Soil microbial residue carbon prediction method and training method based on deep learning

Abstract

The present invention discloses a soil microbial residue carbon prediction method and training method based on deep learning. The training method comprises: obtaining normalized multi-category environmental parameter data corresponding to different soil sites, each soil site corresponding to a normalized soil microbial residue carbon content; for each soil site, inputting the normalized multi-category environmental parameter data into an autoencoder for layer-by-layer unsupervised pre-training to extract high-level features of environmental variables; inputting the high-level features of environmental variables into a regressor to output soil microbial residue carbon prediction data, wherein the regressor and the pre-trained autoencoder form a stacked autoencoder network; and adjusting the network parameters of the stacked autoencoder network according to the soil microbial residue carbon content and the soil microbial residue carbon prediction data to obtain a trained soil microbial residue carbon prediction model. The present invention can simply and accurately predict soil microbial residue carbon content under climate change conditions.

Description

Technical Field

[0001] The present invention relates to the technical field of environmental impact assessment, and in particular to a soil microbial residue carbon prediction method and training method based on deep learning. Background Art

[0002] With the introduction of the concept of the soil microbial "carbon pump," microbial necromass carbon (MNC) is considered an important component of the soil carbon pool, accounting for as much as 30-87% of soil organic carbon (SOC). MNC is accumulated through the continuous production of residual substances by microorganisms and is stored in the soil for a long time due to its difficult-to-degrade properties. As an inert component of the soil carbon pool, the fluctuations of MNC under climate change can represent the soil carbon sequestration capacity. Therefore, estimating regional or global changes in MNC under future climate change conditions can help people explore future changes in soil carbon sequestration capacity and carbon storage.

[0003] Numerous environmental factors influence the formation, accumulation, and degradation of MNC. Existing research is largely limited to qualitative exploration of the environmental drivers of MNC, lacking large-scale, regional quantitative studies. Studies have shown that annual mean temperature indirectly influences surface soil MNC through its effects on drought index and net primary productivity; plant carbon input and mineral protection are the most significant drivers of surface soil MNC. Amino sugars are commonly used as biomarkers for quantification of MNC. MNC content is calculated by multiplying muramic acid and glucosamine, which represent bacterial and fungal residues, respectively, by corresponding conversion factors. This measurement method requires large-scale instrumentation such as gas chromatography-mass spectrometry or high-performance liquid chromatography, and the pre-treatment process is relatively cumbersome, making it suitable for small-scale laboratory quantitative studies. Previous studies investigating the impact of climate change on MNC have been limited to small-scale experiments conducted at field stations.

[0004] However, small-scale, fixed-site field experiments can only control experimental subjects under specific experimental conditions. For example, they can set specific warming times and temperatures to simulate the effects of climate warming on MNC, or artificially add certain nutrients to simulate the effects of land use on MNC. These small-scale, fixed-site experiments often require long experimental periods. Most known warming plots have warming periods ranging from 20 to 50 years, and the warming devices are relatively simple, often failing to simulate the actual natural environment of climate warming. Furthermore, the small scale of fixed-site experiments can only reflect soil changes within a small area and are not suitable for large-scale soil studies. This makes it difficult to simulate the impact of future climate change on persistent soil carbon storage in a short period of time. Summary of the Invention

[0005] In view of this, the main purpose of the present invention is to provide a soil microbial residue carbon prediction method and training method based on deep learning, in order to at least partially solve at least one of the above-mentioned technical problems.

[0006] To achieve the above object, the technical solution of the present invention is as follows:

[0007] As one aspect of the present invention, a soil MNC prediction model training method based on deep learning is provided, wherein the soil microbial residue carbon prediction model includes an autoencoder and a regressor, and the method includes: obtaining normalized multi-class environmental parameter data corresponding to different soil sites, wherein each soil site corresponds to a normalized soil MNC content; for each soil site, inputting the normalized multi-class environmental parameter data into the autoencoder for layer-by-layer unsupervised pre-training to extract high-level features of environmental variables; inputting the high-level features of environmental variables into the regressor to output soil MNC prediction data, wherein the regressor and the pre-trained autoencoder constitute a stacked autoencoder network; and adjusting the network parameters of the stacked autoencoder network according to the soil MNC content and the soil MNC prediction data to obtain a trained soil MNC prediction model.

[0008] As another aspect of the present invention, a soil MNC prediction method based on deep learning is provided, comprising: obtaining normalized multi-category environmental parameter data of the soil to be predicted as a data sample to be predicted; inputting the data sample to be predicted into a soil MNC prediction model to obtain a predicted value of the MNC content of the soil to be predicted; wherein the soil MNC prediction model is pre-trained using the training method described above.

[0009] As another aspect of the present invention, a soil MNC prediction model training device based on deep learning is provided, wherein the soil microbial residue carbon prediction model includes an autoencoder and a regressor, and the device includes: an acquisition module for acquiring normalized multi-class environmental parameter data corresponding to different soil sites, wherein each soil site corresponds to a normalized soil MNC content; a pre-training module for inputting the normalized multi-class environmental parameter data into the autoencoder for layer-by-layer unsupervised pre-training for each soil site, and extracting high-level features of environmental variables; a regression module for inputting the high-level features of environmental variables into the regressor and outputting soil MNC prediction data, wherein the regressor and the pre-trained autoencoder constitute a stacked autoencoder network; a training module for adjusting the network parameters of the stacked autoencoder network according to the soil MNC content and the soil MNC prediction data to obtain a trained soil MNC prediction model.

[0010] As another aspect of the present invention, an electronic device is provided, comprising: one or more processors; a storage device for storing one or more programs, wherein, when the one or more programs are executed by the one or more processors, the one or more processors execute the method as described above.

[0011] Based on the above technical solution, the soil microbial residue carbon prediction method and training method based on deep learning of the present invention have at least one or part of the following beneficial effects:

[0012] The present invention establishes training samples based on data from multiple environmental parameters and soil MNC content at different soil sites. After training the constructed soil MNC prediction model, the resulting model is used to predict soil MNC for the data samples to be predicted. Because the trained model extracts high-level features of multiple environmental parameters that influence soil MNC changes, it can cover multiple environmental parameters at different sites across a wide area, not just in small, fixed-point field locations. It also eliminates the need for long-term field experiments and enables relatively accurate soil MNC predictions, providing a convenient method for predicting global changes in soil MNC under future climate change conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 is a flowchart of a soil MNC prediction model training method based on deep learning according to an embodiment of the present invention;

[0014] Figure 2 This is a structural diagram of a soil MNC model according to an embodiment of the present invention;

[0015] Figure 3 is a flow chart of a method for adjusting network parameters of a stacked autoencoder network according to an embodiment of the present invention;

[0016] Figure 4 Flowchart of a method for predicting soil microbial residue carbon based on deep learning according to an embodiment of the present invention;

[0017] Figure 5 This is a structural block diagram of a soil microbial residue carbon prediction model training device based on deep learning according to an embodiment of the present invention;

[0018] Figure 6 This is a block diagram of an electronic device suitable for implementing a soil microbial residue carbon prediction model training method based on deep learning according to an embodiment of the present invention. DETAILED DESCRIPTION

[0019] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with specific embodiments and with reference to the accompanying drawings.

[0020] Through research, it was found that various environmental factors such as ecosystem specificity and the duration of climate change have complex effects on the carbon sequestration process mediated by microbial residues, resulting in significant differences in the distribution of MNC under climate change. In related technologies, the prediction method for estimating soil MNC is based on field fixed-point experimental devices to simulate climate change observations. This method requires a long experimental cycle, is less efficient, and cannot fully simulate climate change under natural environmental conditions. In the process of realizing the present invention, it was found that by establishing training samples based on multiple types of environmental parameter data and soil microbial residue carbon content corresponding to different soil sites, the soil MNC prediction model was trained using the training samples to obtain a trained soil MNC prediction model, which provides technical support for the subsequent prediction of soil MNC data, has good soil MNC prediction accuracy, and provides a convenient quantitative method for predicting global changes in soil MNC under future climate change conditions.

[0021] Specifically, according to an embodiment of the present invention, a soil MNC prediction model training method based on deep learning is provided, wherein the soil MNC model includes an autoencoder and a regressor. Figure 1 Flowchart of the soil MNC prediction model training method based on deep learning according to an embodiment of the present invention. Figure 1 As shown, the training method includes operations S101 to S104.

[0022] In operation S101, normalized multi-category environmental parameter data corresponding to different soil sites are obtained, wherein each soil site corresponds to a normalized soil MNC content;

[0023] In operation S102, for each soil site, the normalized multi-category environmental parameter data is input into an autoencoder to perform layer-by-layer unsupervised pre-training to extract high-level features of environmental variables;

[0024] In operation S103, the high-level features of the environmental variables are input into the regressor, and soil MNC prediction data is output, wherein the regressor and the pre-trained autoencoder form a stacked autoencoder network;

[0025] In operation S104 , the network parameters of the stacked autoencoder network are adjusted according to the soil MNC content and the soil MNC prediction data to obtain a trained soil MNC prediction model.

[0026] According to embodiments of the present invention, obtaining normalized environmental parameter data and corresponding soil MNC data supports the construction of training samples and the training of a soil MNC prediction model. Normalization restricts the environmental parameters and soil MNC data to a specific range, thereby eliminating the adverse effects of outliers and improving the accuracy of soil MNC prediction results. Using these training samples to train a soil MNC prediction model, the trained soil MNC prediction model incorporates high-level features of environmental parameter variations and can cover diverse sites across a wide area. This method is simple and easy to implement, providing technical support for the subsequent, efficient and accurate prediction of soil MNC data.

[0027] According to an embodiment of the present invention, in step S101, the multiple environmental parameters include climate and geographic parameters, vegetation parameters, soil parameters, and microbial parameters. More specifically, climate and geographic parameters may include: annual average temperature, annual average precipitation, drought index, and altitude; vegetation parameters may include: normalized difference vegetation index and net primary productivity; soil parameters may include: total carbon, total nitrogen, total phosphorus, total potassium, total organic carbon, pH, soil bulk density, silt content, sand content, clay content, gravel content, soil thickness, cation exchange capacity, ammonia content, and nitrate content; and microbial parameters include: bacterial biomass, fungal biomass, relative bacterial abundance, and relative fungal abundance. More specifically, the aforementioned multiple environmental parameter data can be obtained from meteorological websites or global soil datasets as ArcGIS raster data corresponding to each soil site.

[0028] According to an embodiment of the present invention, in step S101, soil MNC content is quantitatively measured using amino sugars as biomarkers. Specifically, this can be achieved through direct sampling and / or by collecting a dataset of MNC-related data measured using this method. This expansion of existing datasets provides technical support and a basis for subsequent training sample construction and model training, which is of great significance.

[0029] According to an embodiment of the present invention, in step S101, the normalization method may be, for example, mean-variance normalization. Specifically, each soil site corresponds to a sample, and various environmental parameter data or soil MNC content are sample data. The various environmental parameter data or soil MNC content of each soil site are normalized based on the following formula:

[0030]

[0031] In formula (1), x nor is the normalized sample data, x is the sample data, x mean is the mean of the sample data, x std is the standard deviation of the sample data.

[0032] According to an embodiment of the present invention, training samples are constructed using normalized multi-category environmental parameter data and normalized soil MNC content to train the soil MNC prediction model. Since the normalization method restricts the various types of environmental parameter data and soil MNC data to a certain range, the adverse effects of singular data are eliminated, which is conducive to improving the accuracy of MNC prediction results.

[0033] According to an embodiment of the present invention, Figure 2 : is a structural diagram of the soil MNC model according to an embodiment of the present invention, such as Figure 2 As shown in Figure 2, the soil MNC model includes autoencoders (AEs) to generate the model, attempting to reconstruct the input of various environmental variables by extracting high-level features. Figure 2 As shown in Figure 1, the autoencoder structure consists of an input layer, a hidden layer, and an output layer. The input layer is used to input various environmental parameter data, the hidden layer is used to compress the environmental parameter data and release the variable dimension, and the output layer is used to output the extracted high-level features of the environmental variables.

[0034] More specifically, in operation S102, the main steps of pre-training the autoencoder include setting model hyperparameters, defining the model structure, and pre-training the model. The main hyperparameters of the autoencoder include: the number of samples, the learning rate, the number of hidden layers and the compression dimension, the activation function, etc. The number of samples primarily affects the calculation of the loss function and the training speed, while the learning rate primarily affects the weight adjustment and the training speed.

[0035] According to an embodiment of the present invention, by pre-training the autoencoder to extract high-level features of various environmental parameters and initialize the model structure, model overfitting in the subsequent supervised training process can be reduced, and learning the input distribution helps to learn the mapping from input to output.

[0036] According to an embodiment of the present invention, Figure 2 As shown, the soil MNC model also includes a regressor for prediction using the high-level features of environmental variables extracted by the autoencoder. The regressor employs a neural network for modeling and parameter adjustment. Specifically, the regressor also includes a hidden layer. After the autoencoder pre-training in operation S102, the extracted high-level features of each environmental variable are incorporated into the weights of the hidden layer.

[0037] According to an embodiment of the present invention, Figure 3 FIG. 1 is a flow chart of a method for adjusting network parameters of a stacked autoencoder network according to an embodiment of the present invention. Figure 3 As shown, operation S104 specifically includes:

[0038] S1041: determining a loss function based on soil MNC content and soil MNC prediction data;

[0039] S1042: Adjusting the network parameters of the stacked autoencoder network according to the loss function, and obtaining a trained soil MNC prediction model when the loss function is less than a first preset threshold.

[0040] According to an embodiment of the present invention, the calculation formula of the loss function is:

[0041]

[0042] In formula (2), L sr is the loss function, N is the number of soil MNC content, that is, the number of samples, T model is the soil MNC prediction data, T obs is the soil MNC content.

[0043] According to an embodiment of the present invention, fine-tuning the network parameters of the stacked autoencoder network after pre-training can extract high-level features for more robust prediction.

[0044] According to an embodiment of the present invention, in order to improve the prediction accuracy of the soil MNC prediction model, operation S104 further includes: performing linear fitting on the soil MNC content and the soil MNC prediction data, and calculating R , which represents the goodness of fit. 2 , root mean square error (RMSE); in R 2 When the second preset threshold is exceeded and the root mean square error is less than the third preset threshold, the trained soil MNC prediction model is obtained. Further, optionally, the average of the 10 model performance evaluation results can be taken as the final root mean square error result.

[0045] According to an embodiment of the present invention, the calculation formula of the root mean square error is as follows:

[0046]

[0047] Where x RMSE is the root mean square error, x MSE is the mean square error, N is the amount of soil MNC content, that is, the number of samples, x obs，i is the soil MNC content of the i-th soil site, x model，i is the soil MNC prediction data of the i-th soil site.

[0048] According to an embodiment of the present invention, in order to improve the prediction accuracy of the soil MNC prediction model, the sample data corresponding to different soil sites are further divided into training samples and verification samples, wherein the sample data includes normalized multi-category environmental parameter data and soil MNC content. The training samples may, for example, account for 90% of the total samples and are used to perform the above operations S102 to S104. The verification samples account for 10% of the total samples and are used to verify the trained soil microbial residue carbon prediction model to evaluate the accuracy of the model.

[0049] According to an embodiment of the present invention, a soil microbial residue carbon prediction method based on deep learning is also provided. Figure 4 Flowchart of the soil microbial residue carbon prediction method based on deep learning according to an embodiment of the present invention. Figure 4 As shown, the method includes operations S401 to S402.

[0050] In operation S401, normalized multi-category environmental parameter data of the soil to be predicted is obtained as a data sample to be predicted;

[0051] In operation S402, the data sample to be predicted is input into the soil MNC prediction model to obtain a predicted value of the MNC content of the soil to be predicted;

[0052] The soil MNC prediction model is pre-trained using the training method described above.

[0053] According to an embodiment of the present invention, soil MNC content is predicted using a trained soil MNC model based on multiple environmental parameter data of the soil to be predicted, thereby improving the prediction efficiency of soil MNC. This method is simple and easy to implement, avoids long-term field experiments, ensures the accuracy of soil MNC prediction values, and provides a convenient method for predicting global changes in soil MNC under future climate change conditions.

[0054] According to an embodiment of the present invention, operation S401 specifically includes: obtaining various environmental parameter data of the soil to be predicted through methods such as satellite remote sensing data and soil physical and chemical measurements; and normalizing the various environmental parameter data of the soil to be predicted to obtain a data sample to be predicted. More specifically, based on the above formula (1), the various environmental parameter data of the soil to be predicted are combined with the various environmental parameter data obtained at different soil sites to normalize the various environmental parameter data of the soil to be predicted.

[0055] According to an embodiment of the present invention, a soil microbial residue carbon prediction model training device based on deep learning is also provided. Figure 5 This is a structural block diagram of a soil microbial residue carbon prediction model training device based on deep learning according to an embodiment of the present invention. Figure 5As shown, the device includes: an acquisition module 501, a pre-training module 502, a regression module 503 and a training module 504.

[0056] The acquisition module 501 is used to acquire normalized multi-category environmental parameter data corresponding to different soil sites, wherein each soil site corresponds to a normalized soil MNC content.

[0057] The pre-training module 502 is used to input the normalized multi-category environmental parameter data into the autoencoder for layer-by-layer unsupervised pre-training for each soil site, thereby extracting high-level features of the environmental variables.

[0058] The regression module 503 is used to input the high-level features of environmental variables into the regressor and output soil MNC prediction data, wherein the regressor and the pre-trained autoencoder constitute a stacked autoencoder network.

[0059] The training module 501 is used to adjust the network parameters of the stacked autoencoder network according to the soil MNC content and the soil MNC prediction data to obtain a trained soil MNC prediction model.

[0060] According to an embodiment of the present disclosure, any multiple modules of the acquisition module 501, the pre-training module 502, the regression module 503 and the training module 504 can be combined into one module for implementation, or any one of the modules can be split into multiple modules. Alternatively, at least part of the functions of one or more of these modules can be combined with at least part of the functions of other modules and implemented in one module. According to an embodiment of the present disclosure, at least one of the acquisition module 501, the pre-training module 502, the regression module 503 and the training module 504 can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application specific integrated circuit (ASIC), or can be implemented by hardware or firmware such as any other reasonable way of integrating or packaging the circuit, or implemented in any one of the three implementation methods of software, hardware and firmware or in an appropriate combination of any of them. Alternatively, at least one of the acquisition module 501 , the pre-training module 502 , the regression module 503 and the training module 504 may be at least partially implemented as a computer program module, which may perform corresponding functions when executed.

[0061] Figure 6 This is a block diagram of an electronic device suitable for implementing a soil microbial residue carbon prediction model training method based on deep learning according to an embodiment of the present invention.

[0062] like Figure 6As shown, the electronic device 600 according to an embodiment of the present invention includes a processor 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage portion 608 into a random access memory (RAM) 603. The processor 601 may, for example, include a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or a related chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 601 may also include an onboard memory for caching purposes. The processor 601 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present invention.

[0063] Various programs and data required for the operation of the electronic device 600 are stored in the RAM 603. The processor 601, ROM 602, and RAM 603 are connected to each other via a bus 604. The processor 601 executes the programs in the ROM 602 and / or RAM 603 to perform various operations according to the method flow of the embodiment of the present invention. It should be noted that the programs may also be stored in one or more memories other than the ROM 602 and RAM 603. The processor 601 may also execute the programs stored in the one or more memories to perform various operations according to the method flow of the embodiment of the present invention.

[0064] According to an embodiment of the present invention, electronic device 600 may further include an input / output (I / O) interface 605, which is also connected to bus 604. Electronic device 600 may further include one or more of the following components connected to I / O interface 605: an input section 606 including a keyboard, a mouse, etc.; an output section 607 including devices such as a cathode ray tube (CRT), a liquid crystal display (LCD), and a speaker; a storage section 608 including a hard disk; and a communication section 609 including a network interface card such as a LAN card or a modem. Communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to I / O interface 605 as needed. Removable media 611, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed in drive 610 as needed, so that computer programs read from the removable media can be installed into storage section 608 as needed.

[0065] The specific embodiments described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above are only specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for training a soil microbial residue carbon prediction model based on deep learning, wherein the soil microbial residue carbon prediction model includes an autoencoder and a regressor, and the method comprises: Obtaining normalized multi-category environmental parameter data corresponding to different soil sites, wherein each soil site corresponds to a normalized soil microbial residue carbon content, the multi-category environmental parameters including climate and geographic parameters, vegetation parameters, soil parameters, and microbial parameters, and the soil microbial residue carbon content is quantitatively measured using amino sugars as biomarkers; For each soil site, inputting the normalized multi-category environmental parameter data into the autoencoder to perform layer-by-layer unsupervised pre-training to extract high-level features of environmental variables; Inputting the high-level features of the environmental variables into the regressor to output soil microbial residue carbon prediction data, wherein the regressor and the pre-trained autoencoder constitute a stacked autoencoder network; According to the soil microbial residue carbon content and the soil microbial residue carbon prediction data, the network parameters of the stacked autoencoder network are adjusted to obtain a trained soil microbial residue carbon prediction model, including: determining a loss function based on the soil microbial residue carbon content and the soil microbial residue carbon prediction data, adjusting network parameters of the stacked autoencoder network based on the loss function, and obtaining a trained soil microbial residue carbon prediction model when the loss function is less than a first preset threshold; Furthermore, a linear fit is performed on the soil microbial residue carbon content and the soil microbial residue carbon prediction data to calculate the R value indicating the goodness of fit. 2 , root mean square error; in R 2 When the second preset threshold is exceeded and the root mean square error is less than the third preset threshold, the trained soil microbial residue carbon prediction model is obtained.

2. The method according to claim 1, wherein: The climate and geographical parameters include: annual average temperature, annual average precipitation, drought index, and altitude; The vegetation parameters include: normalized vegetation index, net primary productivity; The soil parameters include: total carbon, total nitrogen, total phosphorus, total potassium, total organic carbon, pH, soil bulk density, silt content, sand content, clay content, gravel content, soil thickness, cation exchange capacity, ammonia content, and nitrate content; The microbial parameters include: bacterial biomass, fungal biomass, bacterial relative abundance, and fungal relative abundance.

3. The method according to claim 1, wherein The calculation formula of the loss function is: in, is the loss function, N is the carbon content of the soil microbial residues, is the soil microbial residue carbon prediction data, is the carbon content of soil microbial residues.

4. The method according to claim 1, wherein: The normalization method is mean variance normalization; The autoencoder includes an input layer, a hidden layer and an output layer; The regressor includes multiple hidden layers.

5. A deep learning-based soil microbial residue carbon prediction method, comprising: Obtain normalized multi-category environmental parameter data of the soil to be predicted as a data sample to be predicted; Inputting the data sample to be predicted into the soil microbial residue carbon prediction model to obtain a predicted value of the microbial residue carbon content of the soil to be predicted; Wherein, the soil microbial residue carbon prediction model is pre-trained by the training method according to any one of claims 1 to 4.

6. A deep learning-based soil microbial residue carbon prediction model training device, wherein the soil microbial residue carbon prediction model includes an autoencoder and a regressor, and the device includes: An acquisition module is used to obtain normalized multi-category environmental parameter data corresponding to different soil sites, wherein each soil site corresponds to a normalized soil microbial residue carbon content; A pre-training module is used to input the normalized multi-category environmental parameter data into the autoencoder for layer-by-layer unsupervised pre-training for each soil site to extract high-level features of environmental variables; A regression module, configured to input the high-level features of the environmental variables into the regressor and output soil microbial residue carbon prediction data, wherein the regressor and the pre-trained autoencoder constitute a stacked autoencoder network; A training module is used to adjust the network parameters of the stacked autoencoder network according to the soil microbial residue carbon content and the soil microbial residue carbon prediction data to obtain a trained soil microbial residue carbon prediction model, including: determining a loss function according to the soil microbial residue carbon content and the soil microbial residue carbon prediction data, adjusting the network parameters of the stacked autoencoder network according to the loss function, and obtaining the trained soil microbial residue carbon prediction model when the loss function is less than a first preset threshold; and performing linear fitting on the soil microbial residue carbon content and the soil microbial residue carbon prediction data to calculate R representing the goodness of fit. 2 , root mean square error; in R 2 When the second preset threshold is exceeded and the root mean square error is less than the third preset threshold, the trained soil microbial residue carbon prediction model is obtained.

7. An electronic device comprising: one or more processors; a storage device for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors are enabled to perform the method according to any one of claims 1 to 4.

Images