A method, device, medium and product for predicting photosynthetic parameters in field

Abstract

The present application discloses a method, device, medium and product for predicting photosynthetic parameters of a field, and relates to the field of intelligent agricultural monitoring. The method comprises obtaining visible light images and multispectral images of a set area of ​​a field to be predicted; determining the photosynthetic parameters of the set area of ​​a field to be predicted by using a field photosynthetic parameter prediction model according to the visible light images and the multispectral images; wherein the field photosynthetic parameter prediction model is obtained by training an improved convolutional neural network using a training data set combined with a gradient normalization algorithm; the training data set includes visible light images and multispectral images of a field for training and corresponding photosynthetic parameters; the improved convolutional neural network is obtained by replacing the convolutional layer of the residual module in the convolutional neural network with a depth-separable convolutional layer, and the improved convolutional neural network is multi-output. The present application improves the efficiency and accuracy of photosynthetic parameter prediction.

Description

Technical Field

[0001] The present application relates to the field of intelligent agricultural monitoring, and in particular to a method, equipment, medium and product for predicting photosynthetic parameters in a field. Background Art

[0002] Photosynthesis is an important process for crops to absorb solar energy and convert CO2 and water into organic matter, providing materials and energy for crop growth. Photosynthetic parameters are important indicators reflecting the ability of photosynthesis. Therefore, inversion of photosynthetic parameters is of great significance for understanding crop growth conditions and predicting yield and quality. Stomatal conductance and intercellular CO2 concentration are two important photosynthetic parameters that reflect the ability of gas exchange and are closely related to the physiological state of crops. They can be used to monitor the growth of field wheat.

[0003] In order to obtain the true value of photosynthetic parameters in the field, a portable photosynthesis measurement system Li-6400 or Li-6800 is usually used. However, this device is only used to measure the photosynthetic parameters at a certain point. If you want to measure a larger area, it is time-consuming, labor-intensive and inefficient. Therefore, developing a method to predict crop photosynthetic parameters in a low-cost automated manner can enable breeders and researchers to make data-driven decisions and intervene in the growth process. Given the importance of photosynthetic parameters, new methods for estimating photosynthetic parameters are constantly being developed.

[0004] With the development of UAV technology, its advantages such as simple operation, flexible flight mode and high resolution make its application research in the acquisition of photosynthetic parameters of field crops more and more extensive. Many current studies are based on feature selection and traditional machine learning models for modeling. Although they do not require too much computing resources and storage space, they usually require artificial selection and design of features. This process may require some professional knowledge and may not capture the most critical feature information, thus failing to accurately predict photosynthetic parameters. In addition, feature selection is also used for feature screening, which is time-consuming and labor-intensive. For different data sets, the selected feature sets are also different, and there are limitations in the later use and development of the model. These problems can be solved by selecting a deep learning model with strong feature extraction capabilities and trying to use the original remote sensing images as the input of the model to achieve "end-to-end" photosynthetic parameter prediction.

[0005] Multi-source data fusion can provide extensive information and achieve feature complementarity. In addition, most of the existing photosynthetic parameter prediction models predict a single photosynthetic parameter separately, which has the problem of low prediction efficiency. Summary of the invention

[0006] The purpose of this application is to provide a method, device, medium and product for predicting photosynthetic parameters in the field to improve the accuracy and efficiency of predicting photosynthetic parameters in the field.

[0007] To achieve the above objectives, this application provides the following solutions:

[0008] In a first aspect, the present application provides a method for predicting photosynthetic parameters in a field, comprising:

[0009] Obtain visible light images and multispectral images of a set area of ​​a field to be predicted;

[0010] According to the visible light image and the multispectral image, the photosynthetic parameters of the set area of ​​the field to be predicted are determined using a field photosynthetic parameter prediction model; the photosynthetic parameters include stomatal conductance and intercellular CO2 concentration; wherein the field photosynthetic parameter prediction model is obtained by training an improved convolutional neural network using a training data set in combination with a gradient normalization algorithm; the training data set includes visible light images and multispectral images of the training field and corresponding photosynthetic parameters; the improved convolutional neural network is obtained by replacing the convolutional layer of the residual module in the convolutional neural network with a depth-separable convolutional layer, and the improved convolutional neural network has multiple outputs.

[0011] Optionally, the improved convolutional neural network is trained using the training data set, specifically including:

[0012] Obtain training data sets and build improved convolutional neural networks;

[0013] Preprocessing the training data set to obtain a preprocessed training data set; the preprocessed training data set includes a plurality of groups of processed visible light images and processed multispectral images and corresponding standardized photosynthetic parameters;

[0014] The processed visible light image and the processed multispectral image are used as input, and the corresponding standardized photosynthetic parameters are used as output, and the improved convolutional neural network is trained to obtain the field photosynthetic parameter prediction model.

[0015] Optionally, obtain a training data set, specifically including:

[0016] Use DJI Phantom 4 drones to collect visible light and multispectral images of the training fields;

[0017] The photosynthetic parameters of each set area in the training field were collected using the LI-6400 photosynthetic meter.

[0018] Optionally, preprocessing the training data set to obtain a preprocessed training data set specifically includes:

[0019] The visible light image and the multispectral image are stitched and radiometrically corrected respectively to obtain a corrected visible light image and a corrected multispectral image;

[0020] The corrected visible light image is cropped according to the set area of ​​the training field to obtain the corrected visible light image of each set area;

[0021] The corrected multispectral image is cropped according to the set area of ​​the training field to obtain the corrected multispectral image of each set area;

[0022] The corrected visible light image of each set area is randomly scaled, randomly cropped, horizontally flipped, and randomly rotated to obtain a processed visible light image;

[0023] The corrected multispectral image of each set area is randomly scaled, randomly cropped, horizontally flipped and randomly rotated to obtain a processed multispectral image;

[0024] The photosynthetic parameters were normalized by Z-Score to obtain the standardized photosynthetic parameters.

[0025] Optionally, the improved convolutional neural network is trained with the processed visible light image and the processed multispectral image as input and the corresponding standardized photosynthetic parameters as output to obtain the field photosynthetic parameter prediction model, specifically comprising:

[0026] The processed visible light image and the processed multispectral image are input into the current improved convolutional neural network to obtain the predicted values ​​of photosynthetic parameters;

[0027] Determine a weighted loss function value by combining a weighted loss function with a gradient normalization algorithm according to the predicted value of the photosynthetic parameter and the corresponding standardized photosynthetic parameter;

[0028] Determine whether the set number of iterations has been reached;

[0029] If so, the trained improved convolutional neural network corresponding to the weighted loss function value within the set range is used as the field photosynthetic parameter prediction model;

[0030] If not, adjust the model parameters of the current improved convolutional neural network, and return "input the processed visible light image and the processed multispectral image into the current improved convolutional neural network to obtain the predicted values ​​of photosynthetic parameters".

[0031] Optionally, the weighted loss function is:

[0032] ;

[0033] in, is the weighted loss function value; is the stomatal conductance loss function value; is the weight of the stomatal conductance loss function value; is the intercellular CO2 concentration loss function value; is the weight of the intercellular CO2 concentration loss function value.

[0034] Optionally, the improved convolutional neural network includes a second fully connected layer, a third fully connected layer, and a first convolutional layer, a maximum pooling layer, a second convolutional layer, a third convolutional layer, a fourth convolutional layer, a fifth convolutional layer, a global average pooling layer, a Flatten function layer and a first fully connected layer connected in sequence; the first fully connected layer is connected to the second fully connected layer and the third fully connected layer respectively; the second convolutional layer, the third convolutional layer, the fourth convolutional layer and the fifth convolutional layer each include two residual modules connected in sequence.

[0035] In a second aspect, the present application provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement any of the above-described methods for predicting photosynthetic parameters in a field.

[0036] In a third aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the above-described methods for predicting photosynthetic parameters in a field.

[0037] In a fourth aspect, the present application provides a computer program product, including a computer program, which, when executed by a processor, implements any of the above-mentioned methods for predicting field photosynthetic parameters.

[0038] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0039] The present application provides a method, device, medium and product for predicting photosynthetic parameters of a field, by obtaining visible light images and multispectral images of a set area of ​​a field to be predicted; according to the visible light images and multispectral images, using a field photosynthetic parameter prediction model, the photosynthetic parameters of the set area of ​​the field to be predicted are determined; wherein the field photosynthetic parameter prediction model is obtained by training an improved convolutional neural network using a training data set in combination with a gradient normalization algorithm; the training data set includes visible light images and multispectral images of the field for training and the corresponding photosynthetic parameters; the improved convolutional neural network is obtained by replacing the convolutional layer of the residual module in the convolutional neural network with a depth-separable convolutional layer, and the improved convolutional neural network is multi-output. The present application uses multi-source remote sensing data as input, and predicts two photosynthetic parameters end-to-end based on a multi-task deep learning model to simultaneously predict two photosynthetic parameters, solving the problems of time-consuming, labor-intensive, low efficiency and difficult development of manually measuring photosynthetic parameters and predicting photosynthetic parameters with a classical machine learning model, and the present application improves the efficiency and accuracy of photosynthetic parameter prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0041] Figure 1 A schematic diagram of a process for predicting field photosynthetic parameters provided in one embodiment of the present application;

[0042] Figure 2 This is the network architecture diagram of the multi-task deep learning model;

[0043] Figure 3 A scatter plot of the predicted and true values ​​of stomatal conductance determined based on the multi-task deep learning model;

[0044] Figure 4 A scatter plot of the predicted and true values ​​of intercellular CO2 concentration determined based on the multi-task deep learning model;

[0045] Figure 5 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0046] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0047] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0048] This application takes into account the close relationship between photosynthetic parameters and establishes a model for simultaneously predicting stomatal conductance and intercellular CO2 concentration.

[0049] In an exemplary embodiment, Figure 1 As shown, a method for predicting photosynthetic parameters in a field is provided, comprising the following steps:

[0050] S1: Obtain visible light images and multispectral images of a set area of ​​a field to be predicted.

[0051] S2: According to the visible light image and the multispectral image, the photosynthetic parameters of the set area of ​​the field to be predicted are determined using a field photosynthetic parameter prediction model; the photosynthetic parameters include stomatal conductance and intercellular CO2 concentration; wherein the field photosynthetic parameter prediction model is obtained by training an improved convolutional neural network using a training data set combined with a gradient normalization algorithm; the training data set includes visible light images and multispectral images of the training field and the corresponding photosynthetic parameters; the improved convolutional neural network is obtained by replacing the convolutional layer of the residual module in the convolutional neural network with a depth-separable convolutional layer, and the improved convolutional neural network has multiple outputs.

[0052] As an optional implementation, the improved convolutional neural network is trained using a training data set, specifically including:

[0053] Step 1: Obtain the training dataset and build an improved convolutional neural network.

[0054] As an optional implementation, obtaining a training data set specifically includes:

[0055] The DJI Phantom 4 drone was used to collect visible light and multispectral images of the training field.

[0056] The photosynthetic parameters of each set area in the training field were collected using the LI-6400 photosynthetic meter.

[0057] In practical applications, the test area (training field) in this application is located in Wuqiao County, Cangzhou City, Hebei Province. The crop planted is winter wheat. The sowing time is October 2022 and the harvest time is June 2023. A total of 18 wheat varieties were planted, and three water treatments (W0, W1, W2) were designed for each wheat variety, and three repetitions were set for each water treatment. All three water treatments were irrigated with 75 mm before sowing, W1 was re-irrigated with 75 mm during the jointing stage, and W2 was re-irrigated with 75 mm during the jointing stage and flowering stage. The experimental area is divided into 54 rectangular areas, each of which is 2m×20m in size, including 3 repetitions of the same variety and the same water treatment, so a total of 162 plots.

[0058] The image acquisition platform is the DJI Phantom 4, which is equipped with a visible light sensor and five monochrome sensors: blue, green, red, red edge, and near infrared, to facilitate the acquisition of multi-sensor data. The flight mission was carried out from 10:00 to 12:00 on May 8, 2023 under clear and light cloud conditions. The flight altitudes were set to 20m, 30m, and 40m, respectively, and the heading and side overlap rates were set to 80%. Before each flight, radiation calibration plates with three reflectivity of 25%, 50%, and 75% were photographed for radiation calibration.

[0059] On the same day when the drone collected data, the LI-6400 photosynthetic instrument was used to collect the true values ​​of photosynthetic parameters of wheat in each rectangular area: stomatal conductance and intercellular CO2 concentration. Three plants were randomly selected from each plot, and the average value was taken as the true value of the plot.

[0060] The structure of the constructed multi-task deep learning model (improved convolutional neural network) is as follows Figure 2 As shown, the improved convolutional neural network includes a second fully connected layer, a third fully connected layer, and a first convolutional layer, a maximum pooling layer, a second convolutional layer, a third convolutional layer, a fourth convolutional layer, a fifth convolutional layer, a global average pooling layer, a Flatten function layer and a first fully connected layer connected in sequence; the first fully connected layer is connected to the second fully connected layer and the third fully connected layer respectively; the second convolutional layer, the third convolutional layer, the fourth convolutional layer and the fifth convolutional layer each include two residual modules connected in sequence.

[0061] Based on the convolutional neural network ResNet, the model is designed to take into account data fusion and simultaneous prediction of multiple parameters. The input of the model is two data sources, namely visible light images and multispectral images. The output of the model is two branches, namely stomatal conductance and intercellular CO2 concentration, two photosynthetic parameters. As the output branches of the model increase, the number of parameters also increases. Therefore, the convolutional layer in the ResNet residual module is replaced by a depthwise separable convolution to reduce the number of parameters and speed up the training. The loss function of the model is designed to be the weighted value of the loss values ​​of the two photosynthetic parameters. The gradient normalization algorithm (GradNorm) is introduced to realize multi-task adaptive optimization of loss weights and dynamically update the loss value weights.

[0062] The weighted loss function is:

[0063] .

[0064] in, is the weighted loss function value; is the stomatal conductance loss function value; is the weight of the stomatal conductance loss function value; is the intercellular CO2 concentration loss function value; is the weight of the intercellular CO2 concentration loss function value.

[0065] Among them, w (including and ) is difficult to set to a fixed value, so a gradient normalization algorithm (GradNorm) for deep multi-task networks is introduced. This algorithm automatically balances the training of multi-task deep learning models by dynamically adjusting the gradient size, avoiding one task dominating the model and ignoring other tasks, and realizes multi-task adaptive optimization of loss weights and dynamic update of w.

[0066] Step 2: preprocessing the training data set to obtain a preprocessed training data set; the preprocessed training data set includes multiple groups of processed visible light images and processed multispectral images and corresponding standardized photosynthetic parameters.

[0067] As an optional implementation, step 2 specifically includes:

[0068] The visible light image and multispectral image are stitched and radiometrically corrected to obtain the corrected visible light image and the corrected multispectral image. In practical applications, the DJI Zhitu software is used to stitch and radiometrically correct the images to obtain visible light images and single-band orthophotos (multispectral images) at different flight altitudes.

[0069] The corrected visible light image is cropped according to the set area of ​​the training field to obtain a corrected visible light image of each set area.

[0070] The corrected multispectral image is cropped according to the set area of ​​the training field to obtain the corrected multispectral image of each set area.

[0071] The corrected visible light image of each set area is randomly scaled, randomly cropped, horizontally flipped and randomly rotated to obtain a processed visible light image.

[0072] The corrected multispectral image of each set area is randomly scaled, randomly cropped, horizontally flipped and randomly rotated to obtain a processed multispectral image.

[0073] The photosynthetic parameters were normalized by Z-Score to obtain the standardized photosynthetic parameters.

[0074] In practical applications, the input of the multi-task deep learning model is a two-dimensional image. The stitched complete remote sensing image is cropped according to the small area to obtain the visible light image and the multispectral image as input. The online data enhancement method of random scaling and random cropping and the offline data enhancement method of horizontal flipping and random rotation are used to alleviate the overfitting phenomenon in the training process.

[0075] The two photosynthetic parameters, stomatal conductance and intercellular CO2 concentration, have different dimensions. Therefore, the Z-Score standardization method was used to make the two parameters in the same order of magnitude, which is suitable for comprehensive comparative evaluation.

[0076] Step 3: Taking the processed visible light image and the processed multispectral image as input and the corresponding standardized photosynthetic parameters as output, the improved convolutional neural network is trained to obtain the field photosynthetic parameter prediction model.

[0077] As an optional implementation, step 3 specifically includes:

[0078] The processed visible light image and the processed multispectral image are input into the current improved convolutional neural network to obtain the predicted values ​​of photosynthetic parameters.

[0079] According to the predicted value of the photosynthetic parameter and the corresponding standardized photosynthetic parameter, a weighted loss function is combined with a gradient normalization algorithm to determine a weighted loss function value.

[0080] Determine whether the set number of iterations has been reached.

[0081] If yes, the improved convolutional neural network after training corresponding to the weighted loss function value in the set range is used as the field photosynthetic parameter prediction model. In practical applications, the best accuracy model in the stable interval of the loss value change curve is selected as the field photosynthetic parameter prediction model.

[0082] If not, adjust the model parameters of the current improved convolutional neural network, and return "input the processed visible light image and the processed multispectral image into the current improved convolutional neural network to obtain the predicted values ​​of photosynthetic parameters".

[0083] The cropped visible light image and the multispectral image are used as input, and the predicted value of the model is compared with the true value. 2 And RMSE are used to evaluate the prediction effect.

[0084] .

[0085] .

[0086] in, is the true value of photosynthetic parameters, is the average value of the true value of photosynthetic parameters, is the predicted value of photosynthetic parameters; n is the number of plots.

[0087] In order to verify the effectiveness of data fusion and multi-task learning, four different network architecture models (single-input single-output model, multi-input single-output model, single-input multi-output model, and multi-input multi-output model) were trained for comparison. The trained prediction model was tested using the test data set, and the results are shown in Table 1.

[0088] In order to verify the effectiveness of data fusion, a multi-input single-output model was constructed and compared with the best single-input single-output. Visible light and multi-spectrum were selected as data sources to train stomatal conductance and intercellular CO2 prediction models respectively. For stomatal conductance, compared with the best single-input single-output, the multi-input single-output model has a better performance at 20m (R 2 =0.768), 30m (R2 =0.785) and 40m (R 2 =0.761) performed better, with prediction accuracy exceeding 0.086, 0.053 and 0.163; for intercellular CO2 concentration, compared with the best single-input single-output model, the multi-input single-output model performed better at 20m (R 2 =0.725), 30m (R 2 =0.768) and 40m (R 2 =0.776) all performed better, with prediction accuracies exceeding 0.097, 0.057, and 0.082.

[0089] In order to verify the effectiveness of multi-task learning, this example constructs a single-input multi-output to simultaneously predict stomatal conductance and intercellular CO2 concentration, and compares it with the best single-input single-output. The results show that the single-input multi-output has high accuracy for both stomatal conductance and intercellular CO2 concentration. At a flight altitude of 20m, R 2 The values ​​are 0.719 and 0.677, respectively, which are 0.037 and 0.049 higher than the best single-input single-output. At a flight altitude of 30m, R 2 The values ​​are 0.751 and 0.759, respectively, which are 0.019 and 0.048 higher than the best single-input single-output. At a flight altitude of 40m, R 2 The values ​​are 0.728 and 0.699 respectively, which are 0.13 and 0.005 higher than the best single-input single-output, proving the feasibility of multi-task learning.

[0090] Combining the advantages of data fusion and multi-task learning, the multi-task deep learning model constructed by the present invention, i.e., the multi-input multi-output model, realizes the efficient prediction of photosynthetic parameters. The method has a high accuracy rate for both stomatal conductance and intercellular CO2 concentration. At a flight altitude of 20m, R 2 (RMSE) are 0.771 (0.064) and 0.747 (19.446) respectively; at a flight altitude of 30m, R 2 (RMSE) are 0.814 (0.058) and 0.783 (18.035) respectively; at a flight altitude of 40m, R 2 The RMSE is 0.809 (0.059) and 0.777 (18.265) respectively, which can achieve the same or even better prediction accuracy as multi-input single output. It not only significantly improves the accuracy of the model, but also optimizes the model efficiency, that is, only one forward propagation is needed to simultaneously predict two photosynthetic parameters. In addition, the experimental results show that this method is applicable to a variety of flight altitudes. In this embodiment, the prediction result at an altitude of 30m is the best. The predicted values ​​of stomatal conductance and intercellular CO2 concentration are visualized with the true values, such as Figure 3 and Figure 4 shown.

[0091] Table 1 Comparison of prediction performance of stomatal conductance and intercellular CO2 concentration of different models

[0092]

[0093] Ablation test.

[0094] Ablation experiments were conducted on the depthwise separable convolution and GradNorm algorithm introduced in the method to prove the effectiveness of the two modules. The results are shown in Table 2.

[0095] By replacing ordinary convolution with depthwise separable convolution, the number of parameters is reduced from 11.325M to 1.597M, a reduction of 86%, while the prediction accuracy remains comparable, proving the effectiveness of depthwise separable convolution.

[0096] Without adding GradNorm, the prediction accuracy of stomatal conductance and intercellular CO2 concentration varies greatly. 2 The differences are 0.069, 0.04, and 0.046, respectively. After adding GradNorm, R 2 The differences are 0.024, 0.031, and 0.032 respectively, which alleviates the prediction gap between multiple tasks and illustrates the effectiveness of GradNorm in adaptively adjusting the loss value for multi-task learning.

[0097] Table 2 Comparison of ablation tests

[0098]

[0099] The field photosynthetic parameter prediction method of the present application uses multi-source remote sensing data as input, and predicts two photosynthetic parameters end-to-end based on a multi-task deep learning model, solving the time-consuming, labor-intensive, low-efficiency, and difficult-to-develop problems of manually measuring photosynthetic parameters and predicting photosynthetic parameters with a classical machine learning model, and has the advantages of high precision, high efficiency, and ease of application development. The method can also be applied to other crops, providing a new idea for predicting photosynthetic parameters, and helping to accurately and efficiently predict agricultural production.

[0100] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 5As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for predicting photosynthetic parameters in a field is implemented.

[0101] Those skilled in the art will understand that Figure 5 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0102] In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the above-mentioned method for predicting photosynthetic parameters in a field when executing the computer program.

[0103] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the above-mentioned method for predicting photosynthetic parameters in a field is implemented.

[0104] In an exemplary embodiment, a computer program product is provided, including a computer program, which implements the above-mentioned field photosynthetic parameter prediction method when executed by a processor.

[0105] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0106] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0107] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0108] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0109] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A method for predicting photosynthetic parameters in field, characterized in that: include: Obtain visible light images and multispectral images of a set area of ​​a field to be predicted; According to the visible light image and the multispectral image, the photosynthetic parameters of the set area of ​​the field to be predicted are determined using a field photosynthetic parameter prediction model; the photosynthetic parameters include stomatal conductance and intercellular CO2 concentration; wherein the field photosynthetic parameter prediction model is obtained by training an improved convolutional neural network using a training data set combined with a gradient normalization algorithm; the training data set includes visible light images and multispectral images of the training field and corresponding photosynthetic parameters; the improved convolutional neural network is obtained by replacing the convolution layer of the residual module in the convolutional neural network with a depthwise separable neural network. The improved convolutional neural network is obtained by convolutional layer, and the improved convolutional neural network is multi-output; the improved convolutional neural network includes a second fully connected layer, a third fully connected layer, and a first convolutional layer, a maximum pooling layer, a second convolutional layer, a third convolutional layer, a fourth convolutional layer, a fifth convolutional layer, a global average pooling layer, a Flatten function layer and a first fully connected layer connected in sequence; the first fully connected layer is connected to the second fully connected layer and the third fully connected layer respectively; the second convolutional layer, the third convolutional layer, the fourth convolutional layer and the fifth convolutional layer each include two residual modules connected in sequence; The improved convolutional neural network is trained using the training data set, including: Obtain training data sets and build improved convolutional neural networks; Preprocessing the training data set to obtain a preprocessed training data set; the preprocessed training data set includes a plurality of groups of processed visible light images and processed multispectral images and corresponding standardized photosynthetic parameters; Taking the processed visible light image and the processed multispectral image as input and the corresponding standardized photosynthetic parameters as output, the improved convolutional neural network is trained to obtain the field photosynthetic parameter prediction model; Taking the processed visible light image and the processed multispectral image as input and the corresponding standardized photosynthetic parameters as output, the improved convolutional neural network is trained to obtain the field photosynthetic parameter prediction model, which specifically includes: The processed visible light image and the processed multispectral image are input into the current improved convolutional neural network to obtain the predicted values ​​of photosynthetic parameters; Determine a weighted loss function value by combining a weighted loss function with a gradient normalization algorithm according to the predicted value of the photosynthetic parameter and the corresponding standardized photosynthetic parameter; Determine whether the set number of iterations has been reached; If so, the trained improved convolutional neural network corresponding to the weighted loss function value within the set range is used as the field photosynthetic parameter prediction model; If not, adjust the model parameters of the current improved convolutional neural network, and return "input the processed visible light image and the processed multispectral image into the current improved convolutional neural network to obtain the predicted values ​​of photosynthetic parameters".

2. The method for predicting field photosynthetic parameters according to claim 1, characterized in that: Obtain the training data set, including: Use DJI Phantom 4 drones to collect visible light and multispectral images of the training fields; The photosynthetic parameters of each set area in the training field were collected using the LI-6400 photosynthetic meter.

3. The method for predicting field photosynthetic parameters according to claim 1, characterized in that: Preprocessing the training data set to obtain a preprocessed training data set specifically includes: The visible light image and the multispectral image are stitched and radiometrically corrected respectively to obtain a corrected visible light image and a corrected multispectral image; The corrected visible light image is cropped according to the set area of ​​the training field to obtain the corrected visible light image of each set area; The corrected multispectral image is cropped according to the set area of ​​the training field to obtain the corrected multispectral image of each set area; The corrected visible light image of each set area is randomly scaled, randomly cropped, horizontally flipped, and randomly rotated to obtain a processed visible light image; The corrected multispectral image of each set area is randomly scaled, randomly cropped, horizontally flipped and randomly rotated to obtain a processed multispectral image; The photosynthetic parameters were normalized by Z-Score to obtain the standardized photosynthetic parameters.

4. The method for predicting field photosynthetic parameters according to claim 1, characterized in that: The weighted loss function is: ； in, is the weighted loss function value; is the stomatal conductance loss function value; is the weight of the stomatal conductance loss function value; is the intercellular CO2 concentration loss function value; is the weight of the intercellular CO2 concentration loss function value.

5. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method for predicting field photosynthetic parameters according to any one of claims 1 to 4.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for predicting field photosynthetic parameters according to any one of claims 1 to 4 is implemented.

7. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for predicting field photosynthetic parameters according to any one of claims 1 to 4 is implemented.

Images