corresponding to the preset number of days after rice transplanting of the single rice plant.

[0042] Another object of the present application is to provide a rice early growth vigor prediction device, comprising:

[0043] A spectral feature determination module is configured to acquire multispectral image data and laser radar point cloud data corresponding to the preset number of days after rice transplanting of a plurality of single rice plants of a target variety, and determine spectral feature data of the single rice plants according to the multispectral image data;

[0044] A structural feature determination module is configured to generate a canopy height model of the single rice plant according to the laser radar point cloud data to extract the plant height of the single rice plant, determine the leaf age and tiller number of the single rice plant based on a structural feature extraction model trained to a convergent state according to the laser radar point cloud data corresponding to the preset number of days after rice transplanting, and construct structural feature data of the single rice plant according to the plant height, leaf age, and tiller number.

[0045] A data set acquisition module is configured to acquire a first sample data set, wherein the first sample data set includes a plurality of first training samples and corresponding first sample labels, the first training samples represent spectral feature time series data constructed by spectral feature data collected on the preset number of days after rice transplanting of a single rice plant, and structural feature time series data constructed by structural feature data, and the first sample labels represent early growth vigor scores of the rice.

[0046] A growth vigor prediction module is configured to input current spectral feature data and current structural feature data of a single rice plant of a target variety into an early growth vigor prediction model trained using the first sample data set to predict the early growth vigor score of the single rice plant of the target variety on the preset number of days after rice transplanting, thereby completing the prediction of the early growth vigor of the rice.

[0047] Another object of the present application is to provide an electronic device comprising a central processing unit and a memory, wherein the central processing unit is configured to call and run a computer program stored in the memory to perform the steps of the rice early growth vigor prediction method described in the present application.

[0048] Another object of the present application is to provide a computer readable storage medium storing a computer program implemented according to the rice early growth vigor prediction method in the form of computer readable instructions, which, when invoked and run by a computer, performs the steps included in the corresponding method.

[0049] Compared with the prior art, the present application is directed to the problems in the prior art that the multispectral index can only indirectly reflect the tillering trend of the group, but cannot realize accurate analysis at the single plant scale, and that the three-dimensional characteristics are lost due to the traditional gridding method, and the segmentation network based on regular voxels is difficult to adapt to the topological heterogeneity of rice tillering, which leads to the fact that the current LiDAR technology has not realized high-throughput analysis of the tiller number, and the present application includes but is not limited to the following beneficial effects:

[0050] Firstly, the rice early growth vigor prediction method of the present application solves the problem of difficulty in extracting fine structure features such as tiller number and leaf age, designs an end-to-end point cloud deep learning network, realizes point cloud spatial alignment and standardization through a T-Net micro network, completes feature dimensionality reduction and dimensionality reduction through a multilayer perceptron, and integrates a time parameter, thereby breaking through the bottleneck of extracting the tiller number and leaf age at the single plant scale, the determination coefficient (R²) of the network for predicting the tiller number reaches 0.91, and the R² for predicting the leaf age reaches 0.97, which solves the contradiction between the insufficient group inversion accuracy and the low single plant measurement efficiency of the traditional method, and realizes high-throughput accurate analysis of key growth parameters. At the same time, the LSTM network with a two-level feature fusion structure is innovatively adopted to integrate multi-modal time series data, and the importance of data at different growth stages is quantified through an attention mechanism, thereby overcoming the defects of insufficient stability and explainability of traditional static phenotype analysis.

[0051] Secondly, the rice early growth vigor prediction method of the present application integrates the LiDAR point cloud deep learning features and the model of multispectral data, and the R² for predicting the early growth vigor reaches 0.79, and the RMSE is only 0.07, which is significantly better than the model using LiDAR (R²=0.72, RMSE=0.08) or multispectral data (R²=0.54, RMSE=0.11) alone. The three-dimensional structure parameters provided by LiDAR make up for the deficiency of the spectral index in the three-dimensional characterization of the canopy, and the physiological parameters of multispectral data enhance the understanding of the model for biochemical processes, and the two are dynamically complementary at different growth stages, and the structural parameters contribute more within 12 days after transplanting, and the contribution of the spectral index is highlighted after the canopy is closed. Further optimization of the input parameter combination finds that the prediction accuracy is the highest (R²=0.84) when the tiller number, leaf age and plant height are fused with the multispectral index, which fully verifies the complementary effect of multi-modal data.

[0052] Thirdly, the rice early growth vigor prediction method of the present application can accurately locate the period of 12 to 20 days after transplanting as the key period affecting the prediction of early growth vigor through the time attention mechanism visualization, and the canopy phenotype data at this stage has the highest contribution weight, which shortens the window period of traditional experience judgment. At the same time, it is found that the leaf age has the highest prediction value among single parameters, and the dynamic synergy of structural parameters and spectral index is the core influencing factor of early growth vigor. In addition, the relationship between early growth vigor and yield is clarified, and high-yield varieties generally have high early growth vigor, but high-vigor varieties are not necessarily high-yield, which confirms that early growth and fast development are necessary but not sufficient conditions for high yield, and the proportion of "high yield + high score" and "low yield + low score" varieties is more than 75%, which provides a new perspective for the study of yield formation mechanism.

[0053] Fourthly, the rice early growth vigor prediction method of the present application realizes the quantitative evaluation of rice early growth vigor at the single plant scale in the field environment, breaks the traditional mode relying on manual judgment, and provides a standardized technical means for early growth vigor breeding of rice varieties. By comparing the early growth vigor scores of different rice varieties, data support is provided for variety adaptability evaluation. At the same time, the research clearly defines the management priorities of different growth type varieties, such as strengthening the late stage regulation for "high score + low yield" varieties and optimizing the early management for "low score + high yield" varieties, which provides individualized solutions for precision agricultural management. In addition, the constructed technical system can be popularized to the field of crop phenotype research, and lays a solid foundation for the application of multi-modal remote sensing technology in agricultural intelligence. BRIEF DESCRIPTION OF DRAWINGS

[0054] The above and / or additional aspects and advantages of the present application will become apparent and more readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings, in which:

[0055] Figure 1 A flowchart of the rice early growth vigor prediction method in the embodiments of the present application is shown;

[0056] Figure 2 An exemplary network architecture of the point cloud deep learning network in the embodiments of the present application is shown;

[0057] Figure 3 A scatter plot of the tiller number prediction of the point cloud deep learning network in the embodiments of the present application is shown;

[0058] Figure 4 A scatter plot of the leaf age prediction of the point cloud deep learning network in the embodiments of the present application is shown;

[0059] Figure 5 A time series change graph of the leaf age prediction of the point cloud deep learning network in the embodiments of the present application is shown;

[0060] Figure 6A time series change diagram of the point cloud deep learning network for tiller number prediction in an embodiment of the present application;

[0061] Figure 7 A diagram showing the number of days required for the hybrid American silk seedlings to reach the effective tiller number in an embodiment of the present application;

[0062] Figure 8 A flowchart showing the number of days required for the time node at which the leaf age of the hybrid American silk seedlings increases at the fastest speed in an embodiment of the present application;

[0063] Figure 9 An exemplary network architecture of the long short-term memory network based on the time attention mechanism in an embodiment of the present application;

[0064] Figure 10 A scatter plot of the predicted value and the true value of the multispectral and laser radar fusion model in an embodiment of the present application;

[0065] Figure 11 A histogram of the error of the prediction of the multispectral and laser radar fusion model in an embodiment of the present application;

[0066] Figure 12 A principle block diagram of the early growth vigor prediction device for rice in an embodiment of the present application;

[0067] Figure 13 A structural schematic diagram of the computer device in an embodiment of the present application. DETAILED DESCRIPTION

[0068] The embodiments of the present application will be described in detail below, examples of which are shown in the accompanying drawings, in which the same or similar reference numerals represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the accompanying drawings are exemplary and are only used to explain the present application, and cannot be interpreted as a limitation on the present application.

[0069] Unless it is explicitly stated that the various embodiments to be disclosed herein are mutually exclusive, the technical features involved in each of the embodiments can be combined flexibly to construct new embodiments, as long as such combination does not deviate from the spirit of the present application and can meet the needs of the prior art or solve some deficiencies in the prior art. For this variation, those skilled in the art should know.

[0070] Please refer to Figure 1 The early growth vigor prediction method for rice in an embodiment of the present application comprises:

[0071] In step S10, multispectral image data and laser radar point cloud data corresponding to a preset number of days after rice transplanting of a plurality of single rice plants of a target variety are acquired, and spectral feature data of the single rice plants is determined according to the multispectral image data.

[0072] The rice early growth vigor prediction system in the terminal device can acquire multispectral image data and laser radar point cloud data of a plurality of single rice plants of a target variety corresponding to a preset number of days after rice transplanting, and determine spectral feature data of the single rice plants according to the multispectral image data; the preset number of days can be 4 days, 8 days, 12 days, 16 days, etc., and a person skilled in the art can determine the preset number of days as needed according to business scenario requirements; the spectral feature data includes normalized difference vegetation index (NDVI), chlorophyll index (LCI), green normalized difference vegetation index (GNDVI), optimized soil-adjusted vegetation index (OSAVI), normalized difference red edge (NDRE), and canopy coverage (Canopy coverage).

[0073] In some embodiments, the normalized difference vegetation index (NDVI) is calculated according to the following formula:

[0074]

[0075] wherein NDVI represents the normalized difference vegetation index; NIR represents the near-infrared band reflectance; and R represents the red band reflectance.

[0076] The chlorophyll index (LCI) is calculated according to the following formula:

[0077]

[0078] wherein LCI represents the chlorophyll index; NIR represents the near-infrared band reflectance; RE represents the red edge band reflectance; and R represents the red band reflectance.

[0079] The green normalized difference vegetation index (GNDVI) is calculated according to the following formula:

[0080]

[0081] wherein GNDVI represents the green normalized difference vegetation index; NIR represents the near-infrared band reflectance; and G represents the green band reflectance.

[0082] The optimized soil-adjusted vegetation index (OSAVI) is calculated according to the following formula:

[0083]

[0084] wherein OSAVI represents the optimized soil-adjusted vegetation index; NIR represents the near-infrared band reflectance; and R represents the red band reflectance.

[0085] The normalized difference red edge (NDRE) is calculated according to the following formula: ​​​​

[0086] ,

[0087] wherein, NDRE represents a normalized red edge difference index; NIR represents a near infrared band reflectance; and RE represents a red edge band reflectance.

[0088] The formula for calculating the canopy coverage is:

[0089] ,

[0090] In the process of calculating the spectral feature data by using the unmanned aerial vehicle multi-spectral data to analyze the vegetation growth condition, there are soil pixels (i.e. bare soil area) and shadow pixels (such as shadow area formed by object obstruction). These pixels are not vegetation themselves, and if they are included in the calculation of the vegetation index, they will interfere with the results and produce confounding effects, that is, the vegetation index cannot accurately reflect the real vegetation growth situation. Therefore, a normalized difference vegetation index (NDVI) threshold is set, wherein the NDVI threshold can be 0.45, and the pixels with a normalized difference vegetation index (NDVI) less than 0.45 are marked as background pixels, which means that in the subsequent calculation process of the vegetation index, these background pixels (soil, weeds and shadows, etc. non-crown layer pixels) will be excluded and will not participate in the calculation, so that the vegetation index can more accurately reflect the actual growth state of the vegetation (rice), and the accuracy of the analysis result is improved. After excluding the soil, weeds and shadows, etc. non-crown layer pixels, the number of effective crown layer pixels of the single rice plant is determined.

[0091] In some embodiments, strong early growth vigor refers to the characteristic of vigorous growth of rice during the vegetative growth period after transplanting, which improves the photosynthesis time and efficiency of rice, and is an important basis for rice variety breeding and high-yield and efficient cultivation, especially in double-cropping rice planting areas. However, the current strong early growth vigor characteristic lacks precise quantitative description and efficient evaluation technology in field environment, which restricts the efficiency of rice variety breeding and the research and development of high-yield cultivation technology.

[0092] In some embodiments, the experimental area can be selected as a double-cropping rice planting area in a certain place. In order to ensure the diversity of experimental data, 200 international rice varieties from all over the world and 100 rice varieties of different years in a certain place are selected, a total of 300 rice varieties, and a repetition area is set. Therefore, there are a total of 600 plots. Each plot is planted with 7 rows and 8 columns and 56 plants, with a row spacing of 6 inches and a plant spacing of 5 inches. Each hole is uniformly planted with one seedling to facilitate the later tiller statistics. The cultivation measures are in accordance with the local traditional methods, and three kinds of fertilizers, urea, potassium chloride and superphosphate, are applied. The date of transplanting is August 6, 2024. Within 40 days after transplanting, unmanned aerial vehicle multispectral and laser radar data are collected every 4 days, and the tillers and leaf ages of 3 rice plants at fixed positions in the plot are manually counted, a total of 10 periods of data are collected.

[0093] In some embodiments, the step of determining the spectral feature data of the single rice plant according to the multispectral image data comprises:

[0094] Step S1001, obtaining the multispectral image data of the single rice plant of the target variety corresponding to the preset number of days after rice transplanting, preprocessing the multispectral image data to determine the multispectral reflectance of the single rice plant corresponding to the preset number of days after rice transplanting, wherein the multispectral reflectance includes green waveband reflectance, red waveband reflectance, red edge waveband reflectance and near-infrared waveband reflectance;

[0095] Step S1002, calculating and determining the spectral feature data of the single rice plant corresponding to the preset number of days after rice transplanting according to the multispectral reflectance.

[0096] Specifically, DJI Mavic 3M can be used for multispectral image data acquisition. The aircraft integrates a multispectral sensor and a visible light sensor camera, which can simultaneously collect 5 million pixel multispectral image data containing green waveband (560±16nm), red waveband (650±16nm), red edge waveband (730±16nm) and near-infrared waveband (860±26nm), and 20 million pixel visible light image. Moreover, the unmanned aerial vehicle integrates a differential positioning module (RTK), which can realize spatial positioning within 1 cm horizontally and 1.5 cm vertically. In order to balance the reconstruction success rate, ground resolution and collection efficiency, the flight height is 12 meters when collecting data, the visible light ground resolution is 0.36 cm, and the multispectral ground resolution is 0.62 cm. The image heading overlap rate is 80%, and the lateral overlap rate is 70%. All data are collected between 11am and 2pm in sunny weather. After data collection, the preset 3D modeling software (DJITerra) is used to stitch the images taken by the unmanned aerial vehicle into an orthographic map, and three standard reflectivity gray panels with reflectivity of 25%, 50% and 75% and area of 20cm*20cm are used for radiation correction to convert the gray value (DN) to reflectivity.

[0097] In some embodiments, the spectral feature data includes a normalized difference vegetation index, a chlorophyll index, a green normalized difference vegetation index, an optimized soil adjusted vegetation index, a normalized difference red edge index, and a canopy coverage; the base network architecture of the structure feature extraction model is a point cloud deep learning network, wherein the point cloud deep learning network includes an input layer, an encoder, and a decoder, the encoder includes two T-Net micro networks and two multilayer perceptrons, wherein the encoder includes a first T-Net micro network, a first multilayer perceptron, a second T-Net micro network, a second multilayer perceptron, and a maximum pooling layer connected in sequence; the decoder includes a third multilayer perceptron and a fourth multilayer perceptron connected in sequence;

[0098] The basic network architecture of the early growth vigor prediction model is a long short-term memory network based on a time attention mechanism, wherein the long short-term memory network based on the time attention mechanism comprises an input layer, a double-branch LSTM module, a feature fusion module and a prediction layer connected in sequence, the double-branch LSTM module comprises a first LSTM branch module and a second LSTM branch module, wherein the first LSTM branch module is used for processing spectral feature time series data, and the second LSTM branch module is used for processing structural feature time series data, and the first LSTM branch module and the second LSTM branch module each comprise an LSTM processing unit corresponding to a plurality of time steps, and each time step corresponding LSTM processing unit is connected with a time attention mechanism module.

[0099] In step S20, the canopy height model of the single rice plant is generated according to the laser radar point cloud data to extract the plant height of the single rice plant, the leaf age and the tiller number of the single rice plant are determined based on the structural feature extraction model trained to a convergent state according to the laser radar point cloud data corresponding to the preset number of days after rice transplanting, and the structural feature data of the single rice plant is constructed according to the plant height, the leaf age and the tiller number.

[0100] In step S20, the canopy height model of the single rice plant is generated according to the laser radar point cloud data to extract the plant height of the single rice plant, the leaf age and the tiller number of the single rice plant are determined based on the structural feature extraction model trained to a convergent state according to the laser radar point cloud data corresponding to the preset number of days after rice transplanting, and the structural feature data of the single rice plant is constructed according to the plant height, the leaf age and the tiller number.

[0101] In some embodiments, the step of generating the canopy height model of the single rice plant according to the laser radar point cloud data to extract the plant height of the single rice plant comprises:

[0102] Step S201, acquiring laser radar point cloud data corresponding to a preset number of days after rice seedling transplanting of a single rice plant of a target variety;

[0103] Step S202, generating a digital surface model and a digital elevation model of the single rice plant according to the laser radar point cloud data, and determining a canopy height model of the single rice plant according to a difference between the digital surface model and the digital elevation model;

[0104] Step S203, extracting a plant height of the single rice plant according to the canopy height model of the single rice plant.

[0105] Specifically, the DJI M350 RTK aircraft can be used to carry the ZENMUSE L2 laser radar to collect point cloud data. This system integrates Livox laser radar module, high-precision inertial navigation (IMU), differential positioning module (RTK), surveying camera, three-axis gimbal and other modules. The flight speed during data collection is 0.7 m / s, the flight height is 20 m, and the point cloud density is 70452 points / m2. The 3D modeling software (DJI Terra) is used to read the laser radar point cloud data, inertial navigation record, track file and other related data, and perform POS solving, point cloud coloring, splicing and other operations. After completion, the las point cloud file is generated. The las file data includes the three-dimensional coordinates of each point, scanning angle, echo number, return pulse intensity and GPS timestamp. The LiDAR360 software is used for cutting, denoising, point cloud registration, ground filtering and other preprocessing to generate digital surface model (DSM), digital elevation model (DEM) and canopy height model (CHM).

[0106] After generating the digital surface model (DSM), the digital elevation model (DEM) and the canopy height model (CHM), the canopy height model of the single rice plant is determined according to the difference between the digital surface model and the digital elevation model, and the plant height of the single rice plant is extracted according to the canopy height model of the single rice plant.

[0107] In some embodiments, the step of training the structure feature extraction model comprises:

[0108] Step S2001, acquiring a second sample data set, wherein the second sample data set includes a plurality of second training samples and their corresponding second sample labels, the second training samples represent laser radar point cloud data corresponding to a preset number of days after rice seedling transplanting of a single rice plant, and the second sample labels represent leaf age and tiller number of the single rice plant;

[0109] Step S2002, input the plurality of second training samples and their corresponding second sample labels into a preset structure feature extraction model, generate a 3x3 affine transformation matrix in the first T-Net micro network to perform three-dimensional space alignment on the lidar point cloud data, and perform dimensionality reduction from 3 dimensions to 64 dimensions in the first multi-layer perception to extract local features,

[0110] Step S2003, generate a 64x64 affine transformation matrix in the second T-Net micro network to perform feature space alignment on the 64-dimensional local features, perform dimensionality reduction in the second multi-layer perception to 1024 dimensions, and aggregate the 1024-dimensional features in the maximum pooling layer to generate a global feature vector;

[0111] Step S2004, after the decoder receives the global feature vector, dimensionality reduction is performed in the third multi-layer perception to 2 dimensions, and nonlinear transformation is performed in the fourth multi-layer perception to fuse the 2-dimensional global feature vector and the preset number of days after rice transplanting to output the leaf age prediction value and the tiller number prediction value of the single rice plant;

[0112] Step S2005, repeat the above steps S2001 to S2004 until the structure feature extraction model is trained to a convergent state to determine the structure feature extraction model trained to a convergent state.

[0113] Specifically, the laser radar point cloud data corresponding to the preset number of days after rice transplanting of the single rice plant collected by the unmanned aerial vehicle can be used as the second training sample, and the leaf age and tiller number of the single rice plant can be used as the second sample label. The second training sample and the second sample label are one-to-one corresponding, and are used to construct the second sample data set. First, find the manually investigated single rice plant in each plot in the laser radar point cloud data, and manually label the plant center point of the single rice plant. Then, draw a circle with a radius of 20 cm to extract the point cloud within the circle. Finally, ground point filtering and outlier point filtering are performed on the point cloud sample to obtain pure rice plant point cloud data. All point cloud samples are reduced to 10,000 points for subsequent processing.

[0114] The basic network architecture of the structural feature extraction model of the present application is an end-to-end point cloud deep learning network, including an input layer, an encoder and a decoder. The laser radar point cloud data corresponding to the preset number of days after rice seedling transplanting of the single rice plant collected by the unmanned aerial vehicle is taken as the input of the structural feature extraction model, and the model outputs the tiller prediction and leaf age prediction values. The core idea of the structural feature extraction model of the present application is to solve the permutation invariance of the point cloud through the symmetric function-max pooling. In the encoding part, the encoder includes a first T-Net micro network, a first multi-layer perceptron (MLP), a second T-Net micro network, a second multi-layer perceptron (MLP) and a max pooling layer connected in sequence. The main role of the T-Net micro network is to align and normalize the point cloud data in the three-dimensional (3D) space, which is like straightening the skewed object for subsequent processing. The multi-layer perceptron (MLP) extracts features by dimensionality increasing, and the information of each point is increased from the original 3 dimensions (x, y, z) to 1024 dimensions to form a global feature vector.

[0115] The T-Net micro network can generate an affine transformation matrix to realize the spatial changes such as rotation and translation of the point cloud. It learns a k x k rotation matrix by mapping the point cloud to a k-dimensional space. Since the rotation matrix has orthogonality, a regularization penalty term needs to be introduced in the loss function to make it as close to an orthogonal matrix as possible. The expression of the regularization penalty term is:

[0116]

[0117] In the formula, represents the regularization penalty term; E is the unit matrix; A is the affine transformation matrix output by the T-Net micro network, and the Frobenius norm term is added in the loss function to make the output affine transformation matrix A as close to an orthogonal matrix as possible; represents the transpose matrix of the affine transformation matrix A; represents the Frobenius norm.

[0118] In the decoder part, the decoder includes a third multi-layer perceptron and a fourth multi-layer perceptron connected in sequence. The global features are reduced in dimension by two multi-layer perceptrons (MLPs). The third multi-layer perceptron reduces the global features to 2 dimensions, and the fourth multi-layer perceptron has an input of a time feature parameter, i.e. the number of days after rice seedling transplanting, which is a time parameter highly related to the tillering and leaf age of rice. It is fused with the extracted point cloud feature vector to jointly predict the tillering and leaf age, so the fourth multi-layer perceptron has 3-dimensional input and 2-dimensional output, which includes leaf age and tiller number.

[0119] In some embodiments, please refer to Figure 2 ,​Figure 2 This is an exemplary network architecture for a point cloud deep learning network, where Input Points represents the input LiDAR point cloud data; T-Net represents a T-Net micro-network; Transformationmatrix represents an affine transformation matrix; MLP represents a multilayer perceptron; Max pool represents a max pooling layer; Encode represents an encoder; Decoder represents a decoder; Tillers represent the number of tillers; and Leaf age represents the leaf age.

[0120] In some embodiments, please refer to Figure 3 , Figure 3 The horizontal axis represents the number of tillers predicted by the model, and the vertical axis represents the number of tillers measured in the field. The coefficient of determination (R²) between the predicted and measured tillers is 0.91, indicating a strong correlation between the two. The root mean square error (RMSE) is 1.60, reflecting the average error between the predicted and measured tillers, demonstrating that the point cloud deep learning network has good performance in tiller prediction.

[0121] Please see Figure 4 , Figure 4 The horizontal axis represents the leaf ages predicted by the model, and the vertical axis represents the field-measured leaf ages. The coefficient of determination (R²) between the model-predicted leaf ages and the field-measured leaf ages reaches 0.97, indicating a strong correlation between the two. The root mean square error (RMSE) is 0.43, reflecting the average error between the model-predicted leaf ages and the field-measured leaf ages, demonstrating that the point cloud deep learning network has good performance in leaf age prediction.

[0122] Please see Figure 5 , Figure 5 The horizontal axis represents the date (such as 0806, 0810, etc.), and the vertical axis represents the leaf ages. Figure 5 It shows the trend of leaf age change predicted by the model over time. By combining the distribution interval, we can intuitively see the concentration and fluctuation of leaf age prediction values ​​at different periods, which helps to analyze the prediction stability of the model in the process of dynamic changes in leaf age.

[0123] Please see Figure 6 , Figure 6 The horizontal axis represents the date, and the vertical axis represents the number of tillers. Figure 6The model predicted change trend of tiller number over time is presented, and the dispersion degree of tiller number prediction value in different periods can be observed through the distribution interval, so as to judge the prediction reliability of the model in the change process of tiller number over time.

[0124] In step S30, a first sample data set is obtained, wherein the first sample data set includes a plurality of first training samples and corresponding first sample labels, the first training samples represent spectral feature time series data constructed by spectral feature data collected from a single rice plant at a preset number of days after rice transplanting, and structure feature time series data constructed by structure feature data, and the first sample labels represent early growth vigor scores of the rice plants.

[0125] According to the laser radar point cloud data, a canopy height model of the single rice plant is generated to extract the plant height of the single rice plant. Based on the structure feature extraction model trained to a convergent state, laser radar point cloud data corresponding to a preset number of days after rice transplanting is used to determine the leaf age and tiller number of the single rice plant. After the plant height, the leaf age, and the tiller number are used to construct structure feature data of the single rice plant, a first sample data set is obtained, wherein the first sample data set includes a plurality of first training samples and corresponding first sample labels, the first training samples represent spectral feature time series data constructed by spectral feature data collected from a single rice plant at a preset number of days after rice transplanting, and structure feature time series data constructed by structure feature data, and the first sample labels represent early growth vigor scores of the rice plants.

[0126] In some embodiments, the step of determining the early growth vigor score includes:

[0127] In step S301, the number of days required for a target variety of rice to reach an effective tiller number, the maximum value of the number of days required for all varieties of rice to reach an effective tiller number, the minimum value of the number of days required for all varieties of rice to reach an effective tiller number, the number of days required for a target variety of rice to reach a time node at which the leaf age increases at the fastest speed, the maximum value of the number of days required for all varieties of rice to reach a time node at which the leaf age increases at the fastest speed, and the minimum value of the number of days required for all varieties of rice to reach a time node at which the leaf age increases at the fastest speed are obtained.

[0128] In step S302, a first difference value between the number of days required for the target variety of rice to reach an effective tiller number and the minimum value of the number of days required for all varieties of rice to reach an effective tiller number is calculated, a second difference value between the maximum value of the number of days required for all varieties of rice to reach an effective tiller number and the minimum value of the number of days required for all varieties of rice to reach an effective tiller number is calculated, and a first ratio value between the first difference value and the second difference value is calculated.

[0129] Step S303, calculate a third difference value between the minimum value of the days required for the target variety of rice to reach the time node of the fastest leaf age increase speed and the days required for all varieties of rice to reach the time node of the fastest leaf age increase speed, calculate a fourth difference value between the maximum value of the days required for all varieties of rice to reach the time node of the fastest leaf age increase speed and the minimum value of the days required for all varieties of rice to reach the time node of the fastest leaf age increase speed, and calculate a second ratio value between the third difference value and the fourth difference value;

[0130] Step S304, calculate a first sum value between the first ratio value and the second ratio value, calculate a third ratio value between the first sum value and a numerical value two, and determine the early growth vigor score according to the difference value between the numerical value one and the third ratio value.

[0131] Specifically, according to the characteristics of early growth vigor type of rice “early return green and early tillering”, two indexes are selected to represent the degree of early or late return green and tillering of the experimental rice varieties. It should be noted that the tillering of rice is divided into effective tillering and ineffective tillering, and the ineffective tillering cannot develop into a normal ear eventually, so the number of days required to reach the effective tillering number is investigated during manual annotation.

[0132] Please refer to Figure 7 and Figure 8 Taking the rice variety Heumeisimian as an example, the maximum effective tillering number at harvest is 7, and according to the manual field investigation results, the number of days required for the rice variety Heumeisimian to reach the effective tillering number and the number of days required to reach the time node of the fastest leaf age increase speed are calculated. Among them, Figure 7 the horizontal axis of the graph represents the number of days after transplanting (Days after transplanting), and the vertical axis represents the tiller number (Tillers), Figure 7 the graph shows the change of the tiller number of the rice variety Heumeisimian with time after transplanting, and the upward trend of the curve reflects the growth process of the tiller number. Among them, the largest dot mark corresponds to the number of days required for the rice variety to reach the effective tillering number, and by observing the curve, the time node when the tiller number grows to the effective number can be directly observed.

[0133] Figure 8 the horizontal axis of the graph represents the number of days after transplanting (Days after transplanting), and the vertical axis represents the leaf age increase / 4 days (Leafage increase / 4 days), Figure 8 the different columns in the graph represent the leaf age increase speed in different time periods after transplanting. The higher the column, the faster the leaf age increases in that time period. Among them, the time period corresponding to the higher column is the time node of the fastest leaf age increase speed, and by observing the graph, the time node of the fastest leaf age increase speed can be directly observed.Figure 8 It can clearly determine the number of days required to reach the point where the leaf age increases most rapidly.

[0134] Furthermore, the early growth activity scores of the target rice varieties were comprehensively calculated. Since the 300 varieties planted in this study basically cover a wide range of early growth activity from extremely weak to extremely strong, the early growth activity scores were mapped to a range of 0 to 1 using the maximum-minimum normalization method. Finally, the rice varieties... The formula for calculating the early growth viability score is expressed as follows:

[0135] ,

[0136] In the formula, Indicates rice varieties Early growth viability score; This indicates the number of days required for the target rice variety to reach the effective number of tillers; This represents the minimum number of days required for all rice varieties to reach the effective number of tillers. This represents the maximum number of days required for all rice varieties to reach the effective number of tillers. Indicates rice varieties The number of days required to reach the point where leaf age increases most rapidly; This represents the maximum number of days required for all rice varieties to reach the point where their leaf age increases at the fastest rate. This represents the minimum number of days required for all rice varieties to reach the point where their leaf age increases at its fastest rate.

[0137] Step S40: Input the current spectral feature data and current structural feature data of a single rice plant of the target variety into the early growth vigor prediction model trained using the first sample dataset, so as to predict the early growth vigor score of a single rice plant of the target variety at a preset number of days after rice transplanting, and complete the prediction of the early growth vigor of rice.

[0138] After obtaining the first sample data set, the current spectral feature data and the current structural feature data of the single rice plant of the target variety are input into the early growth vigor prediction model trained by using the first sample data set to predict the early growth vigor score of the single rice plant of the target variety at a preset number of days after rice transplanting, thereby completing the prediction of the early growth vigor of the rice.

[0139] Specifically, the long short-term memory network (LSTM) based on time attention mechanism is used as the basic network architecture of the early growth vigor prediction model of the present application. The long short-term memory network based on time attention mechanism includes an input layer, a double-branch LSTM module, a feature fusion module and a prediction layer connected in sequence. The double-branch LSTM module includes a first LSTM branch module (multi-spectral module) and a second LSTM branch module (laser radar point cloud module). The first LSTM branch module is used to process spectral feature time series data, and the second LSTM branch module is used to process structural feature time series data. The feature parameters extracted from the spectral feature time series data and the structural feature time series data are input into the feature fusion module in the long short-term memory network based on time attention mechanism. Finally, the feature parameters form a global feature vector through vector connection in the feature fusion module. The loss functions of the first LSTM branch module (multi-spectral module) and the second LSTM branch module (laser radar point cloud module) are fused by weighted gradient to realize the training of the early growth vigor prediction model. For each module , the loss function , the gradient of the feature parameters can be expressed as , and the gradient of the first LSTM branch module (multi-spectral module) and the second LSTM branch module (laser radar point cloud module) after fusion can be expressed as:

[0140] ,

[0141] wherein, represents the first LSTM branch module (multi-spectral module) or the second LSTM branch module (lidar point cloud module); represents the gradient of the objective function of the early growth vigor prediction model with respect to the feature parameters . represents the objective function of the early growth vigor prediction model; represents the total number of modules participating in gradient fusion; represents the weight of the first module; represents the objective function of the first module; represents the objective function of the first module with respect to the feature parameters . represents the gradient of the objective function of the first module with respect to the feature parameters

[0142] , i.e., the gradient information of the first

[0143] Further, the first LSTM branch module and the second LSTM branch module each include a plurality of LSTM processing units corresponding to time steps, each LSTM processing unit is connected with a time attention mechanism module, and each LSTM processing unit is connected with a time attention mechanism module, which can adaptively adjust the attention weight of each time step, so that the early growth vigor prediction model gives greater weight to key time steps, and the weight can be visualized to determine the weight relied on by the model during prediction. In some embodiments, the structural feature extraction model and the early growth vigor prediction model of the present application can be built using the Pytorch framework and trained on an NVIDIA RTX 4090 24G graphics memory. The structural feature extraction model and the early growth vigor prediction model in the present application are implemented by Pytorch, with stochastic gradient descent (SGD) as the optimizer and L2 regularization to prevent overfitting. For the final model prediction evaluation, the coefficient of determination (R2), the root mean square error (RMSE), and the relative root mean square error (RRMSE) are calculated on the test set, wherein the calculation formula of the coefficient of determination (R2) is represented as:

[0144] ,

[0145] The calculation formula of the root mean square error (RMSE) is represented as:

[0146] ,

[0147] The relative root mean square error (RRMSE) is calculated as: ​​The formula for calculating ) is expressed as:

[0148] ,

[0149] in, The coefficient of determination measures the goodness of fit between the model's predicted values ​​and the actual values. It ranges from 0 to 1, with values ​​closer to 1 indicating a better fit. The model represents the first Predicted values ​​for each sample; Represents actual value The average value; Indicates the first The actual value of each sample; Indicates the number of samples; Root mean square error (RMSE) measures the average error between the model's predicted value and the actual value. A smaller value indicates higher model prediction accuracy. Relative root mean square error is an indicator that relativizes the root mean square error. It is used to more intuitively reflect the proportion of the error relative to the average level of the actual value, and is presented in the form of a percentage.

[0150] The early growth vigor prediction model of this application is trained using the first sample dataset mentioned above. After the early growth vigor prediction model is trained to convergence, the early growth vigor score of a single rice plant of the target variety can be predicted based on the current spectral feature data and current structural feature data of the single rice plant of the target variety at a preset number of days after rice transplanting.

[0151] In a specific embodiment, the step of inputting the current spectral feature data and current structural feature data of a single rice plant of the target variety into an early growth vigor prediction model trained using the first sample dataset to predict the early growth vigor score of a single rice plant of the target variety at a preset number of days after rice transplanting includes:

[0152] Step S401: Input the current spectral feature time series data and the current structural feature time series data of the target variety of single rice plant into the input layer of the early growth vigor prediction model;

[0153] Step S402: In the first LSTM branch module, the LSTM processing unit at each time step performs time dynamic feature extraction on the spectral feature time series data, and in the second LSTM branch module, the LSTM processing unit at each time step performs time dynamic feature extraction on the structural feature time series data.

[0154] Step S403, in the time attention mechanism module corresponding to each time step, the weight of the spectral feature data or the structural feature data output by the corresponding LSTM processing unit is calculated to obtain the spectral feature weighted vector or the structural feature weighted vector corresponding to the LSTM processing unit of each time step;

[0155] Step S404, after the spectral feature data of all time steps is weighted and aggregated into a spectral feature global weighted vector by the respective corresponding time attention mechanism module, the structural feature data of all time steps is weighted and aggregated into a structural feature global weighted vector by the respective corresponding time attention mechanism module;

[0156] Step S405, in the feature fusion module, the spectral feature global weighted vector and the structural feature global weighted vector are spliced and fused to obtain a global feature vector, and in the prediction layer, the global feature vector is processed by a full connection layer and a Sigmoid activation function for nonlinear mapping, and an early growth vigor score of the target variety of single rice at a preset number of days after rice transplanting is output.

[0157] In some embodiments, referring to Figure 9 , Figure 9 is an exemplary network architecture of a long short-term memory network (LSTM) based on a time attention mechanism, wherein Input represents input, LSTM represents a long short-term memory network, MS Module represents a multi-spectral module, Dense represents a dense connection layer, Softmax represents a Softmax activation function, Feature Vector represents a feature vector, Full Connect represents a full connection layer, Sigmoid represents a Sigmoid activation function, Fusion Model represents a fusion model, LiDAR Module represents a laser radar point cloud module, Attention Mechanism represents a time attention mechanism module, Early growth vigor score (MS) represents an early growth vigor score output by a multi-spectral module: Early growth vigor score (MS+LiDAR) represents an early growth vigor score output by a fusion model constructed by a multi-spectral module and a laser radar point cloud module; Early growth vigor score (LiDAR) represents an early growth vigor score output by a laser radar point cloud module.

[0158] In some embodiments, referring to Figure 10 , Figure 10Figure 1 is a scatter plot of predicted values and true values of the fusion model of multispectral (MS) and LiDAR data, with the horizontal axis representing predicted values and the vertical axis representing true values. The figure is marked with , indicating that the predicted values of the fusion model have a strong correlation with the true values; the figure is marked with and , reflecting the small average error between predicted values and true values, and overall indicating that the fusion model of multispectral (MS) and LiDAR data performs well in early growth vigor prediction.

[0159] Referring to Figure 11 , Figure 11 Figure 2 is a histogram of prediction errors of the fusion model of multispectral (MS) and LiDAR data, with the horizontal axis representing errors and the vertical axis representing counts. The figure shows the distribution of sample numbers in different error intervals, and the errors overall show a trend of approximate normal distribution, with most errors concentrated between -0.2 and 0.2, indicating that the fusion model has high stability in prediction.

[0160] In some embodiments, the relationship between early growth vigor and final yield of rice shows a certain positive correlation between the two, but this relationship is not absolutely linear. Varieties with higher early growth vigor are more likely to achieve higher yield, however, not all high-vigor varieties can achieve high yield, and all high-yield varieties almost have higher early growth vigor. This result shows that early growth vigor is a necessary condition for high yield of rice, but not a sufficient condition.

[0161] Combining the scatter plot and the proportion distribution, it can be seen that most varieties are concentrated in the "low yield + low score" (50.9%) and "high yield + high score" (24.9%) quadrants, indicating that the growth vigor and yield trends of most varieties are consistent. Varieties in the "high yield + high score" quadrant show strong early growth ability and ultimately achieve high individual yield, which is consistent with the growth and physiological rules of rice, i.e. early growth advantage usually lays the foundation for high yield in later period. Varieties in the "low yield + low score" quadrant have weak growth potential during the entire growth period and are difficult to accumulate enough biomass, so the final yield is low.

[0162] Notably, some varieties fell into the "high yield + low score" (2.3%) and "low yield + high score" (22.0%) quadrants, and the performance of these varieties revealed the multidimensionality of rice yield formation. Among them, the "high yield + low score" varieties achieved higher yield despite lower early growth vigor scores, which may be due to their stronger ability to compensate for growth later, such as higher photosynthetic efficiency, stronger grain filling capacity, or better tiller-to-ear conversion rate. Conversely, the "low yield + high score" varieties had stronger early growth vigor but lower final yield, which may be affected by factors such as late tiller degradation, nutrient competition, limited grain number per spike, or environmental stress.

[0163] These results suggest that, in the process of rice variety breeding and precision management, focusing only on early growth vigor is not sufficient to fully predict yield, and the performance and physiological characteristics of later growth need to be combined. For example, the existence of some "high yield + low score" varieties indicates that, although early growth vigor is low, if growth disadvantage can be compensated for through later management (such as reasonable fertilization and improved filling efficiency), higher yield can still be achieved. Therefore, in the precision management strategy of rice, differentiated management schemes can be developed for different growth types of varieties, such as strengthening late tiller regulation and water and fertilizer management for "high score + low yield" varieties to improve final yield, and optimizing early growth regulation for "low score + high yield" varieties to promote better growth balance.

[0164] From the above embodiments, compared with the prior art, the present application addresses the problems in the prior art that the multi-spectral index can only indirectly reflect the group tiller trend, but cannot achieve precise analysis at the single plant scale, and the traditional gridding method causes loss of three-dimensional characteristics, and the segmentation network based on regular voxels is difficult to adapt to the topological heterogeneity of rice tillers, which leads to the current LiDAR technology not being able to achieve high-throughput analysis of tiller number. The present application includes but is not limited to the following beneficial effects:

[0165] Firstly, the rice early growth vigor prediction method of the present application addresses the problem of difficulty in extracting fine structure features such as tiller number and leaf age, designs an end-to-end point cloud deep learning network, realizes point cloud spatial alignment and standardization through a T-Net micro network, completes feature dimensionality and dimensionality reduction through a multi-layer perceptron, and integrates a time parameter, breaking through the bottleneck of single plant scale tiller number and leaf age extraction. The determination coefficient (R²) of the network for tiller number prediction reaches 0.91, and the R² for leaf age prediction reaches 0.97, solving the contradiction between "insufficient group inversion accuracy and low single plant measurement efficiency" of traditional methods, and achieving high-throughput precise analysis of key growth parameters. At the same time, the LSTM network with a two-level feature fusion structure is innovatively used to integrate multi-modal time series data, and the attention mechanism is used to quantify the importance of data at different growth stages, overcoming the defects of insufficient stability and explainability of traditional static phenotype analysis.

[0166] Secondly, the early growth vigor prediction method of the present application fuses the LiDAR point cloud deep learning features and the model of multispectral data, and the R² of the early growth vigor prediction reaches 0.79, and the RMSE is only 0.07, which is significantly better than the model using LiDAR (R²=0.72, RMSE=0.08) or multispectral data (R²=0.54, RMSE=0.11) alone. The three-dimensional structure parameters provided by LiDAR make up for the deficiency of spectral index in the three-dimensional characterization of the canopy, and the physiological parameters of multispectral data enhance the understanding of the biochemical process of the model, and the two are dynamically complementary at different growth stages, and the structure parameters contribute more within 12 days after transplanting, and the spectral index contribution is highlighted after the canopy is closed. Further optimization of the input parameter combination found that "tiller number + leaf age + plant height" fused with multispectral index has the highest prediction accuracy (R²=0.84), which fully verifies the complementary effect of multi-modal data.

[0167] Thirdly, the early growth vigor prediction method of the present application can visualize the time attention mechanism, accurately locate the period of 12 to 20 days after transplanting as the key period affecting the early growth vigor prediction, and the canopy phenotype data have the highest contribution weight at this stage, which shortens the window period of traditional experience judgment. It is found that the leaf age has the highest prediction value among single parameters, and the dynamic coordination of structure parameters and spectral index is the core influencing factor of early growth vigor. In addition, the relationship between early growth vigor and yield is clarified, and high-yield varieties generally have high early growth vigor, but high-vigor varieties are not necessarily high-yield, which confirms that early growth and fast development are necessary but not sufficient conditions for high yield, and the proportion of "high yield + high score" and "low yield + low score" varieties is more than 75%, which provides a new perspective for the study of yield formation mechanism.

[0168] Fourthly, the early growth vigor prediction method of the present application realizes the quantitative evaluation of early growth vigor of single rice plant in the field environment, breaks the traditional mode relying on manual judgment, and provides a standardized technical means for early growth vigor breeding of rice varieties. By comparing the early growth vigor scores of different rice varieties, data support is provided for variety adaptability evaluation. At the same time, the research clearly points out the management focus of different growth type varieties, such as strengthening the late regulation of "high score + low yield" varieties and optimizing the early management of "low score + high yield" varieties, to provide individualized solutions for precision agricultural management. In addition, the technical system constructed can be popularized to the field of crop phenotype research, and lay a solid foundation for the application of multi-modal remote sensing technology in agricultural intelligence.

[0169] Please refer to Figure 12, a rice early growth vigor prediction device provided for adapting to one of the purposes of the present application, comprising a spectral feature determination module 1100, a structure feature determination module 1200, a data set acquisition module 1300, and a growth vigor prediction module 1400. The spectral feature determination module 1100 is configured to acquire multi-spectral image data and laser radar point cloud data corresponding to a preset number of days after rice transplanting of a plurality of single rice plants of a target variety, and determine spectral feature data of the single rice plants according to the multi-spectral image data. The structure feature determination module 1200 is configured to generate a canopy height model of the single rice plants according to the laser radar point cloud data to extract plant heights of the single rice plants, determine leaf ages and tiller numbers of the single rice plants according to the laser radar point cloud data corresponding to the preset number of days after rice transplanting based on a structure feature extraction model trained to a converged state, and construct structure feature data of the single rice plants according to the plant heights, the leaf ages, and the tiller numbers. The data set acquisition module 1300 is configured to acquire a first sample data set, wherein the first sample data set includes a plurality of first training samples and corresponding first sample labels. The first training samples represent spectral feature time series data constructed by spectral feature data collected at a preset number of days after rice transplanting of single rice plants, and structure feature time series data constructed by structure feature data. The first sample labels represent early growth vigor scores of the rice. The growth vigor prediction module 1400 is configured to input current spectral feature data and current structure feature data of a single rice plant of a target variety into an early growth vigor prediction model trained by the first sample data set, to predict an early growth vigor score of the single rice plant of the target variety at a preset number of days after rice transplanting, and complete prediction of early growth vigor of the rice.

[0170] Based on any embodiment of the present application, please refer to Figure 13 Another embodiment of the present application further provides an electronic device, which can be implemented by a computer device. As shown in Figure 13 , a schematic diagram of an internal structure of the computer device. The computer device includes a processor, a computer readable storage medium, a memory, and a network interface connected by a system bus. The computer readable storage medium of the computer device stores an operating system, a database, and computer readable instructions. The database can store a control information sequence. When the computer readable instructions are executed by the processor, the processor can implement a rice early growth vigor prediction method. The processor of the computer device is used to provide computing and control capabilities to support the operation of the entire computer device. The memory of the computer device can store computer readable instructions. When the computer readable instructions are executed by the processor, the processor can execute the rice early growth vigor prediction method of the present application. The network interface of the computer device is used to connect and communicate with a terminal. Those skilled in the art can understand thatFigure 13 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0171] The processor in the embodiment is configured to execute the specific functions of each module in Figure 12 The memory stores the program codes and various data required for executing the above modules. The network interface is configured to transmit data between the user terminal and the server. The memory in the embodiment stores the program codes and data required for executing all the modules in the rice early growth vigor prediction device of the present application, and the server can call the program codes and data of the server to execute the functions of all the modules.

[0172] The present application also provides a storage medium storing computer readable instructions, which, when executed by one or more processors, cause the one or more processors to perform the steps of the rice early growth vigor prediction method described in any of the embodiments of the present application.

[0173] The present application also provides a computer program product, including computer programs / instructions, which, when executed by one or more processors, implement the steps of the rice early growth vigor prediction method described in any of the embodiments of the present application.

[0174] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiments of the present application can be completed by a computer program instructing related hardware. The computer program can be stored in a computer readable storage medium. When the program is executed, it can include the processes of the above-mentioned embodiments of each method. The storage medium can be a computer readable storage medium such as a magnetic disc, an optical disc, a read-only memory (ROM), or a random access memory (RAM).

[0175] The above only describes some embodiments of the present application. It should be noted that those of ordinary skill in the art can make several improvements and refinements without departing from the principles of the present application. These improvements and refinements should also be considered within the scope of the present application.

Claims

1. A method for predicting early growth vigor of rice, characterized in that, include: Multispectral image data and lidar point cloud data of multiple individual rice plants of the target variety at a preset number of days after rice transplanting are obtained. The spectral feature data of the individual rice plants are determined based on the multispectral image data. The spectral feature data includes normalized vegetation index, chlorophyll index, green normalized vegetation index, soil-optimized vegetation index, normalized red edge difference index, and canopy coverage. The canopy height model of a single rice plant is generated based on the lidar point cloud data to extract the plant height. A structural feature extraction model, trained to convergence, is used to determine the leaf age and tiller number of a single rice plant based on lidar point cloud data corresponding to a preset number of days after rice transplanting. Structural feature data of the single rice plant is constructed based on the plant height, leaf age, and tiller number. The basic network architecture of the structural feature extraction model is a point cloud deep learning network, which includes an input layer, an encoder, and a decoder. The encoder includes two T-Net micronetworks and two multilayer perceptrons. The encoder comprises a first T-Net micronetwork, a first multilayer perceptron, a second T-Net micronetwork, a second multilayer perceptron, and a max-pooling layer connected sequentially. The decoder comprises a third multilayer perceptron and a fourth multilayer perceptron connected sequentially. Obtain a first sample dataset, wherein the first sample dataset includes multiple first training samples and their corresponding first sample labels. The first training samples represent spectral feature time series data constructed from spectral feature data collected from a single rice plant at a preset number of days after rice transplanting, and structural feature time series data constructed from structural feature data. The first sample labels represent the early growth vigor score of the rice. The step of determining the early growth vigor score includes: The goal is to obtain the number of days required for the target rice variety to reach the effective number of tillers, the maximum number of days required for all rice varieties to reach the effective number of tillers, the minimum number of days required for all rice varieties to reach the effective number of tillers, the number of days required for the target rice variety to reach the time node with the fastest leaf age increase, the maximum number of days required for all rice varieties to reach the time node with the fastest leaf age increase, and the minimum number of days required for all rice varieties to reach the time node with the fastest leaf age increase. Calculate a first difference between the number of days required for the target rice variety to reach the effective number of tillers and the minimum number of days required for all rice varieties to reach the effective number of tillers; calculate a second difference between the maximum number of days required for all rice varieties to reach the effective number of tillers and the minimum number of days required for all rice varieties to reach the effective number of tillers; and calculate a first ratio between the first difference and the second difference. Calculate the third difference between the number of days required for the target rice variety to reach the fastest leaf age increase point and the minimum number of days required for all rice varieties to reach the fastest leaf age increase point; calculate the fourth difference between the maximum number of days required for all rice varieties to reach the fastest leaf age increase point and the minimum number of days required for all rice varieties to reach the fastest leaf age increase point; and calculate the second ratio between the third difference and the fourth difference. Calculate and determine a first sum between the first ratio and the second ratio, calculate and determine a third ratio between the first sum and the second value, and determine the early growth vitality score based on the difference between the first value and the third ratio; The current spectral and structural feature data of a single rice plant of the target variety are input into an early growth vigor prediction model trained using the first sample dataset to predict the early growth vigor score of a single rice plant of the target variety at a preset number of days after rice transplanting, thus completing the prediction of early growth vigor of rice. The basic network architecture of the early growth vigor prediction model is a Long Short-Term Memory network based on a time attention mechanism. The Long Short-Term Memory network based on the time attention mechanism includes an input layer, a dual-branch LSTM module, a feature fusion module, and a prediction layer connected in sequence. The dual-branch LSTM module includes a first LSTM branch module and a second LSTM branch module. The first LSTM branch module is used to process spectral feature time series data, and the second LSTM branch module is used to process structural feature time series data. The first LSTM branch module and the second LSTM branch module each include multiple LSTM processing units corresponding to multiple time steps, and each LSTM processing unit corresponding to a time step is connected to a time attention mechanism module.

2. The method for predicting early growth vigor of rice according to claim 1, characterized in that, The steps for training a structural feature extraction model include: Obtain a second sample dataset, wherein the second sample dataset includes multiple second training samples and their corresponding second sample labels. The second training samples represent the lidar point cloud data of a single rice plant corresponding to a preset number of days after rice transplanting. The second sample labels represent the leaf age and tiller number of a single rice plant. The multiple second training samples and their corresponding second sample labels are input into a preset structural feature extraction model. A 3×3 affine transformation matrix is ​​generated in the first T-Net micro-network to perform three-dimensional spatial alignment on the lidar point cloud data. In the first multilayer perceptron, the three-dimensionally aligned lidar point cloud data is upsized from 3D to 64D to extract local features. In the second T-Net micro-network, a 64×64 affine transformation matrix is ​​generated to align the 64-dimensional local features in the feature space. In the second multilayer perceptron, the features after feature space alignment are increased to 1024 dimensions. In the max pooling layer, the 1024-dimensional features are aggregated to generate a global feature vector. After receiving the global feature vector, the decoder reduces the 1024-dimensional global feature vector to 2-dimensional in the third multilayer perceptron, and then fuses the 2-dimensional global feature vector with the preset number of days after rice transplanting in the fourth multilayer perceptron to perform a nonlinear transformation, so as to output the predicted leaf age and tiller number of the single rice plant. Repeat the above steps until the structural feature extraction model is trained to a convergent state, in order to determine the structural feature extraction model that has been trained to a convergent state.

3. The method for predicting early growth vigor of rice according to claim 1, characterized in that, The steps of inputting the current spectral and structural feature data of a single rice plant of the target variety into an early growth vigor prediction model trained using the first sample dataset to predict the early growth vigor score of a single rice plant of the target variety at a preset number of days after rice transplanting include: The current spectral feature time series data and current structural feature time series data of a single rice plant of the target variety are input into the input layer of the early growth vigor prediction model; In the first LSTM branch module, the LSTM processing unit at each time step performs time dynamic feature extraction on the spectral feature time series data, and in the second LSTM branch module, the LSTM processing unit at each time step performs time dynamic feature extraction on the structural feature time series data. In the temporal attention mechanism module corresponding to each time step, the weights of the spectral feature data or structural feature data output by the corresponding LSTM processing unit are calculated to obtain the spectral feature weighted vector or structural feature weighted vector corresponding to the LSTM processing unit at each time step. The spectral feature data of all time steps are weighted and aggregated into a global weighted vector of spectral features after the weights are calculated by their respective time attention mechanism modules. The structural feature data of all time steps are weighted and aggregated into a global weighted vector of structural features after the weights are calculated by their respective time attention mechanism modules. In the feature fusion module, the global weighted vector of the spectral features and the global weighted vector of the structural features are concatenated and fused to obtain a global feature vector. In the prediction layer, the global feature vector is nonlinearly mapped by a fully connected layer and a Sigmoid activation function to output the early growth vitality score of a single rice plant of the target variety at a preset number of days after rice transplanting.

4. The method for predicting early growth vigor of rice according to claim 1, characterized in that, The step of generating a canopy height model of a single rice plant based on the lidar point cloud data to extract the plant height of the single rice plant includes: Acquire lidar point cloud data of a single rice plant of the target variety at a preset number of days after rice transplanting; A digital surface model and a digital elevation model of a single rice plant are generated based on the lidar point cloud data. The canopy height model of the single rice plant is determined based on the difference between the digital surface model and the digital elevation model. The plant height of a single rice plant is extracted based on the canopy height model of the single rice plant.

5. The method for predicting early growth vigor of rice according to claim 1, characterized in that, The step of determining the spectral characteristic data of the single rice plant based on the multispectral image data includes: Acquire multispectral image data of a single rice plant of the target variety at a preset number of days after rice transplanting. Preprocess the multispectral image data to determine the multispectral reflectance of the single rice plant at the preset number of days after rice transplanting. The multispectral reflectance includes green band reflectance, red band reflectance, red edge band reflectance, and near-infrared band reflectance. The spectral characteristic data of a single rice plant corresponding to a preset number of days after rice transplanting are determined based on the multispectral reflectance.

6. A device for predicting early growth vigor of rice, characterized in that, include: The spectral feature determination module is configured to acquire multispectral image data and lidar point cloud data of multiple individual rice plants of the target variety at a preset number of days after rice transplanting, and determine the spectral feature data of the individual rice plants based on the multispectral image data. The spectral feature data includes normalized vegetation index, chlorophyll index, green normalized vegetation index, soil-optimized vegetation index, normalized red edge difference index, and canopy coverage. The structural feature determination module is configured to generate a canopy height model of a single rice plant based on the lidar point cloud data to extract the plant height of the single rice plant. Based on a structural feature extraction model trained to convergence, it determines the leaf age and tiller number of the single rice plant using lidar point cloud data corresponding to a preset number of days after rice transplanting. The module then constructs structural feature data of the single rice plant based on the plant height, leaf age, and tiller number. The basic network architecture of the structural feature extraction model is a point cloud deep learning network, which includes an input layer, an encoder, and a decoder. The encoder includes two T-Net micronetworks and two multilayer perceptrons. The encoder comprises a first T-Net micronetwork, a first multilayer perceptron, a second T-Net micronetwork, a second multilayer perceptron, and a max-pooling layer connected sequentially. The decoder comprises a third multilayer perceptron and a fourth multilayer perceptron connected sequentially. The dataset acquisition module is configured to acquire a first sample dataset, wherein the first sample dataset includes multiple first training samples and their corresponding first sample labels. The first training samples represent spectral feature time series data constructed from spectral feature data collected from a single rice plant at a preset number of days after rice transplanting, and structural feature time series data constructed from structural feature data. The first sample labels represent the early growth vigor score of the rice. The step of determining the early growth vigor score includes: The goal is to obtain the number of days required for the target rice variety to reach the effective number of tillers, the maximum number of days required for all rice varieties to reach the effective number of tillers, the minimum number of days required for all rice varieties to reach the effective number of tillers, the number of days required for the target rice variety to reach the time node with the fastest leaf age increase, the maximum number of days required for all rice varieties to reach the time node with the fastest leaf age increase, and the minimum number of days required for all rice varieties to reach the time node with the fastest leaf age increase. Calculate a first difference between the number of days required for the target rice variety to reach the effective number of tillers and the minimum number of days required for all rice varieties to reach the effective number of tillers; calculate a second difference between the maximum number of days required for all rice varieties to reach the effective number of tillers and the minimum number of days required for all rice varieties to reach the effective number of tillers; and calculate a first ratio between the first difference and the second difference. Calculate the third difference between the number of days required for the target rice variety to reach the fastest leaf age increase point and the minimum number of days required for all rice varieties to reach the fastest leaf age increase point; calculate the fourth difference between the maximum number of days required for all rice varieties to reach the fastest leaf age increase point and the minimum number of days required for all rice varieties to reach the fastest leaf age increase point; and calculate the second ratio between the third difference and the fourth difference. Calculate and determine a first sum between the first ratio and the second ratio, calculate and determine a third ratio between the first sum and the second value, and determine the early growth vitality score based on the difference between the first value and the third ratio; The growth vigor prediction module is configured to input the current spectral feature data and current structural feature data of a single rice plant of the target variety into an early growth vigor prediction model trained using the first sample dataset, so as to predict the early growth vigor score of a single rice plant of the target variety at a preset number of days after rice transplanting, thus completing the prediction of the early growth vigor of rice. The basic network architecture of the early growth vigor prediction model is a long short-term memory network based on a time attention mechanism. The long short-term memory network based on the time attention mechanism includes an input layer, a dual-branch LSTM module, a feature fusion module, and a prediction layer connected in sequence. The dual-branch LSTM module includes a first LSTM branch module and a second LSTM branch module. The first LSTM branch module is used to process spectral feature time series data, and the second LSTM branch module is used to process structural feature time series data. The first LSTM branch module and the second LSTM branch module each include multiple LSTM processing units corresponding to multiple time steps, and each LSTM processing unit corresponding to a time step is connected to a time attention mechanism module.

7. An electronic device comprising a central processing unit and a memory, characterized in that, The central processing unit is used to invoke and run a computer program stored in the memory to perform the steps of the method as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, It stores, in the form of computer-readable instructions, a computer program implemented according to any one of claims 1 to 5, which, when invoked by a computer, executes the steps included in the corresponding method.

Images