A deep learning remote sensing monitoring vegetation data management method and system

Abstract

This application proposes a deep learning remote sensing monitoring vegetation data management method and system, which belongs to the field of data management. Among them, the method first conducts vegetation remote sensing monitoring of the target area to obtain monitoring data and pixel size; then collects time and terrain information to identify the distribution of terrain-obstructed shadow areas and their accuracy coefficients; then collects meteorological information to identify the distribution of drought stress areas and their accuracy coefficients; then, based on the pixel size, screens and calculates shadow areas and drought areas that are smaller than the pixel size to achieve accuracy identification and management of remote sensing monitoring data. This application realizes the precise evaluation and management of the accuracy of remote sensing monitoring vegetation data by identifying and quantifying the impact of shadow areas and drought stress areas that are smaller than the pixel size on vegetation data, provides data users with a basis for judging the authenticity of the data, and improves the scientificity and reliability of analysis and decision-making based on remote sensing monitoring data.

Description

Technical Field

[0001] The present invention relates to the field of data management, and in particular to a deep learning-based remote sensing monitoring vegetation data management method and system. Background Art

[0002] Remote sensing monitoring technology is widely used in vegetation surveys and monitoring, enabling rapid acquisition of data on vegetation distribution, coverage, and growth status over a wide area. With the advancement of remote sensing technology, a variety of sensors aboard platforms such as satellites and drones can capture remote sensing images at different wavelengths and resolutions, providing a rich data source for vegetation monitoring.

[0003] However, remote sensing images have inherent spatial resolution limitations. Each pixel represents the average information within a certain area on the ground, usually ranging from hundreds to thousands of meters. At this resolution, factors such as shadows caused by terrain undulations and drought stress caused by meteorological conditions will affect the actual distribution of vegetation. In particular, when these affected areas are smaller than the pixel size, they are more difficult to accurately identify and express in remote sensing images. At present, the processing methods for remote sensing vegetation monitoring data mainly focus on removing macro-interference factors such as cloud removal and atmospheric correction. There is a lack of effective assessment of the impact of fine-scale terrain shadows and drought stress on data accuracy. This lack of assessment makes it impossible to accurately measure the accuracy of vegetation data, and thus cannot provide users with a basis for judging the authenticity of the data, affecting the scientific nature and reliability of analytical decisions based on these data. Summary of the Invention

[0004] The present invention aims to solve the technical problem that the data accuracy of remote sensing monitoring vegetation data in the prior art cannot be accurately measured, and provides a deep learning remote sensing monitoring vegetation data management method and system to solve it.

[0005] The technical solution of the present invention to solve the above technical problems is as follows:

[0006] In a first aspect, the present invention provides a deep learning remote sensing monitoring vegetation data management method, comprising: performing vegetation remote sensing monitoring on a target area, collecting and acquiring remote sensing monitoring data, and obtaining the pixel size of the remote sensing monitoring; collecting time information and terrain information within the target area, performing terrain occlusion shadow identification on the target area, obtaining the shadow area distribution, and obtaining the shadow identification accuracy coefficient; collecting meteorological information within the target area, performing drought stress area identification, obtaining the drought stress area distribution, and obtaining the drought identification accuracy coefficient; based on the shadow area distribution, shadow identification accuracy coefficient, drought stress area distribution, drought identification accuracy coefficient, combined with the pixel size, performing accuracy identification of the remote sensing monitoring data, and performing data management.

[0007] In a second aspect, the present invention provides a deep learning remote sensing monitoring vegetation data management system, comprising: a vegetation monitoring module, used to perform vegetation remote sensing monitoring on a target area, collect and acquire remote sensing monitoring data, and obtain the pixel size of remote sensing monitoring; a terrain shadow recognition module, used to collect time information and terrain information within the target area, perform terrain occlusion shadow recognition on the target area, obtain shadow area distribution, and obtain a shadow recognition accuracy coefficient; a drought stress recognition module, used to collect meteorological information within the target area, perform drought stress area recognition, obtain drought stress area distribution, and obtain a drought recognition accuracy coefficient; a data management module, used to identify the accuracy of the remote sensing monitoring data and perform data management based on the shadow area distribution, shadow recognition accuracy coefficient, drought stress area distribution, and drought recognition accuracy coefficient, combined with the pixel size.

[0008] The beneficial effects of the present invention are:

[0009] Vegetation remote sensing monitoring is conducted in the target area, with remote sensing monitoring data and pixel size collected to provide a benchmark for subsequent assessment of small-scale influencing factors. Temporal and topographic information within the target area is collected to identify terrain shadows, obtain the distribution of shadow areas, and determine the shadow identification accuracy coefficient. This provides key parameters for subsequent accuracy assessment, thereby quantifying the impact of shadow areas, particularly those smaller than the pixel size, on vegetation data accuracy. Meteorological information within the target area is collected to identify drought-stressed areas, obtain their distribution, and determine the drought identification accuracy coefficient. This addresses the issue of quantifying the impact of drought areas, particularly those smaller than the pixel size and without surrounding green vegetation, on vegetation data accuracy. Based on the shadow area distribution, shadow identification accuracy coefficient, drought stress area distribution, and drought identification accuracy coefficient, combined with pixel size, remote sensing monitoring data accuracy is identified and managed, addressing the difficulty in accurately measuring vegetation remote sensing data accuracy and providing data users with a basis for assessing data authenticity.

[0010] Through the above technical solution, this application achieves an effective solution to the accuracy problem of remote sensing monitoring vegetation data caused by fine-scale terrain shadows and drought stress, and achieves the technical effect of accuracy identification and precise management of remote sensing monitoring vegetation data. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 A flowchart of a deep learning remote sensing monitoring vegetation data management method provided by the present invention;

[0012] Figure 2 This is a structural diagram of a deep learning remote sensing monitoring vegetation data management system provided by the present invention.

[0013] In the accompanying drawings, the components represented by the reference numerals are as follows:

[0014] Vegetation monitoring module 11, terrain shadow recognition module 12, drought stress recognition module 13, data management module 14. DETAILED DESCRIPTION

[0015] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. 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 those skilled in the art without making any creative efforts shall fall within the scope of protection of the present invention.

[0016] In the description of the present invention, the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of the technical features indicated. Therefore, a feature specified as "first" or "second" may explicitly or implicitly include one or more of the specified features. In the description of the present invention, "plurality" means two or more, unless otherwise specifically defined.

[0017] In the description of the present invention, the term "for example" is used to mean "used as an example, illustration or illustration". Any embodiment of the present invention described as "for example" is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is given to enable any person skilled in the art to implement and use the present invention. In the following description, details are listed for the purpose of explanation. It should be understood that a person of ordinary skill in the art can recognize that the present invention can be implemented without using these specific details. In other examples, well-known structures and processes are not elaborated in detail to avoid obscuring the description of the present invention with unnecessary details. Therefore, the present invention is not intended to be limited to the embodiments shown, but is consistent with the widest scope consistent with the principles and features disclosed herein.

[0018] Example 1, as Figure 1 As shown, an embodiment of the present invention provides a deep learning remote sensing monitoring vegetation data management method, including:

[0019] S1. Conduct vegetation remote sensing monitoring in the target area, collect remote sensing monitoring data, and obtain the pixel size of remote sensing monitoring.

[0020] Specifically, satellite or aerial remote sensing is first used to acquire vegetation information and remote sensing monitoring data from pre-determined target areas. The target area refers to the specific geographic region where vegetation monitoring and data management are required. This can be areas with vegetation cover, such as nature reserves, agricultural areas, forests, or urban green spaces. Specifically, multispectral remote sensors, hyperspectral remote sensors, or synthetic aperture radars can be used to collect electromagnetic wave reflection or radiation information from the target area, generating remote sensing monitoring data containing vegetation information. This remote sensing monitoring data includes, but is not limited to, vegetation indices (such as NDVI and EVI), leaf area index, biomass distribution, and other parameters reflecting vegetation growth.

[0021] Next, obtain the resolution parameters of the remote sensing monitoring equipment used and use them to calculate the pixel size of the remote sensing monitoring image. Pixel size refers to the actual ground coverage corresponding to a single pixel in the remote sensing image, typically expressed in meters or kilometers, such as 250 meters by 250 meters or 1 kilometer by 1 kilometer. Pixel size is a key indicator of the spatial resolution of remote sensing data, directly affecting the accuracy and reliability of subsequent data processing.

[0022] By acquiring remote sensing monitoring data and the pixel size of the monitoring data, a basic data source and spatial reference scale are provided for subsequent vegetation data management, laying the foundation for accurately assessing the impact of terrain shadows and drought stress on vegetation monitoring accuracy. In practical applications, remote sensing data sources with different spatial resolutions can be selected for different monitoring purposes. For example, for regional vegetation change monitoring, medium- and low-resolution data (pixel size of 250 meters to 1 kilometer) can be used, while for refined vegetation management, high-resolution data (pixel size less than 10 meters) can be used.

[0023] S2. Collecting time information and terrain information within the target area, performing terrain shadow occlusion recognition on the target area, obtaining shadow area distribution, and obtaining a shadow recognition accuracy coefficient.

[0024] Specifically, first, collect time and terrain information within the target area. Time information includes time parameters such as the specific date and time of remote sensing data acquisition. These parameters directly affect the solar altitude and azimuth, and thus the formation and distribution of terrain shadows. Terrain information primarily includes elevation data, slope, and aspect of the target area, and other terrain characteristics, which can be obtained through digital elevation models (DEMs). Specifically, existing DEM databases or specialized terrain surveying methods can be used to obtain elevation information for the target area, typically with meter-level or sub-meter-level accuracy.

[0025] Secondly, based on the collected time and terrain information, terrain shadows in the target area are identified, thereby predicting the formation and distribution of surface shadows and obtaining the distribution of shadow areas within the target area, including information such as the geographic coordinates, shape, and area of ​​each shadow area. This process can be implemented based on deep learning methods. By analyzing the relationship between the sun's position (determined by time information) and terrain features, the specific areas and scopes where light is blocked by terrain to form shadows are determined. At the same time, by comparing the consistency of terrain shadow prediction results (i.e., historical predictions) on known samples with the actual shadow distribution, the accuracy of the above shadow identification process is evaluated, and the shadow identification accuracy coefficient is determined, reflecting the reliability of the above shadow prediction and identification process, providing support for the accuracy assessment of subsequent data.

[0026] By identifying and quantifying the distribution of shadow areas within target regions, we provide a basis for assessing the impact of shadows on the accuracy of vegetation remote sensing monitoring, thereby providing key information support for subsequent data accuracy identification and management. In particular, when the size of shadow areas is smaller than the remote sensing pixel size, vegetation information may not be accurately represented in remote sensing images, thus affecting the accuracy and reliability of the data.

[0027] S3. Collect meteorological information in the target area, identify drought stress areas, obtain drought stress area distribution, and obtain drought identification accuracy coefficient.

[0028] Specifically, meteorological information for the target area is first collected. This information includes, but is not limited to, meteorological parameters such as precipitation, temperature, humidity, and evapotranspiration over a predefined timeframe (e.g., the past 30, 60, or 90 days). This information can be obtained through field data from meteorological stations, satellite inversion data, or the output of regional meteorological models. These meteorological parameters directly affect the water supply and growth conditions of vegetation and are crucial for determining whether vegetation is experiencing drought stress.

[0029] Secondly, based on the collected meteorological information, drought stress areas in the target region are identified to predict the formation and distribution of vegetation drought stress. The distribution of drought stress areas within the target region is determined, including information such as the geographic coordinates, shape, and area of ​​each drought stress area. This process can be implemented using deep learning methods. By analyzing the relationship between meteorological parameters such as precipitation, temperature, and humidity and the state of vegetation drought stress, the specific areas and scope of vegetation potentially affected by drought stress can be determined. Furthermore, by comparing the consistency of drought stress prediction results for known samples (i.e., historical predictions) with the actual drought stress distribution, the accuracy of this drought stress identification process is evaluated and a drought identification accuracy coefficient is determined. This reflects the reliability of the drought stress prediction and identification process and provides support for subsequent data accuracy assessment.

[0030] By identifying and quantifying the distribution of drought-stressed areas within target regions, we provide a basis for assessing the impact of drought stress on the accuracy of vegetation remote sensing monitoring, thereby providing key information support for subsequent data accuracy identification and management. In particular, when the size of drought-stressed areas is smaller than the remote sensing pixel size, the reduction in chlorophyll caused by drought may not be accurately represented as vegetation areas in remote sensing images, thus affecting the accuracy and reliability of the data.

[0031] S4. According to the shadow area distribution, shadow identification accuracy coefficient, drought stress area distribution, drought identification accuracy coefficient, and the pixel size, the accuracy of the remote sensing monitoring data is marked and data management is performed.

[0032] Specifically, first, based on the obtained shadow area distribution, the size parameters of multiple shadow areas in the target area are extracted and calculated, and compared with the pixel size of remote sensing monitoring to determine the proportion of shadow areas smaller than the pixel size. Since shadow areas smaller than the pixel size may not be accurately expressed on remote sensing images, there is uncertainty in the vegetation information in this area, and it is necessary to calculate the basic shadow error coefficient to evaluate its impact on the accuracy of remote sensing monitoring. Secondly, based on the obtained shadow recognition accuracy coefficient, the basic shadow error coefficient is compensated, that is, the basic shadow error coefficient is divided by the shadow recognition accuracy coefficient to obtain the compensated shadow error degree. In this way, the uncertainty in the shadow recognition process is taken into account, thereby obtaining a more accurate error assessment result. Then, by subtracting the shadow error degree from 1, the shadow accuracy is calculated to reflect the reliability level of the remote sensing monitoring data in terms of shadow influence.

[0033] Similarly, for drought-stressed areas, the size parameters of multiple drought-stressed areas were extracted and calculated based on their distribution information. These parameters were then compared with the pixel size of remote sensing monitoring to determine the proportion of drought-stressed areas smaller than the pixel size, and the basic drought stress error coefficient was calculated. Then, based on the drought identification accuracy coefficient, the basic drought stress error coefficient was compensated to obtain the drought stress error degree. The drought stress accuracy was calculated by subtracting the drought stress error degree from 1, reflecting the reliability of remote sensing monitoring data in terms of drought stress impacts.

[0034] The overall accuracy of remote sensing monitoring data is then calculated based on shadow accuracy and drought stress accuracy. This accuracy value is used as a data quality indicator to identify and manage remote sensing monitoring data. This approach enables precise assessment and management of remote sensing vegetation data accuracy, provides data users with a basis for determining data authenticity, and improves the scientific nature and reliability of analytical decisions based on remote sensing monitoring data. Furthermore, this data management approach provides a reference for subsequent data processing and analysis, thereby improving the efficiency and value of remote sensing vegetation monitoring data and effectively supporting scientific decision-making in areas such as vegetation resource management, ecological and environmental monitoring, and agricultural production.

[0035] Furthermore, vegetation remote sensing monitoring is conducted in the target area, remote sensing monitoring data is collected, and the pixel size of remote sensing monitoring is obtained, including:

[0036] S11. Conduct vegetation remote sensing monitoring in the target area and collect remote sensing monitoring data;

[0037] S12. Obtain the resolution of remote sensing monitoring and calculate the pixel size of the minimum pixel for remote sensing monitoring.

[0038] In one feasible implementation, satellite or aerial remote sensing techniques are first used to comprehensively monitor and collect data from the target area. This process involves selecting remote sensing equipment suitable for vegetation monitoring, such as multispectral or hyperspectral sensors carried by satellites like MODIS, Landsat, and Sentinel, to acquire electromagnetic reflectivity or radiance data from the target area. Based on the reflectivity characteristics of different wavebands, various vegetation indices representing vegetation growth, such as the Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI), and Leaf Area Index (LAI), are calculated to generate comprehensive remote sensing monitoring data. This data directly reflects information such as vegetation cover and biomass in the target area.

[0039] Next, obtain the spatial resolution of the remote sensing monitoring equipment used. Spatial resolution is typically expressed as the ground distance per pixel, such as 250, 500, or 1000 meters for MODIS data and 30 meters for Landsat data. Based on the spatial resolution of the remote sensing data, calculate the pixel size of the minimum pixel for remote sensing monitoring, that is, the actual area covered by a single pixel on the ground. For example, for remote sensing data with a resolution of 250 meters, the pixel size is 250 meters by 250 meters, or 62,500 square meters. This pixel size directly reflects the spatial accuracy of remote sensing data and is an important reference scale for subsequent assessments of the impact of terrain shadows and drought stress on monitoring accuracy.

[0040] Through the above steps, not only did we obtain remote sensing vegetation monitoring data for the target area, but we also determined the spatial accuracy characteristics of the data, laying the foundation for subsequent data accuracy assessment and management. In particular, the determination of pixel size is directly related to the ability to detect and identify small-scale vegetation changes and is crucial for accurately assessing the reliability of remote sensing monitoring data.

[0041] Furthermore, collecting time information and terrain information within the target area, identifying terrain shadows in the target area, obtaining shadow area distribution, and obtaining a shadow identification accuracy coefficient include:

[0042] S21, collecting time information and terrain information within the target area, wherein the terrain information includes elevation information;

[0043] S22, inputting the time information and terrain information into a pre-built shadow predictor, and predicting and outputting a shadow area distribution, wherein the shadow area distribution includes the geographic coordinates of the shadow area;

[0044] S23 , testing the accuracy of the shadow predictor, and calculating a shadow recognition accuracy coefficient based on the terrain accuracy of the terrain information collection.

[0045] In a preferred embodiment, detailed time and topographic information of the target area is first collected. Time information includes time parameters such as year, month, day, hour, and minute acquired from remote sensing data. These parameters determine the azimuth and altitude of the sun at the time of observation and are essential for calculating lighting conditions and shadow formation. Topographic information primarily refers to surface elevation data of the target area, including the three-dimensional coordinates of longitude, latitude, and altitude for each point, typically expressed using a digital elevation model (DEM). Topographic information can be obtained through various channels, such as global elevation data products like SRTM and ASTER GDEM, or through more accurate local elevation data obtained through field measurement methods such as topography and lidar.

[0046] The collected time and terrain information is then fed into a pre-built shadow predictor for processing. This shadow predictor, a data processing model based on deep learning methods, calculates the light propagation path and terrain obstruction based on the input time information (determining the sun's position) and terrain information (determining the surface relief), thereby predicting the output shadow area distribution. The resulting shadow area distribution is expressed as geographic coordinates, precisely locating the spatial location of each shadow area and facilitating subsequent spatial correlation analysis with remote sensing pixels. In addition to geographic coordinates, the shadow area distribution can also include information such as shadow shape, area, and duration, comprehensively describing shadow characteristics and their potential impacts.

[0047] The accuracy of the shadow predictor's predictions is then evaluated. The accuracy of the shadow predictor's historically predicted shadow areas is compared with those from actual observations or high-precision simulations, and the degree of consistency between the two is calculated to determine the accuracy of the shadow predictor. Furthermore, to account for potential errors in terrain information acquisition, the accuracy of the terrain data is incorporated into the comprehensive evaluation system. The accuracy of the shadow predictor and the terrain accuracy are combined, and the final shadow recognition accuracy coefficient is calculated using a weighted average or other method. This coefficient is a value between 0 and 1, with values ​​closer to 1 indicating more reliable shadow recognition results. This provides a reliable reference for subsequent data accuracy assessments.

[0048] Through these steps, we achieved accurate identification and reliability assessment of terrain shadows in the target area, laying a solid foundation for subsequent evaluation of the impact of shadows on the accuracy of vegetation remote sensing monitoring. Compared to traditional geometric calculation methods, this deep learning-based shadow recognition method can better handle the formation and changes of shadows in complex terrain conditions, improving the accuracy and efficiency of shadow recognition.

[0049] Furthermore, the step of pre-building the shadow predictor includes:

[0050] S241. Collect a sample time information set and a sample terrain information set based on a vegetation remote sensing database, collect coordinates of all shadow areas within the region under different sample time information and sample terrain information, and mark them to obtain a sample shadow area distribution set;

[0051] S242. Use deep learning to build a network structure for shadow prediction;

[0052] S243 , using the sample time information set, the sample terrain information set, and the sample shadow area distribution set, perform iterative supervised training on the shadow predictor until convergence, thereby completing the pre-construction.

[0053] In a preferred embodiment, first, based on the existing vegetation remote sensing database, multiple sets of sample time information and sample terrain information are collected. The sample time information set includes time parameters of different seasons, different dates and different times, covering various possible combinations of solar altitude angles and azimuth angles; the sample terrain information set includes terrain data of various terrain types and elevation distribution characteristics, such as digital elevation models of different landform units such as plains, hills, and mountains. Then, for each combination of sample time information and sample terrain information, the precise coordinates of all shadow areas in the area under the corresponding conditions are obtained through high-precision lighting simulation or field observation. These shadow area coordinates are processed through annotation to form a sample shadow area distribution set, which serves as the target output for deep learning model training. The diversity and representativeness of the sample data directly affect the generalization ability of the model. Therefore, during the collection process, it is necessary to ensure that the samples cover all possible situations of the target application scenario.

[0054] Then, the shadow prediction network structure is constructed based on deep learning technology. Specifically, a convolutional neural network (CNN), recurrent neural network (RNN), or a hybrid architecture can be used to design a network model that can effectively process spatiotemporal data. The input layer of this network structure receives time information (such as parameters such as year, month, day, hour, and minute) and terrain information (such as elevation raster data). The intermediate layers, including multiple convolutional layers, pooling layers, and fully connected layers, are used to extract features and learn the complex relationship between time and terrain factors and shadow formation. The output layer generates a prediction of the shadow area distribution. The design of this network structure requires a comprehensive consideration of prediction accuracy and computational efficiency. The performance and adaptability of the network can be optimized by combining different types of network layers, adjusting the number of layers and neurons, and selecting appropriate activation functions and connection methods.

[0055] The constructed shadow predictor network is then trained using the collected sample data. The training process utilizes supervised learning, taking sample time and terrain information as input and the distribution of sample shadow areas as the desired output. A backpropagation algorithm continuously adjusts network parameters to minimize the error between the predicted output and the actual annotations. The training process is iterative. In each iteration, the network processes all or part of the training samples, updates weight parameters, and gradually improves prediction accuracy. Training concludes when the model performance reaches the preset target or the training loss function stabilizes (i.e., convergence is achieved), completing the pre-construction of the shadow predictor. To prevent overfitting, cross-validation, early stopping, and regularization techniques can be employed during training to ensure good generalization of the model.

[0056] Through the above steps, a deep learning model capable of accurately predicting terrain shadows was established. This model can quickly and accurately calculate and predict the distribution of shadow areas based on input time and terrain information, providing important support for the subsequent management of vegetation remote sensing monitoring data. Compared with traditional geometric calculations or physical simulation methods, this deep learning-based shadow prediction method can better handle shadow prediction problems under complex terrain and lighting conditions, with higher computational efficiency and prediction accuracy.

[0057] Furthermore, meteorological information in the target area is collected to identify drought stress areas, obtain drought stress area distribution, and obtain drought identification accuracy coefficients, including:

[0058] S31, collecting meteorological information within a preset time range in the target area;

[0059] S32, inputting the meteorological information into a pre-built drought stress predictor, and predicting and outputting a drought stress area distribution, wherein the drought stress area distribution includes the geographical coordinates of the drought stress area;

[0060] S33. Test and obtain the accuracy of the drought stress predictor, and calculate and obtain the drought identification accuracy coefficient in combination with the meteorological accuracy of the meteorological information collection.

[0061] In a preferred embodiment, first, meteorological information of the target area within a preset time range in the past is collected. The preset time range can be 30 days, 60 days, 90 days or longer, and the specific range can be determined according to the response characteristics of vegetation to drought and research needs. The collected meteorological information mainly includes meteorological parameters closely related to the moisture status of vegetation, such as precipitation, temperature, humidity, evapotranspiration, solar radiation, etc. The channels for obtaining meteorological information include measured data from meteorological stations, meteorological satellite inversion data, regional meteorological model output results, etc. In actual operation, a time series data set is constructed as meteorological information to record the changes in meteorological conditions at each time point in order to comprehensively evaluate the cumulative impact of meteorological conditions on vegetation growth, especially the formation and development process of drought stress.

[0062] The collected meteorological information is then input into a pre-built drought stress predictor for processing. This drought stress predictor is a data processing model built based on deep learning methods. It can analyze the spatiotemporal changes in vegetation water stress conditions based on the input meteorological parameter sequence, thereby predicting and outputting the distribution of drought stress areas. The obtained drought stress area distribution is expressed in the form of geographic coordinates, accurately locating the spatial position of each drought stress area, facilitating subsequent spatial correlation analysis with remote sensing pixels. In addition to geographic coordinates, the drought stress area distribution can also include information such as the intensity level, duration, and affected area of ​​the drought, comprehensively describing the characteristics of drought stress and its potential impact. By learning the complex relationship between meteorological conditions and vegetation responses, the drought stress predictor can identify mild or early drought stress conditions that may not be fully reflected in conventional drought indices.

[0063] The accuracy of the drought stress predictor's predictions is then evaluated. The accuracy of the drought stress predictor's predictions is calculated by comparing the drought stress areas predicted by the drought stress predictor with those from actual observations or high-precision assessments. The degree of consistency between the two is then calculated to determine the accuracy of the drought stress predictor. Furthermore, to account for potential errors in meteorological information collection, the accuracy of meteorological data is incorporated into the comprehensive evaluation system. The drought stress predictor's accuracy and meteorological accuracy are combined, and a weighted average or other mathematical method is used to calculate the drought identification accuracy coefficient. This coefficient is a value between 0 and 1, with values ​​closer to 1 indicating more reliable drought stress identification results. This provides a reliable reference for subsequent data accuracy assessments.

[0064] Through these steps, the precise identification and reliability assessment of drought stress conditions in the target area were achieved, laying a solid foundation for subsequent assessments of the impact of drought stress on the accuracy of vegetation remote sensing monitoring. Compared to traditional drought index methods, this deep learning-based drought stress identification method better captures the nonlinear relationship between meteorological conditions and vegetation responses, improving the accuracy and sensitivity of drought stress identification. In particular, it significantly enhances the ability to identify local-scale drought stress in complex geographical conditions.

[0065] Furthermore, the step of pre-building the drought stress predictor includes:

[0066] S341. Collecting a set of sample meteorological information based on a vegetation remote sensing database, and collecting coordinates of all drought stress areas within a region under different sample meteorological information, and marking to obtain a distribution set of sample drought stress areas;

[0067] S342. Use deep learning to build a network structure for drought stress prediction;

[0068] S343. Using the sample meteorological information set and the sample drought stress regional distribution set, perform iterative supervised training on the drought stress predictor until convergence, thereby completing the pre-construction.

[0069] In a preferred embodiment, first, multiple sets of sample meteorological information are collected based on the existing vegetation remote sensing database. The sample meteorological information set includes time series of meteorological parameters in different regions, different seasons, and different climatic conditions, such as precipitation, temperature, humidity, evapotranspiration, solar radiation, etc., covering various possible meteorological combinations and change patterns. For each set of sample meteorological information, the precise coordinates of all drought stress areas in the region under the corresponding meteorological conditions are determined through field surveys, expert evaluations, or analysis methods based on changes in vegetation indices. These drought stress area coordinates are labeled to form a sample drought stress area distribution set, which serves as the target output for deep learning model training. The comprehensiveness and representativeness of the sample data are directly related to the model's ability to identify different types and intensities of drought stress. Therefore, during the collection process, special attention should be paid to including cases of mild, moderate, and severe drought stress, as well as the response characteristics of different vegetation types to drought.

[0070] Then, the network structure of the drought stress predictor is constructed based on deep learning technology. Specifically, architectures such as long short-term memory networks (LSTMs), gated recurrent units (GRUs), or spatiotemporal convolutional networks (STCNs), which have strong time series processing capabilities, can be used to design a composite network model capable of effectively processing meteorological time series and spatially distributed data. The network's input layer receives a time series of meteorological information (such as precipitation, temperature, humidity, and other parameters over a period of time). The intermediate layers, including multi-layer recurrent units or convolutional layers, attention mechanism layers, and fully connected layers, are used to extract time series features and capture the complex relationship between spatiotemporal variations in meteorological conditions and vegetation drought stress. The output layer generates a prediction of the regional distribution of drought stress. The design of the network structure requires a balance between prediction accuracy and model complexity. The accuracy and generalization of the model's drought stress identification can be improved by combining different types of network layers, adjusting the number of layers and units, and selecting appropriate activation functions and optimization algorithms.

[0071] Subsequently, the constructed drought stress predictor network was trained using the collected sample data. The training process utilizes a supervised learning approach, taking a set of sample meteorological information as input and a set of sample drought stress area distributions as the desired output. A backpropagation algorithm continuously adjusts network parameters to minimize the error between the predicted output and the actual annotations. The specific training process includes data preprocessing (standardization, missing value processing, etc.), batch training, and parameter updates. Through multiple rounds of iteration, the model's accuracy in identifying drought stress areas is gradually improved. Training ends when the model performance reaches the preset target or the training loss function stabilizes (i.e., convergence is reached), completing the pre-construction of the drought stress predictor. To enhance the robustness and adaptability of the model, techniques such as data augmentation, cross-validation, and ensemble learning can be employed during training. Appropriate sampling or weighting strategies can be employed for imbalanced data (e.g., when there are fewer drought samples) to ensure that the model has good recognition capabilities for various drought stress conditions.

[0072] Through the above steps, a deep learning model capable of accurately predicting the regional distribution of drought stress was established. Based on input meteorological information, this model can quickly and accurately calculate and predict the regional distribution of vegetation potentially affected by drought stress within a target region, providing important support for the subsequent management of vegetation remote sensing monitoring data. Compared with traditional drought index-based methods, this deep learning-based drought stress prediction method comprehensively considers the combined impact of multiple meteorological factors and their temporal and spatial variations on vegetation, achieving higher prediction accuracy and spatial resolution, making it suitable for refined vegetation drought stress monitoring and assessment in complex geographical environments.

[0073] Furthermore, based on the shadow area distribution, shadow identification accuracy coefficient, drought stress area distribution, drought identification accuracy coefficient, and the pixel size, the accuracy of the remote sensing monitoring data is marked and data management is performed, including:

[0074] S41, extracting and calculating multiple shadow sizes of multiple shadow areas according to the shadow area distribution;

[0075] S42, determining the proportion of shadow sizes smaller than the pixel size, and calculating a basic shadow error coefficient;

[0076] S43, performing compensation calculation processing on the basic shadow error coefficient according to the shadow recognition accuracy coefficient to obtain a shadow error degree, and calculating a shadow accuracy;

[0077] S44, extracting and calculating a plurality of drought stress dimensions of a plurality of drought stress areas according to the distribution of the drought stress areas;

[0078] S45, determining the proportion of drought stress sizes smaller than the pixel size, and calculating a basic drought stress error coefficient;

[0079] S46. Performing compensation calculation processing on the basic drought stress error coefficient according to the drought identification accuracy coefficient to obtain a drought stress error degree, and calculating a drought stress accuracy degree;

[0080] S47. Calculate the accuracy of the remote sensing monitoring data based on the shadow accuracy and the drought stress accuracy, and perform labeling and data management on the remote sensing monitoring data.

[0081] In a preferred embodiment,

[0082] First, based on the obtained shadow area distribution, the geometric characteristic parameters of each shadow area are extracted, and the shadow sizes of multiple shadow areas are calculated. Shadow size can be expressed as geometric quantities such as the maximum length, average width, and area of ​​the shadow area, or as a unified representation using the equivalent diameter (i.e., the diameter of a circle with the same area as the shadow area). These size parameters directly reflect the spatial scale of the shadow area and serve as fundamental data for assessing the impact of shadows on remote sensing monitoring accuracy. Then, the shadow size of each shadow area is compared with the pixel size of the remote sensing monitoring. The number of shadow areas smaller than the pixel size and their proportion in the total number of shadow areas are statistically determined. This proportion reflects the proportion of shadow areas that may not be accurately represented in remote sensing imagery due to being too small to be accurately represented, and serves as a direct indicator for assessing the impact of shadows on remote sensing monitoring accuracy. Based on this proportion and the ratio of shadow size to pixel size, the basic shadow error coefficient is calculated. For example, if the shadow area smaller than the pixel size accounts for 10% of the total shadow area, and the average size of these small shadow areas is 1 / 2 of the pixel size, the basic shadow error coefficient can be calculated as 10% × (pixel size / average shadow size) = 10% × 2 = 20%. Subsequently, the basic shadow error coefficient is compensated based on the obtained shadow recognition accuracy coefficient. Specifically, the basic shadow error coefficient is divided by the shadow recognition accuracy coefficient to amplify the potential error caused by the uncertainty of shadow recognition. For example, if the basic shadow error coefficient is 20% and the shadow recognition accuracy coefficient is 0.8, the shadow error degree = 20% ÷ 0.8 = 25%. This error amplification process ensures that when shadow recognition is not accurate enough, the impact of shadows on remote sensing monitoring accuracy can be more conservatively evaluated. Then, by subtracting the shadow error degree from 1, that is, 1-25% = 75%, the shadow accuracy is calculated as an indicator reflecting the reliability of remote sensing monitoring data under the influence of shadow factors.

[0083] At the same time, similar to S41, based on the obtained distribution of drought stress areas, the drought stress dimensions of multiple drought stress areas are extracted and calculated. The drought stress dimensions can be expressed as spatial geometric characteristics of the drought stress area, such as area, length, or equivalent diameter. These dimension parameters directly reflect the spatial scale of the drought stress area and serve as basic data for assessing the impact of drought stress on remote sensing monitoring accuracy. Then, similar to S42, the drought stress dimension of each drought stress area is compared with the pixel size of remote sensing monitoring, and the number of drought stress areas smaller than the pixel size and their proportion in the total number of drought stress areas are statistically obtained. Based on this proportion and the ratio of the drought stress dimension to the pixel size, a basic drought stress error coefficient is calculated to reflect the degree of remote sensing monitoring error that may be caused by the undersized drought stress area. Subsequently, similar to S43, based on the obtained drought identification accuracy coefficient, the basic drought stress error coefficient is compensated and calculated to obtain the drought stress error degree. Specifically, the basic drought stress error coefficient is divided by the drought identification accuracy coefficient to amplify the potential error caused by the uncertainty of drought stress identification. Then, the drought stress accuracy was calculated by subtracting the drought stress error from 1, which was used as an indicator to reflect the reliability of remote sensing monitoring data under the influence of drought stress factors.

[0084] Based on the obtained shadow accuracy and drought stress accuracy, comprehensive assessment methods such as weighted averaging, minimum value rounding, or probabilistic combination are then used to calculate the overall accuracy of the remote sensing monitoring data. This accuracy is added to the metadata of the remote sensing monitoring data as an indicator of data quality to guide its proper use and interpretation. Furthermore, remote sensing monitoring data can be categorized and managed according to accuracy level, such as high-reliability data, medium-reliability data, and low-reliability data, to facilitate data users in selecting appropriate data sources based on specific application needs.

[0085] Through these steps, we achieved precise assessment and identification of the accuracy of remote sensing vegetation monitoring data, and established a data quality management system that considers factors such as shadows and drought stress. This data management approach objectively reflects the reliability of remote sensing monitoring data, improves the pertinence and effectiveness of data use, and provides reliable data quality assurance for applications such as vegetation monitoring, ecological assessment, and resource management based on remote sensing data.

[0086] Furthermore, the proportion of shadow sizes smaller than the pixel size is determined and the basic shadow error coefficient is calculated, including:

[0087] S421, determining and obtaining a proportion of shadow sizes smaller than the pixel size, and obtaining an error area proportion;

[0088] S422. Calculate the average of shadow sizes smaller than the pixel size as the error average shadow size, calculate the ratio of the pixel size to the error average shadow size, adjust the error area ratio, and obtain a basic shadow error coefficient.

[0089] In a preferred embodiment, first, all shadow areas within the target area are compared according to the shadow size and the pixel size of remote sensing monitoring, and those shadow areas with shadow sizes smaller than the pixel size are screened out. Since the spatial range of these small-sized shadow areas is smaller than the minimum resolution unit of the remote sensing image, they may not be accurately expressed in the remote sensing image or may be completely ignored, resulting in monitoring errors. By calculating the proportion of these small-sized shadow areas in the total shadow areas of the target area, the error area ratio is obtained. For example, if there are 100 shadow areas in the target area, and the shadow sizes of 10 of them are smaller than the pixel size, the error area ratio is 10%.

[0090] Then, for the small-sized shadow areas that have been screened out, the average value of their sizes is calculated as the error-average shadow size. This average value can be calculated using arithmetic mean, area-weighted mean, or other appropriate statistical methods. The error-average shadow size reflects the typical spatial scale of shadow areas that may cause monitoring errors and is an important parameter for assessing the degree of error. Then, the ratio of the pixel size to the error-average shadow size is calculated. This ratio reflects the spatial resolution of the pixel for small-sized shadow areas. The larger the ratio, the larger the pixel size relative to the error-average shadow size, the poorer the remote sensing image's ability to express small-sized shadow areas, and the greater the potential error.

[0091] The basic shadow error coefficient is then calculated by multiplying the ratio of pixel size to the error-averaged shadow size by the error-area ratio. This calculation method considers both the proportion of small shadow areas (measured by the error-area ratio) and their size characteristics (measured by the ratio of pixel size to the error-averaged shadow size), providing a more comprehensive assessment of the impact of shadows on remote sensing monitoring accuracy. For example, if the error-area ratio is 10%, the error-averaged shadow size is 500 meters, and the pixel size is 1000 meters, the ratio of pixel size to the error-averaged shadow size is 1000 meters ÷ 500 meters = 2. Multiplying this ratio by the error-area ratio (2 × 10% = 20%) yields a basic shadow error coefficient of 20%. This means that the basic remote sensing monitoring error caused by shadows smaller than the pixel size is 20%.

[0092] Through the above steps, we achieved an accurate assessment of remote sensing monitoring errors caused by shadow factors. This assessment takes into account both the distribution of small shadow areas and their relative relationship to remote sensing resolution, providing a basis for subsequent data accuracy identification and management. This error assessment method is more reasonable than simply using the error area ratio because it considers the amplification effect of size differences on error and can more accurately reflect the actual impact of shadow factors on remote sensing monitoring accuracy.

[0093] Example 2, as Figure 2 As shown, based on the same inventive concept as the deep learning remote sensing monitoring vegetation data management method provided in Example 1, the embodiment of the present invention also provides a deep learning remote sensing monitoring vegetation data management system, including:

[0094] The vegetation monitoring module 11 is used to perform vegetation remote sensing monitoring on the target area, collect remote sensing monitoring data, and obtain the pixel size of remote sensing monitoring;

[0095] A terrain shadow recognition module 12 is configured to collect time information and terrain information within the target area, identify terrain shadows in the target area, obtain shadow area distribution, and acquire a shadow recognition accuracy coefficient;

[0096] The drought stress identification module 13 is used to collect meteorological information in the target area, identify drought stress areas, obtain drought stress area distribution, and obtain drought identification accuracy coefficient;

[0097] The data management module 14 is used to identify the accuracy of the remote sensing monitoring data and perform data management based on the shadow area distribution, shadow identification accuracy coefficient, drought stress area distribution, drought identification accuracy coefficient, and the pixel size.

[0098] Furthermore, the vegetation monitoring module 11 includes the following execution steps:

[0099] Conduct vegetation remote sensing monitoring in target areas and collect remote sensing monitoring data;

[0100] Obtain the resolution of remote sensing monitoring and calculate the pixel size of the minimum pixel for remote sensing monitoring.

[0101] Furthermore, the terrain shadow recognition module 12 includes the following execution steps:

[0102] Collecting time information and terrain information within the target area, wherein the terrain information includes elevation information;

[0103] Inputting the time information and terrain information into a pre-built shadow predictor, and predicting output to obtain shadow area distribution, wherein the shadow area distribution includes geographic coordinates of the shadow area;

[0104] The accuracy of the shadow predictor is tested, and the shadow recognition accuracy coefficient is calculated in combination with the terrain accuracy of the terrain information collection.

[0105] Furthermore, the step of pre-building the shadow predictor includes:

[0106] According to the vegetation remote sensing database, a sample time information set and a sample terrain information set are collected, and the coordinates of all shadow areas in the region under different sample time information and sample terrain information are collected, and the sample shadow area distribution set is obtained by marking;

[0107] Use deep learning to build a shadow predictor network structure;

[0108] The sample time information set, the sample terrain information set and the sample shadow area distribution set are used to perform iterative supervised training on the shadow predictor until convergence, thereby completing the pre-construction.

[0109] Furthermore, the drought stress identification module 13 includes the following execution steps:

[0110] Collecting meteorological information within a preset time range in the past within the target area;

[0111] Inputting the meteorological information into a pre-built drought stress predictor, and predicting and outputting a drought stress area distribution, wherein the drought stress area distribution includes the geographical coordinates of the drought stress area;

[0112] The accuracy of the drought stress predictor is obtained by testing, and the drought identification accuracy coefficient is calculated in combination with the meteorological accuracy of the meteorological information collection.

[0113] Furthermore, the step of pre-building the drought stress predictor includes:

[0114] According to the vegetation remote sensing database, a set of sample meteorological information is collected, and the coordinates of all drought stress areas in the region under different sample meteorological information are collected, and the distribution set of sample drought stress areas is obtained by marking;

[0115] Using deep learning, we build a network structure for drought stress prediction;

[0116] The sample meteorological information set and the sample drought stress regional distribution set are used to perform iterative supervised training on the drought stress predictor until convergence, thereby completing the pre-construction.

[0117] Furthermore, the data management module 14 includes the following execution steps:

[0118] Extracting and calculating multiple shadow sizes of multiple shadow areas according to the shadow area distribution;

[0119] Determine the proportion of shadow sizes smaller than the pixel size and calculate a basic shadow error coefficient;

[0120] According to the shadow recognition accuracy coefficient, the basic shadow error coefficient is compensated and calculated to obtain the shadow error degree, and the shadow accuracy is calculated;

[0121] extracting and calculating a plurality of drought stress dimensions of a plurality of drought stress regions according to the distribution of the drought stress regions;

[0122] Determine and obtain the proportion of drought stress sizes smaller than the pixel size, and calculate and obtain the basic drought stress error coefficient;

[0123] According to the drought identification accuracy coefficient, the basic drought stress error coefficient is compensated and calculated to obtain the drought stress error degree, and the drought stress accuracy is calculated;

[0124] The accuracy of the remote sensing monitoring data is calculated based on the shadow accuracy and the drought stress accuracy, and the remote sensing monitoring data is marked and managed.

[0125] Furthermore, the data management module 14 further includes the following execution steps:

[0126] Determine the proportion of shadow sizes smaller than the pixel size to obtain the proportion of error areas;

[0127] The average of the shadow sizes smaller than the pixel size is calculated as the error average shadow size, and the ratio of the pixel size to the error average shadow size is calculated. The error area ratio is adjusted and calculated to obtain a basic shadow error coefficient.

[0128] It should be noted that, in the above embodiments, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant description of other embodiments.

[0129] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0130] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded computer, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0131] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0132] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0133] Although preferred embodiments of the present invention have been described, additional changes and modifications to these embodiments may occur to those skilled in the art once the basic inventive concepts become known.

[0134] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the present invention and its equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A deep learning remote sensing monitoring vegetation data management method, characterized in that: The method comprises: Conduct vegetation remote sensing monitoring in the target area, collect and obtain remote sensing monitoring data, obtain the resolution of remote sensing monitoring, and calculate the pixel size of the minimum pixel for remote sensing monitoring; Collecting time information and terrain information within the target area, performing terrain shadow recognition on the target area, obtaining shadow area distribution, and obtaining a shadow recognition accuracy coefficient; Collecting meteorological information in the target area, identifying drought stress areas, obtaining drought stress area distribution, and obtaining drought identification accuracy coefficient; According to the shadow area distribution, shadow identification accuracy coefficient, drought stress area distribution, drought identification accuracy coefficient, and pixel size, the accuracy of the remote sensing monitoring data is marked and data management is performed, including: Extracting and calculating multiple shadow sizes of multiple shadow areas according to the shadow area distribution; Determine the proportion of shadow sizes smaller than the pixel size and calculate the basic shadow error coefficient, including: Determine the proportion of shadow sizes smaller than the pixel size to obtain the proportion of error areas; Calculating the average value of the shadow sizes smaller than the pixel size as the error average shadow size, calculating the ratio of the pixel size to the error average shadow size, adjusting the error area ratio, and obtaining a basic shadow error coefficient; Dividing the basic shadow error coefficient by the shadow recognition accuracy coefficient to obtain a shadow error degree, and calculating the shadow accuracy; extracting and calculating a plurality of drought stress dimensions of a plurality of drought stress regions according to the distribution of the drought stress regions; Determine and obtain the proportion of drought stress sizes smaller than the pixel size, and calculate and obtain the basic drought stress error coefficient; Dividing the basic drought stress error coefficient by the drought identification accuracy coefficient to obtain a drought stress error degree, and calculating the drought stress accuracy; The accuracy of the remote sensing monitoring data is calculated based on the shadow accuracy and the drought stress accuracy, and the remote sensing monitoring data is marked and managed.

2. The deep learning remote sensing monitoring vegetation data management method according to claim 1 is characterized in that: Collecting time information and terrain information within the target area, identifying terrain shadows in the target area, obtaining shadow area distribution, and obtaining a shadow identification accuracy coefficient, including: Collecting time information and terrain information within the target area, wherein the terrain information includes elevation information; Inputting the time information and terrain information into a pre-built shadow predictor, and predicting output to obtain shadow area distribution, wherein the shadow area distribution includes geographic coordinates of the shadow area; The accuracy of the shadow predictor is tested, and the shadow recognition accuracy coefficient is calculated in combination with the terrain accuracy of the terrain information collection.

3. The deep learning remote sensing monitoring vegetation data management method according to claim 2 is characterized in that: The steps of pre-building the shadow predictor include: According to the vegetation remote sensing database, a sample time information set and a sample terrain information set are collected, and the coordinates of all shadow areas in the region under different sample time information and sample terrain information are collected, and the sample shadow area distribution set is obtained by marking; Use deep learning to build a shadow predictor network structure; The sample time information set, the sample terrain information set and the sample shadow area distribution set are used to perform iterative supervised training on the shadow predictor until convergence, thereby completing the pre-construction.

4. The deep learning remote sensing monitoring vegetation data management method according to claim 1 is characterized in that: Collecting meteorological information in the target area, identifying drought stress areas, obtaining drought stress area distribution, and obtaining drought identification accuracy coefficients, including: Collecting meteorological information within a preset time range in the past within the target area; Inputting the meteorological information into a pre-built drought stress predictor, and predicting and outputting a drought stress area distribution, wherein the drought stress area distribution includes the geographical coordinates of the drought stress area; The accuracy of the drought stress predictor is obtained by testing, and the drought identification accuracy coefficient is calculated based on the meteorological accuracy of the meteorological information collection.

5. The deep learning remote sensing monitoring vegetation data management method according to claim 4 is characterized in that: The steps of pre-building the drought stress predictor include: According to the vegetation remote sensing database, a set of sample meteorological information is collected, and the coordinates of all drought stress areas in the region under different sample meteorological information are collected, and the distribution set of sample drought stress areas is obtained by marking; Using deep learning, we build a network structure for drought stress prediction; The sample meteorological information set and the sample drought stress regional distribution set are used to perform iterative supervised training on the drought stress predictor until convergence, thereby completing the pre-construction.

6. A deep learning remote sensing monitoring vegetation data management system, characterized by: A method for managing vegetation data monitored by remote sensing using deep learning according to any one of claims 1 to 5, the system comprising: Vegetation monitoring module, used to perform vegetation remote sensing monitoring of the target area, collect remote sensing monitoring data, obtain the resolution of remote sensing monitoring, and calculate the pixel size of the minimum pixel for remote sensing monitoring; A terrain shadow recognition module is used to collect time information and terrain information within the target area, identify terrain shadows in the target area, obtain shadow area distribution, and obtain a shadow recognition accuracy coefficient; A drought stress identification module is used to collect meteorological information in the target area, identify drought stress areas, obtain drought stress area distribution, and obtain drought identification accuracy coefficient; A data management module, configured to identify the accuracy of the remote sensing monitoring data and perform data management based on the shadow area distribution, shadow identification accuracy coefficient, drought stress area distribution, drought identification accuracy coefficient, and pixel size; The data management module further includes the following execution steps: Extracting and calculating a plurality of shadow sizes of a plurality of the shadow areas according to the shadow area distribution; Determine the proportion of shadow sizes smaller than the pixel size and calculate the basic shadow error coefficient, including: Determine the proportion of shadow sizes smaller than the pixel size to obtain the proportion of error areas; Calculating the average value of the shadow sizes smaller than the pixel size as the error average shadow size, calculating the ratio of the pixel size to the error average shadow size, adjusting the error area ratio, and obtaining a basic shadow error coefficient; Dividing the basic shadow error coefficient by the shadow recognition accuracy coefficient to obtain the shadow error degree, and calculating the shadow accuracy; extracting and calculating a plurality of drought stress dimensions of a plurality of drought stress regions according to the distribution of the drought stress regions; Determining and obtaining a proportion of drought stress sizes smaller than the pixel size, and calculating and obtaining a basic drought stress error coefficient based on a ratio relationship between the drought stress size and the pixel size; Dividing the basic drought stress error coefficient by the drought identification accuracy coefficient to obtain a drought stress error degree, and calculating the drought stress accuracy; The accuracy of the remote sensing monitoring data is calculated based on the shadow accuracy and the drought stress accuracy, and the remote sensing monitoring data is marked and managed.

Images