A method for inversion and reconstruction of farmland temperature based on iterative depth perception machine

Abstract

The invention proposes a farmland temperature inversion and reconstruction method based on an iterative deep and shallow sensor, which belongs to the technical field of farmland meteorological monitoring. The invention uses historical temperature grid data, farmland temperature data, longitude and latitude coordinates of farmland, altitude of farmland and other information to train a farmland temperature meteorological data inversion model and a temperature meteorological grid data reconstruction model. After the model training is completed, the farmland temperature meteorological data inversion model and the temperature meteorological grid data reconstruction model are connected in series. The original temperature grid point actual data are first input into the temperature meteorological grid data reconstruction model for correction, and the corrected surface temperature data are then input into the farmland temperature meteorological data inversion model to invert the farmland temperature in the temperature grid data. The invention can be reconstructed according to the farmland microclimate observation station data, and the reconstructed grid point actual data can have the characteristics of the farmland microclimate and be more in line with the actual situation of the farmland.

Description

Technical Field

[0001] The present invention belongs to the technical field of farmland meteorological monitoring, and in particular, relates to a farmland temperature inversion and reconstruction method based on an iterative depth perception machine. Background Art

[0002] Real-time monitoring of meteorological environmental information is very important for crop growth and development. However, the cost of full coverage monitoring of farmland meteorological data is very high. To solve this problem, it is necessary to invert farmland meteorological data based on the actual data of meteorological grid points. The actual data of grid points are obtained from ground-based automatic meteorological observation stations through interpolation methods such as Kriging interpolation and inverse distance weighted interpolation. However, most of the ground-based automatic meteorological observation stations are built in cities and cannot accurately reflect the characteristics of farmland microclimate. Therefore, due to the problems of meteorological grid data itself, inverting farmland meteorological data based entirely on existing meteorological grid data will result in large errors.

[0003] Existing methods for inverting farmland temperature mainly include statistical methods and mathematical modeling. Statistical methods often average or interpolate observation station data to obtain relatively uniform grid data. However, the growth of crops will affect the surrounding meteorological environment, and this rough processing method cannot effectively capture the complexity of the microclimate of farmland; mathematical modeling methods usually use meteorological models, but these models often cannot provide grid data with sufficiently high spatial resolution when dealing with micro-meteorological phenomena. On the other hand, the meteorological grid data used for inversion are not accurate in themselves, so using these data to invert farmland meteorological data will increase errors, and existing methods do not consider correcting or further processing the original meteorological grid data. In general, the lack of sufficiently accurate inversion models and the low-confidence meteorological grid data required for inversion jointly affect the accuracy of temperature data inversion by existing methods.

[0004] Most of the existing temperature inversion models use some physical methods, statistical methods and artificial intelligence methods to invert data. The physical method simulates the energy exchange process between the atmosphere, the surface and the earth's surface, thereby inferring the surface or near-surface temperature. Although the model generated by the physical method is highly explanatory, the physical method requires a large number of physical processes and parameters, has a large amount of calculation, and runs slowly. The statistical method has a small amount of calculation and runs quickly. Artificial intelligence methods, such as using satellite images to invert the temperature, need to be compared to see whether they fit the actual temperature on the ground.

[0005] In addition, a new fully coupled framework that integrates physical methods, statistical methods and deep learning to establish a group of near-surface temperature inversion equations and uses deep learning methods to optimize the algorithm can stably and directly invert the near-surface temperature from thermal infrared remote sensing data. However, the input data used in the existing inversion methods are all highly confident data, and do not consider data reconstruction or data correction to use more accurate data to invert the target.

[0006] Existing reconstruction methods for meteorological grid data can be roughly divided into traditional statistical methods and deep learning-based methods. Traditional reconstruction methods use time series analysis methods to reconstruct data. However, these methods often use fixed analysis patterns and do not take into account the impact of time, location, season, and different crops on the reconstruction results, and do not use a large amount of historical data.

[0007] At the same time, the deep learning-based method only uses historical multi-mode meteorological element grid prediction data for modeling. Although it can improve the temporal and spatial resolution, the corrected grid prediction data cannot accurately reflect the actual meteorological conditions; even if it is based on automatic regional meteorological observation stations, although it covers a wide range in time and space, it cannot reflect the specific microclimate conditions.

[0008] Although the above methods have achieved good results, they do not consider the impact of ground crops on meteorological elements, and do not make good use of the impact of auxiliary information such as time, date, and terrain on the reconstruction results. Summary of the invention

[0009] In view of the shortcomings of the existing technology, in order to more accurately invert farmland temperature data, the present invention proposes a farmland temperature inversion and reconstruction method based on an iterative depth and shallow perception machine, which can reconstruct the grid actual data according to the farmland microclimate observation station data. The reconstructed grid actual data can have the characteristics of the farmland microclimate and be more in line with the actual situation of the farmland.

[0010] The present invention is achieved through the following technical solutions:

[0011] A method for inversion and reconstruction of farmland temperature based on iterative depth perception machine:

[0012] The method specifically comprises the following steps:

[0013] Step 1: Use historical temperature grid data, farmland temperature data, farmland longitude and latitude coordinates, and farmland altitude information to train the farmland temperature meteorological data inversion model and the temperature meteorological grid data reconstruction model;

[0014] Step 2: connect the farmland temperature meteorological data inversion model and the temperature meteorological grid data reconstruction model trained in step 1 in series;

[0015] Step 3: Input the original temperature grid point data into the model connected in series in step 2. The data is first input into the temperature meteorological grid data reconstruction model for correction to generate a new set of grid point information.

[0016] Step 4: The corrected surface temperature data is then input into the farmland temperature meteorological data inversion model. The farmland temperature within the temperature grid data is inverted and calculated using the data reconstructed in step 3 combined with auxiliary information. Finally, the farmland temperature meteorological data inversion model outputs the inverted farmland temperature data.

[0017] Furthermore, the farmland temperature meteorological data inversion model and the temperature meteorological grid data reconstruction model both adopt a deep and shallow perceptron model structure, and both models include three parts;

[0018] The first part is a deep neural network with three fully connected layers; the deep neural network inputs auxiliary information;

[0019] The second part is a wide neural network with a fully connected layer. The wide neural network inputs grid information, which is used to provide high-order cross features between auxiliary information and grid information, and provides feature representation for the output of the model.

[0020] The third part is the final output layer which contains two fully connected layers; the final output layer converts the feature vectors of the connected deep neural network and wide neural network into the final output value of the model.

[0021] Furthermore, the input of the temperature meteorological grid data reconstruction model is the original grid point information and the auxiliary information, and the output is the four sets of new grid point information after reconstruction;

[0022] The input of the farmland temperature meteorological data inversion model is the original grid point information, the newly generated grid point information and auxiliary information after reconstruction by the temperature meteorological grid data reconstruction model, and the output is a farmland temperature data.

[0023] Furthermore, in step 1, the training method of the farmland temperature meteorological data inversion model and the temperature meteorological grid data reconstruction model is an optimization process based on cyclic iteration;

[0024] Step 1.1, in the first cycle, the farmland temperature meteorological data inversion model is first trained. The model inputs auxiliary information and original grid information. After the model is trained, the gradient of the inversion model parameters is calculated by the back propagation algorithm, and the optimization algorithm is used to update the parameters so that the loss function is gradually reduced;

[0025] Step 1.2, then train the temperature meteorological grid data reconstruction model, input auxiliary information and original grid data into the temperature meteorological grid data reconstruction model, calculate the parameter gradient of the reconstruction model through the back propagation algorithm after the reconstruction model is trained, use the optimization algorithm to update the parameters, so that the loss function gradually decreases, and finally output the new four groups of grid information after reconstruction;

[0026] Step 1.3, the second cycle first trains the farmland temperature meteorological data inversion model. At this time, the input data of the farmland temperature meteorological data inversion model includes auxiliary information and the newly generated grid data after the first cycle of temperature meteorological grid data reconstruction model training; after the farmland temperature meteorological data inversion model training, the parameter gradient is calculated and the parameters are updated to reduce the loss function;

[0027] Step 1.4, then train the temperature meteorological grid data reconstruction model and output four new sets of grid point data;

[0028] Step 1.5, continue the optimization cycle according to the second loop, input the auxiliary information and the new grid information generated by the reconstruction model, and output the inverted farmland temperature data until the loss function of the inversion model no longer decreases, then stop training.

[0029] Furthermore, in step 1, the performance of the model is measured by calculating the sum of the absolute values ​​of the differences between the model output and the true label through the reconstruction difference and the inversion difference;

[0030] Reconstruction difference MAE G Refers to the grid points obtained by reconstructing the temperature meteorological grid data reconstruction model With actual grid The difference between

[0031]

[0032] Where N is the number of samples, Where N is the number of samples, is the grid point output by the i-th sample model, is the actual grid point of the i-th sample;

[0033] The inversion difference MAEp refers to the temperature value obtained by inverting the farmland temperature meteorological data inversion model. The actual temperature value The difference between

[0034]

[0035] Where N is the number of samples, is the temperature value output by the i-th sample model, is the true temperature of the i-th sample. From the above formula, we can see that MAE GThe smaller the MAEp is, the better the model reconstruction and inversion effects are.

[0036] Furthermore, in step 2, the output of the temperature meteorological grid data reconstruction model is used as the input of the farmland temperature meteorological data inversion model.

[0037] A farmland temperature inversion and reconstruction system based on iterative depth perception machine:

[0038] The system includes a training module, a reconstruction module and an inversion module:

[0039] The training module uses historical temperature grid data, farmland temperature data, farmland latitude and longitude coordinates, and farmland altitude information to train the farmland temperature meteorological data inversion model and the temperature meteorological grid data reconstruction model; and then connects the trained farmland temperature meteorological data inversion model and the temperature meteorological grid data reconstruction model in series;

[0040] In the reconstruction module, the original temperature grid point actual data is input into the serial model, and the data is first input into the temperature meteorological grid data reconstruction model for correction to generate a set of new grid point information;

[0041] In the inversion module, the corrected surface temperature data is input into the farmland temperature meteorological data inversion model, and the reconstructed data is used in combination with auxiliary information to invert and calculate the farmland temperature within the temperature grid data; finally, the farmland temperature meteorological data inversion model outputs the inverted farmland temperature data.

[0042] An electronic device comprises a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.

[0043] A computer-readable storage medium is used to store computer instructions, and when the computer instructions are executed by a processor, the steps of the above method are implemented.

[0044] Beneficial effects of the present invention

[0045] The present invention can reconstruct the grid point real-time data issued by the China Meteorological Data Land Surface Assimilation System (CLDAS) based on the farmland microclimate observation station data. The reconstructed grid point real-time data can have the characteristics of farmland microclimate and better fit the actual situation of farmland. At the same time, the technology of the present invention can invert farmland data, and gridded farmland microclimate data can be inverted for any longitude and latitude information.

[0046] (1) For the inversion model, the inversion model can be used to accurately and efficiently interpolate the missing data of the farmland microclimate station, ensure the integrity and consistency of the farmland microclimate observation data, and provide data support for the evaluation of agricultural meteorological conditions, disaster monitoring, early warning, assessment and crop yield estimation.

[0047] (2) For the reconstruction model, the reconstruction model can be used to simulate hourly data at a scale of 1 km*1 km in areas without farmland microclimate observation stations. At the same time, the reconstructed grid data is more consistent with the farmland scale characteristics, thereby improving the refinement and pertinence of agricultural meteorological services.

[0048] (3) By comprehensively utilizing the reconstruction model and the inversion model to construct a 1 km*1 km grid real-time multi-factor dataset based on farmland, the economic cost of building farmland microclimate observation stations and the labor cost of operating and maintaining farmland microclimate observation stations and data quality control can be greatly reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 This is a structural diagram of the farmland temperature inversion and reconstruction model based on the iterative depth perception machine of the present invention.

[0050] Figure 2 Schematic diagram of the model training mechanism of the present invention.

[0051] Figure 3 This is the first experimental model evaluation result of the present invention.

[0052] Figure 4 This is the result of the second experimental model evaluation of the present invention. DETAILED DESCRIPTION

[0053] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0054] The experimental methods used in the following examples are conventional methods unless otherwise specified. The materials, reagents, methods and instruments used are conventional materials, reagents, methods and instruments in the art unless otherwise specified, and can be obtained through commercial channels by those skilled in the art.

[0055] Combination Figures 1 to 4 A method for inversion and reconstruction of farmland temperature based on iterative depth perception machine:

[0056] Step 1: First, use the historical temperature grid data, farmland temperature data, farmland latitude and longitude coordinates, farmland altitude and other information to train the farmland temperature meteorological data inversion model (PointInversion) and temperature meteorological grid data reconstruction model (GirdRegenerator);

[0057] The farmland temperature meteorological data inversion model (PointInversion) and the temperature meteorological grid data reconstruction model (GridRegenerator) both adopt the deep and shallow perceptron model structure, and both models include three parts.

[0058] The first part is a deep neural network with three full-layer links. The deep neural network inputs auxiliary information, which includes time information, location information, altitude information, and other information that is weakly correlated with temperature data. The deep neural network can better mine the relationship between auxiliary information and grid data. Since the auxiliary information is weakly correlated with the output that the model needs to generate, deeper parameters are needed to perform feature learning and representation learning on the input data to capture the nonlinear transformation between the two.

[0059] The second part is a wide neural network with a fully connected layer. The wide neural network inputs grid information, which is mainly used for high-order cross features between auxiliary information and grid information, providing rich feature representation for the output of the model. Finally, the outputs of the deep neural network and the wide neural network are connected together.

[0060] The third part is the final output layer which contains two fully connected layers. The final output layer converts the feature vectors of the connected deep neural network and wide neural network into the final output value of the model.

[0061] The difference between the farmland temperature meteorological data inversion model (PointInversion) and the temperature meteorological grid data reconstruction model (GridRegenerator) is that their input and output are different;

[0062] For the temperature meteorological grid data reconstruction model (GirdRegenerator), its input mainly includes original grid point information and auxiliary information, and the output result is four new sets of grid point information.

[0063] For the farmland temperature meteorological data inversion model (PointInversion), its input is the original grid information, the newly generated grid information of the reconstruction model and the auxiliary information, and the output result is a farmland temperature data.

[0064] The training method of the two models is based on a cyclic iterative optimization process. The training process is as follows Figure 2As shown in Figure 2, the first cycle of the model consists of two parts:

[0065] First, the inversion model of farmland temperature meteorological data is trained. The model inputs auxiliary information and original grid information. After the model is trained, the gradient of the inversion model parameters is calculated through the back propagation algorithm, and the optimization algorithm is used to update the parameters so that the loss function gradually decreases.

[0066] Next, the temperature and meteorological grid data reconstruction model is trained. The reconstruction model inputs auxiliary information and original grid data. After the reconstruction model is trained, the parameter gradient of the reconstruction model is calculated through the back propagation algorithm, and the parameters are updated using the optimization algorithm to gradually reduce the loss function. Finally, the four new sets of grid information after reconstruction are output.

[0067] Similarly, the second cycle also consists of two parts.

[0068] The first is the inversion model training of farmland temperature meteorological data. It is particularly noteworthy that, unlike the first cycle, the input data of the inversion model in the second cycle contains auxiliary information and new grid data generated after the reconstruction model training.

[0069] After the inversion model is trained, the parameter gradient is calculated and updated to reduce the loss function. Next, the temperature and meteorological grid data reconstruction model is trained to output four new grid point data.

[0070] The subsequent cycles are the same as the second cycle, inputting auxiliary information and new grid information generated by the reconstruction model, and outputting the inverted farmland temperature data. Training is stopped when the loss function of the inversion model no longer decreases.

[0071] The reconstruction difference and inversion difference of the model are calculated by calculating the sum of the absolute values ​​of the differences between the model output and the true label to measure the performance of the model.

[0072] Reconstruction difference MAE G Refers to the grid points obtained by reconstructing the GridRegenerator model With actual grid The difference between.

[0073]

[0074] Where N is the number of samples, Where N is the number of samples, is the grid point output by the i-th sample model, is the actual grid point of the ith sample.

[0075] The inversion difference MAEp refers to the temperature value obtained by inversion of the PointInversion model. The actual temperature value The difference between.

[0076]

[0077] Where N is the number of samples, is the temperature value output by the i-th sample model, is the true temperature of the i-th sample. From the above formula, we can see that MAE G The smaller the MAEp is, the better the model reconstruction and inversion effects are.

[0078] Step 2: After the model is trained, the farmland temperature meteorological data inversion model and the temperature meteorological grid data reconstruction model are connected in series.

[0079] The output of the temperature and meteorological grid data reconstruction model (GridRegenerator) becomes the input of the farmland temperature and meteorological data inversion model (PointInversion), and the two are closely linked to form a continuous processing flow. In this way, the prediction ability of the model can be more effectively utilized to improve the performance and accuracy of the entire system.

[0080] Step 3: The original temperature grid point data is first input into the temperature meteorological grid data reconstruction model for correction to generate a new and more accurate set of grid point information.

[0081] Step 4: The corrected surface temperature data (new grid data generated by the reconstruction model) are then input into the farmland temperature meteorological data inversion model. Using these reconstructed data and combined with auxiliary information, the farmland temperature within the temperature grid data is inverted and calculated; finally, the PointInversion model outputs the inverted farmland temperature data.

[0082] Example:

[0083] The model evaluation experiment was conducted twice, the first experiment was a selective site evaluation, and the second experiment was an overall site evaluation. The original data were the data of 22 farmland microclimate observation stations in Shandong Province from 2021 to 2023 and the corresponding grid data. The data input into the model were divided into training set and test set.

[0084] The training set of the selective site evaluation experiment is the data of 19 farmland stations in Shandong Province from 2021 to 2022 and the four most recent sets of matching grid data. The test set is the data of 3 farmland stations in Shandong Province in 2023 and the four most recent sets of matching grid data. The longitude and latitude of the three farmland stations are [117.25, 35.97], [116.64, 36.68], and [119.73, 36.47]. In the training set, these three stations were excluded and did not participate in the training. The first experimental model was iterated a total of five times. After each iteration, the inversion difference and reconstruction difference were output. In each iteration, the model will perform some training and optimization before testing. Figure 3 As shown in the figure, after the first iteration, the inversion difference is 1.36 and the reconstruction difference is 1.29; after the second iteration, the inversion difference is 1.08 and the reconstruction difference is 1.13; after the third iteration, the inversion difference is 0.99 and the reconstruction difference is 1.03; after the fourth iteration, the inversion difference is 0.95 and the reconstruction difference is 0.94; after the fourth iteration, the inversion difference is 0.90 and the reconstruction difference is 0.92. After five iterations, the inversion difference decreases from 1.36 to 0.90, and the reconstruction difference decreases from 1.29 to 0.92. The experimental results show that the inversion and reconstruction effects of the model are improved after each iteration, indicating that in the absence of stations in the area, the model can accurately invert farmland temperature data and provide data support for agricultural production.

[0085] The training set of the holistic site evaluation experiment is the data of 22 farmland stations in Shandong Province from 2021 to 2022 and the matching four sets of grid data, and the test set is the data of 22 farmland stations in Shandong Province in 2023 and the matching four sets of grid data. The second experimental process is the same as the first experiment. Figure 4 As shown in the figure, after the first iteration, the inversion difference is 1.36 and the reconstruction difference is 1.48; after the second iteration, the inversion difference is 1.35 and the reconstruction difference is 1.44; after the third iteration, the inversion difference is 1.18 and the reconstruction difference is 1.25; after the fourth iteration, the inversion difference is 1.16 and the reconstruction difference is 1.15; after the fifth iteration, the inversion difference is 1.21 and the reconstruction difference is 1.11. After five iterations, the inversion difference decreases from 1.36 to 1.21, and the reconstruction difference decreases from 1.48 to 1.11. The experimental results show that the reconstruction and inversion effects of the model are improved after each training, indicating that the model can accurately complete the missing values ​​and abnormal values ​​of existing stations, making the data coherent in time and space and improving the quality of the data.

[0086] An electronic device comprises a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.

[0087] A computer-readable storage medium is used to store computer instructions, and when the computer instructions are executed by a processor, the steps of the above method are implemented.

[0088] The memory in the embodiments of the present application may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory, ROM, a programmable read-only memory, PROM, an erasable programmable read-only memory, EPROM, an electrically erasable programmable read-only memory, EEPROM, or a flash memory. The volatile memory may be a random access memory, RAM, which is used as an external cache. By way of example but not limitation, many forms of RAM are available, such as static random access memory static RAM, SRAM, dynamic random access memory dynamic RAM, DRAM, synchronous dynamic random access memory synchronous DRAM, SDRAM, double data rate synchronous dynamic random access memory double data rate SDRAM, DDR SDRAM, enhanced synchronous dynamic random access memory enhanced SDRAM, ESDRAM, synchronous link dynamic random access memory synchlink DRAM, SLDRAM, and direct memory bus random access memory direct rambus RAM, DR RAM. It should be noted that memory of the methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0089] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented by software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from a website site, computer, server or data center through a wired method such as coaxial cable, optical fiber, digital subscriber line digital subscriber line, DSL or wireless such as infrared, wireless, microwave, etc. to another website site, computer, server or data center. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that includes one or more available media integrated. The available medium can be a magnetic medium such as a floppy disk, a hard disk, a tape, an optical medium such as a high-density digital video disc digital video disc, DVD, or a semiconductor medium such as a solid state hard disk solid state disc, SSD, etc.

[0090] In the implementation process, each step of the above method can be completed by an integrated logic circuit of hardware in a processor or an instruction in the form of software. The steps of the method disclosed in conjunction with the embodiment of the present application can be directly embodied as a hardware processor for execution, or a combination of hardware and software modules in a processor for execution. The software module can be located in a storage medium mature in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in a memory, and the processor reads the information in the memory and completes the steps of the above method in conjunction with its hardware. To avoid repetition, it is not described in detail here.

[0091] It should be noted that the processor in the embodiment of the present application can be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method embodiment can be completed by an integrated logic circuit of hardware in the processor or an instruction in the form of software. The above processor can be a general-purpose processor, a digital signal processor DSP, an application-specific integrated circuit ASIC, a field programmable gate array FPGA or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present application can be implemented or executed. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in the embodiment of the present application can be directly embodied as a hardware decoding processor to perform, or the hardware and software modules in the decoding processor can be combined to perform. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in a memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.

[0092] The above is a detailed introduction to the farmland temperature inversion and reconstruction method based on iterative depth perception machine proposed in the present invention, and the principles and implementation methods of the present invention are explained. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for general technical personnel in this field, according to the idea of ​​the present invention, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A method for inversion and reconstruction of farmland temperature based on iterative deep and shallow perception machine, characterized by: The method specifically comprises the following steps: Step 1: Use historical temperature grid data, farmland temperature data, farmland longitude and latitude coordinates, and farmland altitude information to train the farmland temperature meteorological data inversion model and the temperature meteorological grid data reconstruction model; In step 1, the training method of the farmland temperature meteorological data inversion model and the temperature meteorological grid data reconstruction model is an optimization process based on cyclic iteration; Step 1.1, in the first cycle, the farmland temperature meteorological data inversion model is first trained. The model inputs auxiliary information and original grid information. After the model is trained, the gradient of the inversion model parameters is calculated by the back propagation algorithm, and the optimization algorithm is used to update the parameters so that the loss function is gradually reduced; Step 1.2, then train the temperature meteorological grid data reconstruction model, input auxiliary information and original grid data into the temperature meteorological grid data reconstruction model, calculate the parameter gradient of the reconstruction model through the back propagation algorithm after the reconstruction model is trained, use the optimization algorithm to update the parameters, so that the loss function gradually decreases, and finally output the new four groups of grid information after reconstruction; Step 1.3, the second cycle first trains the farmland temperature meteorological data inversion model. At this time, the input data of the farmland temperature meteorological data inversion model includes auxiliary information and the newly generated grid data after the first cycle of temperature meteorological grid data reconstruction model training; after the farmland temperature meteorological data inversion model training, the parameter gradient is calculated and the parameters are updated to reduce the loss function; Step 1.4, then train the temperature meteorological grid data reconstruction model and output four new sets of grid point data; Step 1.5, continue the optimization cycle according to the second cycle, input the auxiliary information and the new grid information generated by the reconstruction model, and output the inverted farmland temperature data until the loss function of the inversion model no longer decreases, then stop the training; Step 2: connect the farmland temperature meteorological data inversion model and the temperature meteorological grid data reconstruction model trained in step 1 in series; Step 3: Input the original temperature grid point data into the model connected in series in step 2. The data is first input into the temperature meteorological grid data reconstruction model for correction to generate a new set of grid point information. Step 4: The corrected surface temperature data is then input into the farmland temperature meteorological data inversion model. The farmland temperature within the temperature grid data is inverted and calculated using the data reconstructed in step 3 combined with auxiliary information. Finally, the farmland temperature meteorological data inversion model outputs the inverted farmland temperature data.

2. The method according to claim 1, characterized in that: The farmland temperature meteorological data inversion model and the temperature meteorological grid data reconstruction model both adopt the deep and shallow perceptron model structure, and both models include three parts; The first part is a deep neural network with three fully connected layers; the deep neural network inputs auxiliary information; The second part is a wide neural network with a fully connected layer. The wide neural network inputs grid information, which is used to provide high-order cross features between auxiliary information and grid information, and provides feature representation for the output of the model. The third part is the final output layer which contains two fully connected layers; the final output layer converts the feature vectors of the connected deep neural network and wide neural network into the final output value of the model.

3. The method according to claim 2, characterized in that: The input of the temperature meteorological grid data reconstruction model is the original grid point information and auxiliary information, and the output is four sets of new grid point information after reconstruction; The input of the farmland temperature meteorological data inversion model is the original grid point information, the newly generated grid point information and auxiliary information after reconstruction by the temperature meteorological grid data reconstruction model, and the output is a farmland temperature data.

4. The method according to claim 3, characterized in that: In step 1, the performance of the model is measured by calculating the sum of the absolute values ​​of the differences between the model output and the true label through reconstruction difference and inversion difference; Reconstruction difference MAE G Refers to the grid points obtained by reconstructing the temperature meteorological grid data reconstruction model With actual grid The difference between Where N is the number of samples, is the grid point output by the i-th sample model, is the actual grid point of the i-th sample; The inversion difference MAEp refers to the temperature value obtained by inverting the farmland temperature meteorological data inversion model. The actual temperature value The difference between Where N is the number of samples, is the temperature value output by the i-th sample model, is the true temperature of the i-th sample. From the above formula, we can see that MAE G The smaller the MAEp is, the better the model reconstruction and inversion effects are.

5. The method according to claim 4, characterized in that: In step 2, the output of the temperature meteorological grid data reconstruction model is used as the input of the farmland temperature meteorological data inversion model.

6. A farmland temperature inversion and reconstruction system based on iterative depth perception machine, characterized by: The system includes a training module, a reconstruction module and an inversion module: The training module uses historical temperature grid data, farmland temperature data, farmland longitude and latitude coordinates, and farmland altitude information to train a farmland temperature meteorological data inversion model and a temperature meteorological grid data reconstruction model; The training method of the farmland temperature meteorological data inversion model and the temperature meteorological grid data reconstruction model is an optimization process based on cyclic iteration; In the first cycle, the farmland temperature meteorological data inversion model is trained. The model inputs auxiliary information and original grid point information. After the model is trained, the gradient of the inversion model parameters is calculated through the back propagation algorithm, and the optimization algorithm is used to update the parameters so that the loss function is gradually reduced. Then, the temperature and meteorological grid data reconstruction model is trained. The auxiliary information and the original grid point data are input into the temperature and meteorological grid data reconstruction model. After the reconstruction model is trained, the parameter gradient of the reconstruction model is calculated by the back propagation algorithm. The parameters are updated using the optimization algorithm to gradually reduce the loss function. Finally, the four new grid point information after reconstruction is output. The second cycle first trains the farmland temperature meteorological data inversion model. At this time, the input data of the farmland temperature meteorological data inversion model includes auxiliary information and the newly generated grid data after the temperature meteorological grid data reconstruction model training after the first cycle; after the farmland temperature meteorological data inversion model training, the parameter gradient is calculated and the parameters are updated to reduce the loss function; Then the temperature meteorological grid data reconstruction model is trained to output four new sets of grid point data; Continue the optimization cycle according to the second loop, input the auxiliary information and the new grid information generated by the reconstruction model, and output the inverted farmland temperature data until the loss function of the inversion model no longer decreases, then stop the training; Then the trained farmland temperature meteorological data inversion model and the temperature meteorological grid data reconstruction model are connected in series; In the reconstruction module, the original temperature grid point actual data is input into the serial model, and the data is first input into the temperature meteorological grid data reconstruction model for correction to generate a set of new grid point information; In the inversion module, the corrected surface temperature data is input into the farmland temperature meteorological data inversion model, and the reconstructed data is used in combination with auxiliary information to invert and calculate the farmland temperature within the temperature grid data; finally, the farmland temperature meteorological data inversion model outputs the inverted farmland temperature data.

7. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 5 are implemented.

8. A computer-readable storage medium for storing computer instructions, characterized in that: When the computer instructions are executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Images