Vegetation health condition determination method, device and equipment

Abstract

The invention discloses a vegetation health condition determination method, device and equipment, and the method comprises the steps: obtaining remote sensing satellite data of an area where target vegetation is located in a preset period, and the remote sensing satellite data comprises a remote sensing image and meteorological information; determining a first net primary productivity of the target vegetation in a preset period based on the remote sensing satellite data; and determining the health condition of the target vegetation based on the first net primary productivity and a preset first productivity estimation range. According to the invention, the technical problem of low convenience of vegetation health condition monitoring is solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of environmental protection, and in particular relates to a method, device and equipment for determining the health status of vegetation. Background Art

[0002] Environmental pollution and destruction can cause the structure and function of the ecosystem to become unbalanced, leading to a decline in environmental quality and even causing an ecological crisis, directly threatening human survival. At the same time, the health of vegetation is highly correlated with environmental quality, so it is necessary to monitor the health of vegetation to prevent a decline in environmental quality.

[0003] In existing technologies, site observation methods are usually used, which requires researchers to conduct field surveys and collect measured data of different ecosystem types to determine the health status of vegetation.

[0004] However, for large-scale area monitoring, especially global monitoring, in order to obtain reliable vegetation health status, many places need to be visited on the spot and a lot of measured data needs to be collected. However, this also increases the workload of researchers exponentially and is very cumbersome, and the convenience of vegetation health status monitoring is low. Summary of the invention

[0005] The embodiments of the present invention provide a method, device and equipment for determining the health status of vegetation, which solve the technical problem of low convenience in monitoring the health status of vegetation.

[0006] In a first aspect, an embodiment of the present invention provides a method for determining the health status of vegetation, comprising: obtaining remote sensing satellite data of an area where target vegetation is located within a preset period, the remote sensing satellite data including remote sensing images and meteorological information; determining the first net primary productivity of the target vegetation within the preset period based on the remote sensing satellite data; and determining the health status of the target vegetation based on the first net primary productivity and a preset first productivity estimated range.

[0007] In combination with the first aspect of the present invention, in some embodiments, the health status of the target vegetation is determined based on the first net primary productivity and a preset first productivity estimated range, including: obtaining ecological change information of the area where the target vegetation is located within the preset period; determining a second productivity estimated range based on the ecological change information of the area where the target vegetation is located and the first productivity estimated range; and determining the health status of the target vegetation based on the first net primary productivity and the second productivity estimated range.

[0008] In combination with the first aspect of the present invention, in some embodiments, the ecological change information of the area where the target vegetation is located includes the cumulative solar radiation and the soil nutrient content, and the second productivity estimation range is determined based on the ecological change information of the area where the target vegetation is located and the first productivity estimation range, including: based on the cumulative solar radiation and the soil nutrient content of the area where the target vegetation is located, the first productivity estimation range is corrected to obtain the second productivity estimation range.

[0009] In combination with the first aspect of the present invention, in some embodiments, the health status of the target vegetation is determined based on the first net primary productivity and the second productivity estimated range, including: if the first net primary productivity is within the second productivity estimated range, determining that the health status of the target vegetation is normal; if the first net primary productivity is not within the second productivity estimated range, determining that the health status of the target vegetation is abnormal.

[0010] In combination with the first aspect of the present invention, in some embodiments, before determining the health status of the target vegetation based on the first net primary productivity and the preset first productivity estimated range, it also includes: obtaining ecological fixation information of the area where the target vegetation is located; inputting the ecological fixation information of the area where the target vegetation is located into the productivity prediction model to obtain the first productivity estimated range, the ecological fixation information including vegetation type, ecological zoning type and climate zoning type.

[0011] In combination with the first aspect of the present invention, in some embodiments, the productivity prediction model is established through the following steps: obtaining multiple groups of samples, each of the multiple groups of samples including the vegetation type, ecological zone type, climate zone type and actual productivity range of a region; inputting the vegetation type, ecological zone type and climate zone type of each group of samples in the multiple groups of samples into a preset neural network model to obtain a productivity estimation range; determining the loss function of the productivity estimation range and the actual productivity range; based on the loss function of the productivity estimation range and the actual productivity range, iteratively optimizing the neural network model to obtain the productivity prediction model.

[0012] In combination with the first aspect of the present invention, in some embodiments, determining the first net primary productivity of the target vegetation within the preset period based on the remote sensing satellite data includes: determining the second net primary productivity of the target vegetation within the preset period based on the remote sensing satellite data; inputting the second net primary productivity of the target vegetation within the preset period into a preset data denoising model to obtain the first net primary productivity of the target vegetation within the preset period.

[0013] In combination with the first aspect of the present invention, in some embodiments, the data denoising model is established by the following steps: obtaining multiple groups of samples, each of the multiple groups of samples including the altitude, surface temperature, ecological zoning type and climate zoning type of a region, and the noisy net primary productivity and denoised net primary productivity of the region in a period; inputting the altitude, surface temperature, ecological zoning type, climate zoning type and noisy net primary productivity of each group of samples in the multiple groups of samples into the generator of a preset GAN model to obtain an estimated net primary productivity; inputting the estimated net primary productivity and the denoised net primary productivity into the discriminator of the GAN model to obtain the loss function of the estimated net primary productivity and the denoised net primary productivity; based on the loss function of the estimated net primary productivity and the denoised net primary productivity, iteratively optimizing the GAN model to obtain the data denoising model.

[0014] In a second aspect, an embodiment of the present invention provides a device for determining the health status of vegetation, comprising: a data acquisition unit, used to acquire remote sensing satellite data of an area where a target vegetation is located within a preset period, the remote sensing satellite data including remote sensing images and meteorological information; a productivity determination unit, used to determine the first net primary productivity of the target vegetation within the preset period based on the remote sensing satellite data; and a status determination unit, used to determine the health status of the target vegetation based on the first net primary productivity and a preset first productivity estimated range.

[0015] In a third aspect, an embodiment of the present invention provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements any one of the methods described in the first aspect when executing the computer program.

[0016] One or more technical solutions provided by the embodiments of the present invention achieve at least the following technical effects or advantages: The embodiment of the present invention obtains remote sensing satellite data of the area where the target vegetation is located within a preset period, and the remote sensing satellite data includes remote sensing images and meteorological information; based on the remote sensing satellite data, the first net primary productivity of the target vegetation within the preset period is determined; based on the first net primary productivity and the preset first productivity estimation range, the health status of the target vegetation is determined. The first net primary productivity of the target vegetation within the preset period is determined by remote sensing satellite data. The first net primary productivity characterizes the production capacity of the target vegetation in the natural environment, so the health status of the target vegetation can be determined by the first net primary productivity, avoiding the researchers' field investigation to collect measured data, reducing the workload, and thus improving the convenience of monitoring the health status of vegetation. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0018] Figure 1 A flowchart of a method for determining the health status of vegetation according to an embodiment of the present invention; Figure 2 This is a functional module diagram of a device for determining the health status of vegetation in an embodiment of the present invention; Figure 3 Schematic diagram of the structure of an electronic device in an embodiment of the present invention. DETAILED DESCRIPTION

[0019] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions 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.

[0020] In the present invention, descriptions such as "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In addition, the technical solutions between the various embodiments can be combined with each other, but they must be based on the ability of ordinary technicians in the field to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0021] The embodiment of the present invention provides a method for determining the health status of vegetation, referring to Figure 1 As shown, the method includes the following steps S101 to S103: S101: within a preset period, remote sensing satellite data of the target vegetation area is obtained, where the remote sensing satellite data includes remote sensing images and meteorological information.

[0022] It should be noted that the size of the target vegetation area may be 10 square meters, 100 square meters, 1 square kilometer, 100 square kilometers, 1,000 square kilometers or 10,000 square kilometers, etc., and the target vegetation may refer to alpine vegetation, sand forest vegetation, meadow vegetation, desert vegetation, grassland vegetation, wetland vegetation or coniferous forest vegetation, etc. The preset period may refer to 1 day, 1 month, half a year or a year, etc.

[0023] S102: Determine the first net primary productivity of the target vegetation within a preset period based on remote sensing satellite data.

[0024] In some implementations, step S102 may include the following steps S1021-S1022: S1021: Determine the second net primary productivity of the target vegetation within a preset period based on remote sensing satellite data.

[0025] It should be noted that the preset period can be further divided into multiple sub-periods. Assuming that the preset period is one year, the multiple sub-periods can be each month from January to December. Then, when determining the second net primary productivity of the target vegetation within the preset period based on remote sensing satellite data, the remote sensing images and meteorological information can be divided into sub-periods according to 12 months, and the second net primary productivity of the target vegetation in the sub-periods can be obtained respectively. Finally, the second net primary productivity of the target vegetation in one year can be obtained based on the second net primary productivity of the target vegetation in each month. Similarly, the second net primary productivity of the target vegetation in one year can also be directly determined according to the data of one year.

[0026] In some embodiments, step S1021 may include: determining raster data of the area where the target vegetation is located based on remote sensing satellite data, each grid in the raster data represents a different location in the area where the target vegetation is located, and the attribute values ​​of the grid include the first vegetation index, surface moisture index, average temperature, cumulative precipitation and cumulative solar radiation of the location within a preset period; based on the raster data, determining the second net primary productivity of the target vegetation within a preset period.

[0027] It should be noted that the net primary productivity corresponding to the grid can be determined based on the attribute values ​​of the grid. The attribute values ​​of the grid include the first vegetation index, surface moisture index, average temperature, cumulative precipitation and cumulative solar radiation, which achieves a more comprehensive consideration of the factors affecting the net primary productivity, thereby improving the accuracy of the obtained net primary productivity data.

[0028] In some embodiments, based on remote sensing satellite data, the raster data of the target vegetation area is determined, which can be: based on remote sensing images, the first vegetation index and surface moisture index of each grid in the raster data are determined; based on meteorological information, the average temperature, accumulated precipitation and accumulated solar radiation of each grid in the raster data are determined.

[0029] In some embodiments, determining the first vegetation index and surface moisture index of each grid in the raster data based on the remote sensing image can be: obtaining the infrared reflectivity, near-infrared reflectivity and short-wave infrared reflectivity of the remote sensing image; determining the first vegetation index of each grid in the raster data based on the infrared reflectivity and near-infrared reflectivity of the remote sensing image; determining the surface moisture index of each grid in the raster data based on the short-wave infrared reflectivity and near-infrared reflectivity of the remote sensing image.

[0030] In some embodiments, based on the raster data, determining the second net primary productivity of the target vegetation within a preset period can be: taking each grid in the raster data as the target grid in turn; determining the third net primary productivity of the target grid based on the first vegetation index, surface moisture index, average temperature, accumulated precipitation and accumulated solar radiation of the target grid; determining the second net primary productivity of the target vegetation within the preset period based on the third net primary productivity of each grid in the raster data.

[0031] It should be noted that the number of the second net primary productivity can be one value or multiple values. When the number of the second net primary productivity is one value, the sum of the third net primary productivity of each grid in the raster data can be used as the second net primary productivity of the target vegetation within a preset period. When the number of the second net primary productivity is multiple values, the third net primary productivity of each grid in the raster data can be used as the second net primary productivity of the target vegetation within a preset period.

[0032] S1022: Inputting the second net primary productivity of the target vegetation within a preset period into a preset data denoising model to obtain the first net primary productivity of the target vegetation within the preset period.

[0033] It should be noted that the second net primary productivity is the noisy net primary productivity, and the first net primary productivity is the denoised net primary productivity. Since the remote sensing satellite data itself has noise and more or less errors, the second net primary productivity determined based on the remote sensing satellite data also has some errors, that is, the second net primary productivity is the noisy net primary productivity. If the health status of the target vegetation is determined directly based on the second net primary productivity, there may be a situation where the health status detection is wrong due to data errors, and environmental protection cannot be effectively guided. Therefore, the embodiment of the present invention adopts a data denoising model, and the second net primary productivity is processed by the data denoising model to obtain the first net primary productivity, thereby improving the accuracy of the determined net primary productivity, and then improving the accuracy of vegetation health status monitoring.

[0034] In some embodiments, the data denoising model can be established through the following steps: obtaining multiple groups of samples, each of the multiple groups of samples includes the altitude, surface temperature, ecological zone type and climate zone type of a region, and the noisy net primary productivity and denoised net primary productivity of the region in a period; inputting the altitude, surface temperature, ecological zone type, climate zone type and noisy net primary productivity of each group of samples in the multiple groups of samples into the generator of a preset GAN model to obtain the estimated net primary productivity; inputting the estimated net primary productivity and the denoised net primary productivity into the discriminator of the GAN model to obtain the loss function of the estimated net primary productivity and the denoised net primary productivity; based on the loss function of the estimated net primary productivity and the denoised net primary productivity, iteratively optimizing the GAN model to obtain the data denoising model.

[0035] S103: Determine the health status of the target vegetation based on the first net primary productivity and a preset first productivity estimation range.

[0036] In some embodiments, step S103 may include: if the first net primary productivity is within the first productivity estimation range, determining that the health status of the target vegetation is normal; if the first net primary productivity is not within the first productivity estimation range, determining that the health status of the target vegetation is abnormal.

[0037] In some other implementations, step S103 may include the following steps S1031 to S1033: S1031: Within a preset period, obtain ecological change information of the area where the target vegetation is located.

[0038] S1032: Determine a second productivity estimation range based on the ecological change information of the target vegetation area and the first productivity estimation range.

[0039] In some embodiments, the ecological change information of the target vegetation area includes the cumulative solar radiation and the soil nutrient content. Then, step S1032 may include: based on the cumulative solar radiation and the soil nutrient content of the target vegetation area, correcting the first productivity estimation range to obtain a second productivity estimation range.

[0040] It should be noted that the estimated range of the first productivity is determined based on ecological fixed information, which refers to fixed ecological information. The ecological fixed information can determine the approximate range of the net primary productivity of the target vegetation. However, since the ecological environment in which the target vegetation is located is constantly changing, the estimated range of the first productivity determined based on the ecological fixed information will be inaccurate. For example, the greater the cumulative solar radiation, the more photosynthesis the vegetation can carry out, thereby increasing the net primary productivity. Therefore, the embodiment of the present invention limits the correction of the estimated range of the first productivity based on the ecological change information, avoids the wrong estimation of the net primary productivity due to changes in the ecological environment, and achieves a more comprehensive consideration of the factors affecting the net primary productivity, thereby improving the accuracy of the estimated range of the net primary productivity, thereby improving the accuracy of monitoring the health status of the vegetation.

[0041] In some embodiments, based on the cumulative solar radiation and soil nutrient content of the target vegetation area, the first productivity estimation range is corrected to obtain a second productivity estimation range, which can be: comparing the cumulative solar radiation and soil nutrient content of the target vegetation area with the benchmark cumulative solar radiation and benchmark soil nutrient content to obtain a productivity correction; adding the productivity correction and the upper limit value of the first productivity estimation range as the upper limit value of the second productivity estimation range; adding the productivity correction and the lower limit value of the first productivity estimation range as the lower limit value of the second productivity estimation range.

[0042] S1033: Determine the health status of the target vegetation based on the estimated ranges of the first net primary productivity and the second net productivity.

[0043] In some embodiments, step S1033 may include: if the first net primary productivity is within the second productivity estimation range, determining that the health status of the target vegetation is normal; if the first net primary productivity is not within the second productivity estimation range, determining that the health status of the target vegetation is abnormal.

[0044] In some embodiments, before step S103, the method may also include: obtaining ecological fixation information of the area where the target vegetation is located; inputting the ecological fixation information of the area where the target vegetation is located into the productivity prediction model to obtain a first productivity estimation range, wherein the ecological fixation information includes vegetation type, ecological zoning type and climate zoning type.

[0045] In some embodiments, the productivity prediction model can be established through the following steps: obtaining multiple groups of samples, each of the multiple groups of samples includes the vegetation type, ecological zone type, climate zone type and actual productivity range of a region; inputting the vegetation type, ecological zone type and climate zone type of each group of samples in the multiple groups of samples into a preset neural network model to obtain a productivity estimation range; determining the loss function of the productivity estimation range and the actual productivity range; based on the loss function of the productivity estimation range and the actual productivity range, iteratively optimizing the neural network model to obtain a productivity prediction model.

[0046] It should be noted that vegetation is an important component of terrestrial ecosystems, connecting the energy flow and material cycle between the atmosphere and the soil. Net primary productivity (NPP) refers to the organic matter produced by vegetation photosynthesis per unit time and per unit area minus the remaining part of autotrophic respiration. It can reflect the production capacity of vegetation in the natural environment and characterize the quality of terrestrial ecosystems. NPP is the basis for the development of terrestrial ecosystems and an important indicator for evaluating regional ecosystem functions, ecosystem quality and sustainable development. NPP was first proposed by P. Boysen-Jensen in 1995. There are many methods for estimating NPP. Domestic and foreign scholars have conducted research from different angles and disciplines and have achieved many results. From the perspective of spatial scale, it can be divided into NPP site measurement, regional simulation estimation and global simulation estimation. Early scholars mostly used site observation and experimental methods. Based on the NPP ground site observation values, researchers collected measured data of different ecosystem types, and then divided regions according to ecosystem types, using the point-for-surface method to extrapolate and simulate the total NPP of the region and the world. The measured data obtained by these methods are of high accuracy, but they are all sample data, which are only suitable for small-scale measurements and are difficult to reflect the true situation of NPP at the regional scale. It is difficult to obtain NPP measured data at the regional or global scale. With the rapid development of computer technology and remote sensing technology, for regional or global scale NPP estimation, the existing data are generally used to establish model simulation. The use of model methods to estimate regional NPP has become a major research direction at home and abroad. Among them, the more commonly used NPP estimation model is mainly a parameter model, also known as the light use efficiency (LUE) model. From the perspective of resource balance, this model takes into account that vegetation growth tends to be affected by limiting factors. By calculating the solar radiation energy that can be absorbed by vegetation, a parametric equation between solar radiation and the influencing factors that may limit vegetation absorption is established, thereby simulating and estimating NPP. Plant NPP can be estimated by using the photosynthetically active radiation absorbed by plants, which is represented by two factors: photosynthetically active radiation absorbed by vegetation and light energy conversion efficiency. The light energy conversion efficiency is the ratio of the absorption of photosynthetically active radiation affected by limiting factors. Common methods include CASA model, GLO-PEM model, VPM model, etc. With the maturity of remote sensing technology, the technology of simulating and estimating NPP based on remote sensing data has also become more mature. The LUE model integrating remote sensing data mainly uses the index, solar radiation and LUE obtained from remote sensing data to establish the NPP estimation model. The establishment of vegetation NPP estimation model using remote sensing data, environmental factors and ground measured data has gradually become the main research direction for regional estimation of NPP. It can estimate and simulate NPP over a large range, conduct real-time and continuous dynamic monitoring of NPP, and reflect the spatiotemporal distribution characteristics of NPP.

[0047] It should be noted that the methods for determining net primary productivity in the prior art have the following defects: First, there are still some deficiencies in constructing NPP inversion models (such as LUE models) based on remote sensing data. First, it is difficult to have remote sensing data with high spatial resolution and high temporal resolution at the same time, and multi-source data fusion is required. There are inconsistencies in spectral information and space between different data sources. Second, compared with the spatial resolution of remote sensing data (several meters to hundreds of meters), the spatial resolution of environmental factors (such as meteorological data) obtained is lower (10 kilometers or even 1°), and even the coefficient environmental monitoring station data is used for simple spatial interpolation. Moreover, the environmental factor data are mostly reanalysis data inverted by combining models with measured data, which has uncertainty itself. Third, due to the complex mechanism models for simulating the carbon cycle process or photosynthesis process of ecosystems, there is great uncertainty in regional and even global NPP inversion. The above reasons lead to large uncertainty in the NPP inversion mechanism model driven by remote sensing data, low inversion accuracy due to error transmission, and poor model generalization ability. Then, data-driven machine learning inversion methods, such as random forests, decision trees, and deep learning neural network technology, need to train models based on training data, and the training data includes data on factors such as climate and environment related to vegetation growth. Measured NPP data is also required, and measured data are mostly point data that cannot cover the surface, resulting in model inversion results being subject to data quality and its coverage. If there is not enough training data, it cannot be applied to other areas. Moreover, machine learning models are statistical models based on training data, so the models cannot explain the mechanism of ecosystem processes. In addition, FPAR (Fraction of absorbed photosynthetically active radiation) is estimated using the vegetation index NDVI (Normalized Difference Vegetation Index), but there is a nonlinear relationship between NDVI and green vegetation biomass, and saturation is prone to occur when the vegetation biomass is high. Moreover, the vegetation index NDVI itself reflects the health of green leaf vegetation, rather than the photosynthesis of vegetation, resulting in low FPAR estimation accuracy. Secondly, the water stress coefficient ( ) reflects the effect of effective water available to vegetation on LUE. The traditional CASA model calculates It is necessary to use a soil moisture model, which involves many complex soil parameters. Soil parameters are usually difficult to obtain and have poor spatial heterogeneity. The cost data is usually old and cannot be updated in time, making the model difficult to implement. Finally, the resolution of the input meteorological data is limited by the model's requirements for the temporal characteristics of vegetation growth and meteorological environmental elements. Traditional NPP inversion models are mostly based on 250-meter, 500-meter, 1-kilometer or even lower resolution MODIS, AVHRR and other data. These satellite data have high temporal resolution (for example, once a day) and can ensure the acquisition of effective data. Temperature, precipitation and solar radiation are mostly obtained by combining reanalysis data of meteorological satellites, monitoring stations, and physical models, or by using monitoring stations to obtain sunshine hours data by establishing an empirical model with solar radiation, and then through spatial interpolation. The spatial resolution of the input data of these models is low, the spatial heterogeneity is poor, and they cannot represent the refined distribution of surface ecosystems. The mixed pixel problem leads to low accuracy and large errors in the inverted NPP data.

[0048] It should be noted that, regarding the method for determining net primary productivity, the following describes the improvement measures of the embodiment of the present invention: The embodiment of the present invention is a coupling model combining the LUE model and deep learning technology, which not only has the respective advantages of the ecological process carbon cycle mechanism model and the deep learning technology, but also improves the inversion accuracy of the traditional parameter model for NPP, wherein the LUE model of the embodiment of the present invention also improves the existing CASA model. The embodiment of the present invention solves the uncertainty problem caused by the inconsistency of the remote sensing inversion mechanism model due to the inconsistency of data sources, spatiotemporal resolution, etc., and constructs a NPP inversion coupling model combining the remote sensing mechanism model and the deep learning technology to solve the problem of rapid automatic inversion of NPP without measured sample data as training data. The remote sensing mechanism model adopts the LUE model, and the deep learning model adopts the GAN (Generative Adversarial Networks, Generative Adversarial Network) model. The coupling model of the present invention integrates the interpretability of the ecological process mechanism of the remote sensing data-driven mechanism model and the quantifiability of the driving factor, as well as the generalization ability of the deep learning technology. Compared with a single mechanism model, it can improve the inversion accuracy and is suitable for regional and even global NPP inversion. The LUE model has also been optimized, mainly to solve the shortcomings of the existing LUE: the model simulation accuracy is not high due to spectral saturation, the calculation of key parameters is too complicated and the required data is not easy to obtain, the spatial resolution of the input data does not match, and the resolution of meteorological data is low compared to remote sensing index products. The three key points optimized by the embodiment of the present invention are: the calculation of FPAR, the use of kNDVI to replace the original NDVI, the use of the Land Surface Water Index (LSWI) to calculate the water stress coefficient, replacing the original model using radiation dryness and soil data to calculate the water stress coefficient through inversion evaporation, and the fusion downscaling of meteorological data to improve the spatial resolution.

[0049] The following is an explanation of the calculation method of the third net primary productivity of the target grid, assuming that the preset period is one year and the multiple sub-periods include each month from January to December: Among them, NPP(x,t) is the third net primary productivity of the target grid x in month t, APAR(x,t) is the photosynthetically active radiation absorbed by the target grid x in month t, represents the actual light energy utilization rate of the target grid x in month t, PAR(x,t) is the total photosynthetically active radiation of the target grid x in month t, and FPAR(x,t) is the absorption ratio of the total incident photosynthetically active radiation of the vegetation of the target grid x in month t. and It is used to characterize the stress effect of low temperature and high temperature on light energy utilization and reflect the influence of temperature on light energy utilization. is the water stress influence coefficient, reflecting the effect of water conditions on light energy utilization. It is the maximum light energy utilization under ideal conditions.

[0050] It should be noted that in formula (2), for the calculation of FPAR(x, t), FPAR can be calculated by the normalized difference vegetation index (NDVI): NDVI (x,t), and then FPAR NDVI (x, t) as FPAR(x, t). In addition, FPAR can also be calculated using the improved normalized difference vegetation index (kNDVI) kNDVI (x,t), and then FPAR kNDVI (x, t) is used as FPAR(x, t), and the first vegetation index can refer to NDVI or kNDVI. Referring to formulas (4) to (8), the following is the FPAR kNDVI The calculation of (x, t) is further explained: In formula (4), kNDVI (i,max) and kNDVI (i,min) The maximum and minimum values ​​of kNDVI for the i-th vegetation type, respectively, can refer to the 5% and 95% percentiles. kNDVI(x,t) is the improved normalized difference vegetation index of the target grid x in month t, FPAR max is the preset maximum absorption ratio, FPAR min is the preset minimum absorption ratio. In formula (5) and formula (6), k is the kernel function, n and r represent the NIR near infrared band and RED infrared band respectively. The kernel function k is used to measure the similarity between the two band channels. The parameter The concept of the distance between the near-infrared band and the infrared band is controlled, and the value can be 1. Then the kernel function k can be derived into a simple expression form, that is, formula (7). The more commonly used kernel function in theory and practice is the radial basis function (RBF), which takes the length scale parameter The value is the average distance between the near-infrared band and the infrared band, that is , then, we can further derive formula (8). Formula (8) is the kernel function of NDVI, which enables kNDVI to be adaptively stretched pixel by pixel and ensures that all moments of the relationship between the near-infrared band and the infrared band are taken into account. This also enables kNDVI to cope with saturation effects, complex phenological cycles and seasonal changes, and handle mixed pixel problems. Compared with other indices, it transmits lower uncertainty.

[0051] In formula (3), The calculation is further explained: in, and They represent the near infrared reflectivity and shortwave infrared reflectivity in remote sensing images respectively. LSWI(x,t) is the surface moisture index of the target grid x in month t. max (x) represents the maximum value of LSWI in the vegetation growth period (set as one year in this study) corresponding to a single grid, P(x,t) is the cumulative precipitation of the target grid x in time t, P max (x) is the maximum monthly cumulative precipitation in the growing season (assuming one year). Although LSWI and formula (11) can reflect the changes in water and heat conditions of the ecosystem, considering the important role of precipitation on vegetation productivity, the purpose of formula (12) is to add the factor of precipitation to improve the accuracy of the model. It should be noted that As the effective water content in the environment increases, it gradually increases, and its value ranges from 0.5 (under extremely dry conditions) to 1 (under very humid conditions). Using the remote sensing inversion LSWI (Land Surface Water Index) method to estimate the dynamic changes of the water stress coefficient can simplify the complexity of calculating the water stress coefficient in the traditional model, and the satellite remote sensing data for calculating LSWI is easier to obtain. LSWI is an indicator that describes the water content of vegetation leaves.

[0052] The following is the formula (3) and The calculation is further explained: Among them, T opt (x) is the most suitable temperature (℃), that is, the average temperature of the month when the NDVI value of a certain area reaches the highest in a year. When the average temperature of a month is less than or equal to -10℃, Take 0. It is T opt (x), its variation is spatial but not temporal, and its value is between 0.8 (T opt (x) is 0℃ or 40℃) to 1 (T opt (x) is between 20℃. Indicates that the ambient temperature changes from the most suitable temperature T opt When (x) changes to high and low temperatures, the photosynthetic utilization rate of plants gradually decreases. T(x, t) is the average temperature of the target grid x in month t. When the average temperature T(x, t) of a certain month is higher than the optimal temperature T opt (x) When the temperature is 10℃ higher or 13℃ lower, the The value is equal to the average temperature T(x,t) is the optimum temperature T opt (x) hours half of its value.

[0053] It should be noted that the embodiment of the present invention uses the geographical weighted regression (GWR) model to downscale and fuse the temperature, precipitation and solar radiation. As an extension of the traditional linear regression model, the GWR model embeds the spatial location information of the data into the regression parameters, and evaluates the changes in the relationship between the independent variable and the dependent variable in the spatial scale by obtaining local parameters. It is a local statistical model that can effectively detect the non-stationarity and spatial dependence of spatial data. Using the GWR model for downscaling and fusion processing can obtain high-resolution model input parameters: temperature, precipitation and solar radiation, thereby improving the model accuracy and being able to characterize finer spatial heterogeneity. The GWR model is explained below: Among them, formula (15) is the basic form of the GWR model, X ik is the predictor (independent) variable, Y i is the dependent variable, (u i , v i ) is the coordinate of the i-th (i=1,2,3…,n) sampling point, β k (u i , v i ) is the kth local regression parameter of the ith sampling point, which is a function of the geographical location, β0(u i , v i )and are the regression constant and error term of the model at point i, respectively, and p is the number of independent variables X. Since the regression parameters in the GWR model vary with the spatial location, according to the weighted least squares method, the regression parameters of point i can be estimated by minimizing equation (16), where w ij is the geographical distance d between regression point i and other observation points j ij A monotonically decreasing function of the spatial weight function w ij It is the core of the GWR model. The embodiment of the present invention selects a universal Gaussian function as the spatial weight function, and its basic form is shown in formula (17). In formula (17), d ijis the distance between regression point i and other observation points j, and b is the bandwidth parameter. Due to the uneven distribution of sampling data points, the embodiment of the present invention adopts an adaptive bandwidth selection strategy and the Corrected Akaike Information Criterion (AICc) to determine the optimal bandwidth, and then the weight matrix can be obtained by the weight function to obtain the estimated local regression coefficient.

[0054] It should be noted that the LUE-GAN coupled inversion model based on deep learning technology proposed in the embodiment of the present invention not only realizes the theoretical or logical mathematical expression of NPP, but also gives play to the advantages of deep learning and improves the generalization ability of the model. The coupling model uses the inversion results of the LUE model driven by remote sensing data as the input (training data) of the GAN model, which not only overcomes the shortcomings of the traditional mechanism model being too complex and the error transmission caused by the uncertainty of the input data, but also avoids the disadvantage of directly using a deep learning model and requiring a large amount of sample data for learning because it does not require a large amount of measured NPP as real data for training. Generative adversarial network GAN is a deep learning method proposed by Goodfellow et al., mainly including a generator G and a discriminator D, which are used to capture the distribution characteristics of data and estimate the true probability of the input image, respectively. Among them, the generator G takes the variable z as input, which is random noise data. The task of the generator G is to generate realistic data as much as possible to deceive the discriminator, that is, an output G (z) obtained by training is close to the real data x. The input of the discriminator D is the real data x or the generated sample G(z). The task of the discriminator D is to distinguish the real data from the generated data as much as possible. The two play a game of maximum and minimum until the discriminator D cannot distinguish the real and generated data, that is, a dynamic balance is reached. The embodiment of the present invention uses a convolutional neural network CNN as a generator and a discriminator for deep learning. The convolutional neural network CNN is a deep learning method based on convolution and is also an important research branch in the field of neural networks. It provides an end-to-end learning model. Since the local area of ​​the previous layer is excited by the convolution kernel with shared weights to obtain the features of the next layer, the number of connections and the network parameters that need to be trained are greatly reduced. CNN captures the spatial correlation of image data and extracts spatial features through deep learning. The basic structure of CNN includes five components: input layer, convolution layer, pooling layer, fully connected layer and output layer. Different combinations of each part can generate a variety of CNN models. If the input is from multiple sources, the data needs to be normalized first. The convolution layer is the core layer of CNN, which is composed of multiple convolution kernels of different numbers and sizes. The convolution kernel has three dimensions, namely the length, width and depth of the convolution kernel. The length and width can be set manually, and the depth is generally the number of input data channels. The input data and the convolution kernel are convoluted to reduce noise, enhance and extract spatial features. In order to improve the fitting ability and nonlinear expression ability of the neural network, the convolution result is mapped using the activation function after extracting local features through convolution calculation to obtain the output.

[0055] Taking the high-resolution satellite SENtinel-2 and / or LANDSAT series multispectral data, inverted monthly and annual NPP products as examples, the calculation of net primary productivity is illustrated as follows: the first step is to retrieve all the Sentinel-2A / B and / or LANDSAT series long-term multispectral data that have passed through the remote sensing data cloud platform in previous years, and perform orthorectification, atmospheric correction, and cloud mask removal on each image to obtain surface reflectance data, and extract infrared RED, near infrared NIR, and shortwave infrared SWIR bands. The second step is to resample the bands of different resolutions of the same satellite to make their resolutions consistent (for example, the RED and NIR bands of the Sentinel-2 satellite have a resolution of 10 meters, and need to be resampled to 20 meters to be consistent with SWIR); calculate the NDVI and LSWI of each image, NDVI = (NIR-RED) / (NIR+RED), LSWI = (NIR-SWIR) / (NIR+SWIR). The third step is to resample the NDVI and LSWI of all satellite images every month of each year to keep the spatial resolution consistent, and then perform monthly maximum synthesis to generate monthly NDVI and LSWI data. The fourth step is to obtain low-resolution (such as ERA5-Land's 0.1°*0.1° data) raster data of monthly average temperature, cumulative precipitation, and cumulative solar radiation, and use the GWR model to perform downscaling and fusion processing to obtain high-resolution (such as 20-meter resolution) monthly average temperature, cumulative precipitation, and cumulative solar radiation raster data consistent with the resolution in the third step. The fifth step is to calculate the absorption ratio FPAR of the total photosynthetically active radiation for a certain month of a certain year (the same below), and you can refer to formulas (4) to (8). The sixth step is to calculate the total photosynthetically active radiation PAR. The calculation method can be: PAR=0.5*SOL, where SOL is the cumulative solar radiation obtained in the fourth step. The seventh step is to calculate the water stress coefficient , refer to formulas (9) to (12). Step 8: Calculate the actual light energy utilization rate , refer to formulas (9) to (14). The ninth step, according to the total photosynthetically active radiation PAR(x, t) and absorption ratio FPAR(x, t) obtained by the above calculation, calculate the photosynthetically active radiation absorption APAR(x, t), refer to formula (2). The tenth step, according to formula (1), calculate the third net primary productivity NPP(x, t) of the target grid x in month t. The eleventh step, for different grid data, repeat the above steps 5 to 10 to obtain the long-term monthly NPP of the past years. ms Data. Step 12, NPP msThe data and DEM (Digital Elevation Model), surface temperature, surface cover, ecological zoning and climate zoning are used as input data and input into the GAN model to output the monthly long-term NPP. my Step 13: Based on the annual NPP calculated in step 12, the annual NPP data for each year is obtained by summing up the NPP of each month.

[0056] The specific details of the fourth step are explained in detail below: Step 1. Select high-resolution surface cover and DEM data that are consistent with the resolution of NDVI and LSWI as much as possible, extract slope and aspect based on DEM data, otherwise resample to the same resolution; Step 2. Obtain low-resolution monthly 2m temperature, precipitation, solar radiation, and 10-meter high V and U direction wind speed, surface atmospheric pressure, cloud coverage, solar radiation, air humidity, dew point temperature, atmospheric water vapor content and other raster data (such as ERA5, ERA5-Land); Step 3. Resample the NDVI, LSWI in the third step and the surface cover, DEM, slope and aspect data in step 1 to keep the resolution consistent with the low-resolution simulation variables; Step 4. Use monthly temperature, precipitation, and solar radiation as dependent variables, and all other data mentioned above as dependent variables. The data are used as predictive variables (independent variables), and the correlation analysis method is used to screen out predictive variables with strong correlation with each dependent variable; Step 5, the principal component analysis method is used to convert all predictive variables into principal component variables to eliminate the collinearity problem of the original predictive variables; Step 6, the best subset regression method is used to screen all predictive variables and select the best predictive variable subset to solve the problem of model overfitting caused by too many dependent variables; Step 7, using the best predictive variables screened in Step 6, the GWR model is constructed to obtain the local regression parameters, regression constants and error terms of the corresponding dependent variables; Step 8, the low-resolution predictive variables obtained in Step 2 are resampled to keep the same resolution as the predictive variables in Step 1, and the high-resolution predictive variables and the GWR model are used to solve the high-resolution simulation variables (temperature, precipitation and solar radiation).

[0057] The following is a detailed description of the details of step 12: Step 1, NPP ms The data is used as the preliminary inversion result and the auxiliary data an (DEM, surface temperature, ecological zoning and climate zoning data) are used as preliminary training data to be input into the generator G module of GAN for training. Generator G is a convolutional neural network. The output of G is G (NPP mz ); Step 2, G (NPP mz ) and other existing NPP monthly products NPP xAs the input to the GAN discriminator D, the discriminator D is also a convolutional neural network, which outputs the discriminant result D (NPP) based on the minimum loss function. my ); Step 3, D (NPP my ) and the NPP of step 1 ms and data an Input them together to G to continue training, and G outputs the result G (NPP mz ); Step 4, repeat steps 2 and 3 until the generator G and the discriminator D reach a balance, and the output result of D (NPP my ) is the inverted real monthly NPP data NPP my .

[0058] The embodiment of the present invention obtains remote sensing satellite data of the area where the target vegetation is located within a preset period, and the remote sensing satellite data includes remote sensing images and meteorological information; based on the remote sensing satellite data, the first net primary productivity of the target vegetation within the preset period is determined; based on the first net primary productivity and the preset first productivity estimation range, the health status of the target vegetation is determined. The first net primary productivity of the target vegetation within the preset period is determined by remote sensing satellite data. The first net primary productivity characterizes the production capacity of the target vegetation in the natural environment, so the health status of the target vegetation can be determined by the first net primary productivity, avoiding the researchers' field investigation to collect measured data, reducing the workload, and thus improving the convenience of monitoring the health status of vegetation.

[0059] Based on the same inventive concept, Figure 2 As shown, an embodiment of the present invention provides a device 10 for determining the health status of vegetation, including: a data acquisition unit 110, used to acquire remote sensing satellite data of the area where the target vegetation is located within a preset period, the remote sensing satellite data including remote sensing images and meteorological information; a productivity determination unit 120, used to determine the first net primary productivity of the target vegetation within a preset period based on the remote sensing satellite data; a status determination unit 130, used to determine the health status of the target vegetation based on the first net primary productivity and a preset first productivity estimation range.

[0060] It can be understood that the status determination unit 130 includes: an information acquisition subunit, which is used to obtain ecological change information of the area where the target vegetation is located within a preset period; a correction subunit, which is used to determine the second productivity estimation range based on the ecological change information of the area where the target vegetation is located and the first productivity estimation range; and a status determination subunit, which is used to determine the health status of the target vegetation based on the first net primary productivity and the second productivity estimation range.

[0061] It can be understood that the ecological change information of the target vegetation area includes the cumulative solar radiation and the soil nutrient content. The correction subunit is specifically used to correct the first productivity estimation range based on the cumulative solar radiation and the soil nutrient content of the target vegetation area to obtain the second productivity estimation range.

[0062] It can be understood that the condition determination subunit is specifically used to: if the first net primary productivity is within the second productivity estimation range, determine the health status of the target vegetation is normal; if the first net primary productivity is not within the second productivity estimation range, determine the health status of the target vegetation is abnormal.

[0063] It can be understood that the device 10 for determining the health status of vegetation also includes: a range determination unit, which is used to obtain ecological fixed information of the area where the target vegetation is located before determining the health status of the target vegetation based on the first net primary productivity and the preset first productivity estimated range; the ecological fixed information of the area where the target vegetation is located is input into the productivity prediction model to obtain the first productivity estimated range, and the ecological fixed information includes vegetation type, ecological zoning type and climate zoning type. Among them, the productivity prediction model is established by the following steps: obtaining multiple groups of samples, each of the multiple groups of samples includes the vegetation type, ecological zoning type, climate zoning type and actual productivity range of a region; the vegetation type, ecological zoning type and climate zoning type of each group of samples in the multiple groups of samples are input into the preset neural network model to obtain the productivity estimated range; determine the loss function of the productivity estimated range and the actual productivity range; based on the loss function of the productivity estimated range and the actual productivity range, iteratively optimize the neural network model to obtain the productivity prediction model.

[0064] It can be understood that the productivity determination unit 120 is specifically used to: determine the second net primary productivity of the target vegetation within a preset period based on remote sensing satellite data; input the second net primary productivity of the target vegetation within the preset period into a preset data denoising model to obtain the first net primary productivity of the target vegetation within the preset period. Among them, the data denoising model is established through the following steps: obtaining multiple groups of samples, each of the multiple groups of samples includes the altitude, surface temperature, ecological zone type and climate zone type of a region, as well as the noisy net primary productivity and denoised net primary productivity of the region in a period; the altitude, surface temperature, ecological zone type, climate zone type and noisy net primary productivity of each group of samples in the multiple groups of samples are input into the generator of the preset GAN model to obtain the estimated net primary productivity; the estimated net primary productivity and denoised net primary productivity are input into the discriminator of the GAN model to obtain the loss function of the estimated net primary productivity and the denoised net primary productivity; based on the loss function of the estimated net primary productivity and the denoised net primary productivity, the GAN model is iteratively optimized to obtain the data denoising model.

[0065] It should be understood that more implementation details of the apparatus 10 for determining the health status of vegetation in the embodiment of the present invention can be found in the aforementioned method for determining the health status of vegetation, and will not be described in detail here for the sake of brevity.

[0066] Based on the same inventive concept, an embodiment of the present invention further provides an electronic device, such as Figure 3 As shown, it includes a memory 304, a processor 302, and a computer program stored in the memory 304 and executable on the processor 302. The processor 302 executes the program to implement the steps described in any implementation of the method for determining the health status of vegetation.

[0067] Among them, Figure 3 In the embodiment of the present invention, a bus architecture (represented by bus 300) is shown, and bus 300 may include any number of interconnected buses and bridges, and bus 300 links various circuits including one or more processors represented by processor 302 and memory represented by memory 304. Bus 300 may also link various other circuits such as peripherals, voltage regulators, and power management circuits, which are well known in the art and are therefore not further described herein. Bus interface 305 provides an interface between bus 300 and receiver 301 and transmitter 303. Receiver 301 and transmitter 303 may be the same element, namely a transceiver, providing a unit for communicating with various other devices on a transmission medium. Processor 302 is responsible for managing bus 300 and general processing, while memory 304 may be used to store data used by processor 302 when performing operations.

[0068] The functions described herein may be implemented in hardware, software executed by a processor, firmware, or any combination thereof. If implemented in software executed by a processor, the functions may be stored as one or more instructions or codes on a computer-readable medium or transmitted via a computer-readable medium. Other examples and implementations are within the scope and spirit of the present invention and the appended claims. For example, due to the nature of software, the functions described above may be implemented using software executed by a processor, hardware, firmware, hard wiring, or a combination of any of these. In addition, each functional unit may be integrated into a processing unit, each unit may exist physically separately, or two or more units may be integrated into one unit.

[0069] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0070] The units described as separate components may or may not be physically separated, and the components of the control device may or may not be physical units, that is, they may be located in one place or distributed in multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0071] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk and other media that can store program codes.

[0072] The above description is only an embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent substitution, improvement, etc. made within the spirit and principle of the present invention shall be included in the scope of the claims of the present invention.

Claims

1. A method for determining the health status of vegetation, characterized in that: include: Acquiring remote sensing satellite data of the target vegetation area within a preset period, wherein the remote sensing satellite data includes remote sensing images and meteorological information; Determining the first net primary productivity of the target vegetation within the preset period based on the remote sensing satellite data; The health status of the target vegetation is determined based on the first net primary productivity and a preset first productivity estimate range.

2. The method for determining the health status of vegetation according to claim 1, characterized in that: The determining the health status of the target vegetation based on the first net primary productivity and a preset first productivity estimation range includes: Acquiring ecological change information of the area where the target vegetation is located within the preset period; Determining a second productivity estimation range based on ecological change information of the target vegetation region and the first productivity estimation range; The health status of the target vegetation is determined based on the first net primary productivity and the second productivity estimate range.

3. The method for determining the health status of vegetation according to claim 2, characterized in that: The ecological change information of the target vegetation region includes cumulative solar radiation and soil nutrient content, and determining the second productivity estimation range based on the ecological change information of the target vegetation region and the first productivity estimation range includes: Based on the accumulated solar radiation and soil nutrient content in the area where the target vegetation is located, the first productivity estimation range is corrected to obtain the second productivity estimation range.

4. The method for determining the health status of vegetation according to claim 2, characterized in that: The determining the health status of the target vegetation based on the first net primary productivity and the second productivity estimated range includes: If the first net primary productivity is within the second productivity estimation range, determining that the health status of the target vegetation is normal; If the first net primary productivity is not within the second productivity estimation range, the health status of the target vegetation is determined to be abnormal.

5. The method for determining the health status of vegetation according to claim 1, characterized in that: Before determining the health status of the target vegetation based on the first net primary productivity and the preset first productivity estimation range, the method further includes: Obtaining ecological fixation information of the area where the target vegetation is located; The ecological fixation information of the target vegetation area is input into the productivity prediction model to obtain the first productivity estimation range, wherein the ecological fixation information includes vegetation type, ecological zoning type and climate zoning type.

6. The method for determining the health status of vegetation according to claim 5, characterized in that: The productivity prediction model is established through the following steps: Acquire multiple sets of samples, each set of samples in the multiple sets of samples includes vegetation type, ecological zone type, climate zone type and actual range of productivity in a region; Inputting the vegetation type, ecological zone type and climate zone type of each group of samples in the plurality of groups of samples into a preset neural network model to obtain a productivity estimation range; Determining a loss function for the productivity estimate range and the productivity actual range; Based on the loss function of the productivity estimation range and the actual productivity range, the neural network model is iteratively optimized to obtain the productivity prediction model.

7. The method for determining the health status of vegetation according to claim 1, characterized in that: The step of determining the first net primary productivity of the target vegetation within the preset period based on the remote sensing satellite data comprises: Determining the second net primary productivity of the target vegetation within the preset period based on the remote sensing satellite data; The second net primary productivity of the target vegetation within the preset period is input into a preset data denoising model to obtain the first net primary productivity of the target vegetation within the preset period.

8. The method for determining the health status of vegetation according to claim 7, characterized in that: The data denoising model is established through the following steps: Acquire multiple sets of samples, each set of samples in the multiple sets of samples includes the altitude, surface temperature, ecological zone type and climate zone type of a region, and the noisy net primary productivity and the noisy net primary productivity of the region in a period; Inputting the altitude, surface temperature, ecological zone type, climate zone type and noisy net primary productivity of each group of samples in the plurality of groups of samples into a generator of a preset GAN model to obtain an estimated net primary productivity; Inputting the estimated net primary productivity and the denoised net primary productivity into the discriminator of the GAN model to obtain a loss function of the estimated net primary productivity and the denoised net primary productivity; Based on the loss function of the estimated net primary productivity and the denoised net primary productivity, the GAN model is iteratively optimized to obtain the data denoising model.

9. A device for determining the health status of vegetation, characterized in that: include: A data acquisition unit, used to acquire remote sensing satellite data of the target vegetation area within a preset period, wherein the remote sensing satellite data includes remote sensing images and meteorological information; A productivity determination unit, used to determine the first net primary productivity of the target vegetation within the preset period based on the remote sensing satellite data; A condition determination unit is used to determine the health status of the target vegetation based on the first net primary productivity and a preset first productivity estimation range.

10. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method according to any one of claims 1 to 8 when executing the computer program.

Images