Method and device for identifying field irrigation winter wheat and electronic equipment

Abstract

The invention relates to the technical field of artificial intelligence and agricultural big data analysis, and provides a method and device for identifying field irrigation winter wheat and electronic equipment, and the method comprises the steps: obtaining a remote sensing time series data set of a farmland in a target region; and constructing a three-dimensional feature space composed of rainfall data, actual evapotranspiration data and normalized vegetation index data. And fitting a rainfall maintenance line in a two-dimensional characteristic plane formed by the rainfall data and the actual evapotranspiration data. And determining a vertical irrigation line value from any winter wheat pixel in the three-dimensional feature space to the rain maintenance line, and determining a slope length index according to the vertical irrigation line value and the normalized vegetation index data. And determining irrigation attributes of each winter wheat pixel in the three-dimensional feature space according to the slope length index, and mapping the irrigation attributes to a pre-obtained winter wheat crop map layer to obtain an irrigation rain culture classification map. The problem that an existing irrigation mapping method is insufficient in mechanism property and interpretability due to the fact that the existing irrigation mapping method depends on a single vegetation index is solved.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and agricultural big data analysis technology, and in particular to a method, apparatus and electronic device for identifying winter wheat irrigated in the field. Background Technology

[0002] Irrigation plays a crucial role in agricultural production as a key measure to alleviate agricultural drought and increase crop yields. Accurately identifying the distribution of irrigated winter wheat fields is essential for scientifically assessing irrigation conditions and water use efficiency in winter wheat growing areas. Traditional irrigated field surveys rely on manual field observations, which are time-consuming and lack timeliness, failing to achieve precise field-by-field surveys and often resulting in misrepresentation. With the rapid development of satellite remote sensing technology, it has become possible to accurately identify high-resolution winter wheat irrigated fields based on differences in evapotranspiration and water consumption characteristics between irrigated and non-irrigated fields. However, existing irrigation mapping methods are mostly based on single vegetation indices, lacking indicators that directly reflect irrigation-related physical processes, leading to insufficient mechanistic understanding and interpretability. Summary of the Invention

[0003] This invention provides a method, apparatus, and electronic device for identifying winter wheat under field irrigation, addressing the problem that existing irrigation mapping methods rely on a single vegetation index, leading to insufficient mechanistic understanding and interpretability. The technical solution proposed by this invention is as follows: In a first aspect, the present invention provides a method for identifying winter wheat grown in irrigated fields, comprising: Obtain remote sensing time-series datasets of farmland within the target area, including precipitation data, actual evapotranspiration data, and normalized vegetation index data; A three-dimensional feature space is constructed, consisting of precipitation data, actual evapotranspiration data, and normalized vegetation index data; among which, precipitation data and actual evapotranspiration data constitute a two-dimensional feature plane, and normalized vegetation index data serves as the third dimension. A rainfed line is fitted in a two-dimensional feature plane formed by precipitation data and actual evapotranspiration data, wherein the rainfed line represents the relationship between the actual evapotranspiration data and precipitation data of rainfed crops; Determine the vertical irrigation line value from any winter wheat pixel in the three-dimensional feature space to the rainfed line, and determine the slope length index based on the vertical irrigation line value and the normalized vegetation index data; The irrigation attribute of each winter wheat pixel in the three-dimensional feature space is determined based on the slope length index, and mapped to a pre-acquired winter wheat crop layer to obtain an irrigation-rainfed classification map. The irrigation attribute includes irrigated winter wheat and rainfed winter wheat. If the slope length index of a winter wheat pixel is greater than or equal to a preset threshold, the irrigation attribute of that winter wheat pixel is determined to be irrigated winter wheat; if the slope length index of a winter wheat pixel is less than the preset threshold, the irrigation attribute of that winter wheat pixel is determined to be rainfed winter wheat.

[0004] Optionally, the normalized vegetation index data and actual evapotranspiration data in the remote sensing time-series dataset are obtained in the following manner: Surface reflectance data from MOD09A1 and multispectral data from Landsat-8 are acquired. The surface reflectance data and multispectral data are spatiotemporally fused to obtain normalized vegetation index data at the first resolution, which is used as the normalized vegetation index data in the remote sensing time series dataset. The normalized vegetation index data at the first resolution is input into a pre-trained actual evapotranspiration data prediction model, and the actual evapotranspiration prediction value of the actual evapotranspiration data prediction model is obtained as the actual evapotranspiration data in the remote sensing time series dataset.

[0005] Optionally, the actual evapotranspiration data prediction model is trained in the following manner: Obtain time-series data of normalized vegetation index at a first resolution and actual evapotranspiration at a second resolution; wherein the first resolution is higher than the second resolution. The actual evapotranspiration time-series data at the second resolution is upsampled to the first resolution to obtain the actual evapotranspiration time-series data at the first resolution. Multiple sample points are randomly generated within the study area, and the normalized vegetation index value of each sample point at each time step is extracted from the normalized vegetation index time series data as a feature. Based on the preset training sample ratio, multiple training samples are randomly selected from all sample points, and the remaining samples are used as validation samples. The features of the training samples and the corresponding actual evapotranspiration values ​​are input into the random forest model for training to obtain the trained model; wherein, the random forest model has 50 trees and 6 seeds; The features of the verification samples and the corresponding actual evapotranspiration values ​​are input into the trained model to obtain the predicted actual evapotranspiration values ​​output by the trained model. The model performance index is calculated based on the predicted actual evapotranspiration value and the actual actual evapotranspiration value of each validation sample. The trained model is evaluated based on the model performance index. The model training and evaluation process is repeated until the maximum number of iterations is reached or the model performance index meets the preset requirements. The trained random forest model is then obtained as the prediction model for the actual evapotranspiration data.

[0006] Optionally, the method further includes: Obtain measured survey data and existing irrigation product data, match the existing irrigation product data with the geographical location of the measured survey data, and extract the predicted irrigation attribute corresponding to each measured sample point in the existing irrigation products; The geographical locations of the measured survey points are matched with the irrigation and rain-fed classification map to obtain the predicted irrigation attributes of each measured survey point in the irrigation and rain-fed classification map. Based on the measured survey data of each measured survey point, the corresponding predicted irrigation attributes in existing irrigation products, and the predicted irrigation attributes in the irrigation rain-fed classification map, a confusion matrix is ​​constructed. Based on the confusion matrix, multiple evaluation metrics are calculated; these multiple evaluation metrics include at least accuracy, precision, and recall. The accuracy of the irrigation rainfed classification map is evaluated based on the calculated evaluation indicators.

[0007] Optionally, the vertical irrigation line value is determined by the following formula: ; in, This represents the vertical irrigation line value of any winter wheat pixel in the three-dimensional feature space. and These represent the slope and intercept of the rainfed line, respectively. and Representing winter wheat pixels The actual evapotranspiration data and precipitation data.

[0008] Optionally, determining the vertical irrigation line value from any winter wheat pixel in the three-dimensional feature space to the rainfed line, and determining the slope length index based on the vertical irrigation line value and the normalized vegetation index data, includes: Calculate the vertical irrigation line value from any winter wheat pixel in the three-dimensional feature space to the rainfed line, and normalize the vertical irrigation line values ​​of all winter wheat pixels to obtain standardized vertical irrigation line values. The slope length index was determined based on standardized vertical irrigation line values ​​and normalized vegetation index data. ; in, Represents any winter wheat pixel in the three-dimensional feature space The slope length index, and Representing winter wheat pixels Standardized vertical irrigation line values ​​and normalized vegetation index data.

[0009] Secondly, the present invention also provides a device for identifying winter wheat irrigated in the field, comprising the following modules: The data acquisition module is used to acquire remote sensing time-series datasets of farmland within the target area, including precipitation data, actual evapotranspiration data, and normalized vegetation index data. The spatial construction module is used to construct a three-dimensional feature space composed of precipitation data, actual evapotranspiration data, and normalized vegetation index data; among them, precipitation data and actual evapotranspiration data constitute a two-dimensional feature plane, and normalized vegetation index data serves as the third dimension. The data fitting module is used to fit a rainfed line in a two-dimensional feature plane formed by precipitation data and actual evapotranspiration data. The rainfed line represents the relationship between the actual evapotranspiration data and precipitation data of rainfed crops. The index determination module is used to determine the vertical irrigation line value from any winter wheat pixel in the three-dimensional feature space to the rain-fed line, and to determine the slope length index based on the vertical irrigation line value and the normalized vegetation index data. The attribute classification module is used to determine the irrigation attribute of each winter wheat pixel in the three-dimensional feature space based on the slope length index, and map it to a pre-acquired winter wheat crop layer to obtain an irrigation-rainfed classification map. The irrigation attribute includes irrigated winter wheat and rainfed winter wheat. If the slope length index of a winter wheat pixel is greater than or equal to a preset threshold, the irrigation attribute of the winter wheat pixel is determined to be irrigated winter wheat; if the slope length index of a winter wheat pixel is less than the preset threshold, the irrigation attribute of the winter wheat pixel is determined to be rainfed winter wheat.

[0010] Thirdly, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the method for identifying winter wheat irrigated in the field as described in the first aspect above.

[0011] Fourthly, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for identifying winter wheat irrigated in the field as described in the first aspect above.

[0012] Fifthly, the present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the method for identifying winter wheat irrigated in the field as described in the first aspect above.

[0013] Based on the above technical solution, the beneficial effects of the present invention compared with the prior art are as follows: The method, apparatus, and electronic equipment provided by this invention for identifying irrigated winter wheat in the field effectively solve the problem of insufficient mechanistic rationality and interpretability caused by the reliance on a single vegetation index in existing irrigation mapping methods by constructing a three-dimensional feature space, fitting a rainfed line, and calculating the slope length index. Specifically, the method of this invention not only utilizes normalized vegetation index data but also combines precipitation data and actual evapotranspiration data to construct a three-dimensional feature space. This fusion of multi-dimensional data allows irrigation identification to no longer rely solely on a single vegetation index but comprehensively considers the water supply and demand relationship and evapotranspiration process of crop growth, thereby improving the mechanistic rationality and interpretability of the method. Fitting a rainfed line in the two-dimensional feature plane formed by precipitation data and actual evapotranspiration data clarifies the relationship between the actual evapotranspiration data and precipitation data of rainfed crops. This step not only provides a theoretical basis for irrigation identification but also makes the identification results more interpretable because the different distributions of irrigated and rainfed fields on both sides of the rainfed line reflect their essential differences in water use. The slope length index combines vertical irrigation line values ​​(representing irrigation probability) and normalized vegetation index data (representing crop growth) to form a comprehensive indicator. This indicator not only reflects the impact of irrigation on crop growth but also embodies the crop's own growth status, thereby further enhancing the method's mechanistic and interpretable aspects.

[0014] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention are realized and obtained through the structures particularly pointed out in the description and the drawings.

[0015] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating the method for identifying winter wheat under field irrigation provided by the present invention.

[0018] Figure 2 This is a schematic diagram of the rain-fed line and slope length index in the three-dimensional feature space provided by the present invention.

[0019] Figure 3 This is a schematic diagram of the winter wheat irrigation identification results provided by the present invention.

[0020] Figure 4 This is a technical flowchart of the method for identifying winter wheat under field irrigation provided by the present invention.

[0021] Figure 5 This is a schematic diagram of the device for identifying winter wheat irrigated in the field, provided by the present invention.

[0022] Figure 6 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0024] Existing irrigation mapping methods have the following three main problems: (1) Lack of mechanistic support: Most existing irrigation mapping methods are based on a single vegetation index (such as NDVI, EVI, etc.), which lacks indicators that directly reflect irrigation-related physical processes, resulting in insufficient mechanistic and interpretability of the methods.

[0025] (2) Low spatial resolution: Existing irrigation mapping typically has low spatial resolution, with the best product resolution only reaching the kilometer level (1000 meters), which is insufficient to meet the irrigation monitoring needs at the field scale.

[0026] (3) Insufficient crop differentiation: Existing irrigation mapping products cannot differentiate the irrigation distribution of different crops, especially in multi-season crop planting areas (such as rice-wheat or wheat-corn), and fail to identify irrigation for each crop individually, thus affecting the guidance of irrigation optimization.

[0027] To address the aforementioned shortcomings, this invention provides a method, apparatus, and electronic device for identifying irrigated winter wheat in fields. This invention constructs a Slope Length Index (SLI) for mapping irrigated winter wheat at the provincial and national scales using a three-dimensional canopy hydrothermal and spectral feature space (hereinafter referred to as "three-dimensional feature space"). The scalability and portability of this method have been verified in different regions and environments, providing an efficient and convenient solution for field-scale irrigated crop mapping.

[0028] This invention addresses the challenge of identifying irrigated crops at the field scale by proposing the following core technological innovations: (1) By combining precipitation and actual evapotranspiration data, the impact of irrigation on surface-atmosphere water transport is quantitatively assessed from the perspective of crop water supply and demand mechanisms. Based on the dual mechanism of irrigation promoting crop growth, a simple irrigated crop identification index, namely the slope length index (SLI), is innovatively proposed; (2) Based on a winter wheat crop layer, the spatial distribution characteristics of irrigation are depicted, successfully solving the problems of potential misclassification of irrigated farmland and ambiguity of annual irrigation data in double-cropping systems; (3) Focusing on winter wheat, the spatiotemporal characteristics of irrigation during its growth process are accurately captured, further realizing the identification of irrigation status at the field scale.

[0029] Reference Figure 1 As shown, the method for identifying winter wheat grown in irrigated fields includes the following: S110. Obtain remote sensing time-series datasets of farmland within the target area, including precipitation data, actual evapotranspiration data, and normalized vegetation index data.

[0030] Satellite remote sensing technology was used to acquire time-series remote sensing data of farmland within the target area. This data encompasses precipitation, actual evapotranspiration (AET), and Normalized Difference Vegetation Index (NDVI). Precipitation data reflects rainfall patterns in the region over a specific period; AET reflects the actual amount of water lost from farmland to the atmosphere; and NDVI is closely related to vegetation growth and cover, characterizing the growth status of winter wheat. The time-series remote sensing data reflects changes in various farmland indicators across different time periods, providing a rich information foundation for subsequent analysis.

[0031] Specifically, the precipitation data (Pre) comes from monthly precipitation data released by meteorological data centers or other relevant institutions. This precipitation data has a monthly temporal resolution, and the spatial resolution is set according to specific needs or data source characteristics to achieve high spatial accuracy (such as 1km or higher) to ensure that it can reflect the precipitation distribution in the target area in detail. For example, the temporal and spatial resolution of the precipitation data can be monthly and 1km, that is, precipitation data within the target area is provided once a month, and the spatial resolution is 1km, which can reflect the precipitation situation in the target area in a relatively detailed manner.

[0032] Actual evapotranspiration (AET) data comes from global or regional land surface evapotranspiration products such as ETMonitor, published by professional institutions (e.g., the International Research Center for Big Data for Sustainable Development). These products are typically calculated based on a combination of remote sensing and ground observation data. This AET data has a daily temporal resolution, and the spatial resolution is also set according to the characteristics of the data source, usually achieving an accuracy matching or higher than precipitation data (e.g., 1 km or higher) to reflect the real-time evapotranspiration of farmland water. For example, the temporal and spatial resolution of this evapotranspiration data can be set to daily, 1 km, meaning that global land surface evapotranspiration data is provided daily with a spatial resolution of 1 km, which can reflect the real-time actual evapotranspiration of farmland water.

[0033] The Normalized Difference Vegetation Index (NDVI) data of this invention were calculated from MOD09A1 and Landsat-8 reflectance images. MOD09A1 has a spatiotemporal resolution of 8 days and 250 meters, while Landsat-8 has a spatiotemporal resolution of 16 days and 30 meters. To achieve irrigation mapping at the field scale for winter wheat, a remote sensing spatiotemporal fusion method was used to perform spatiotemporal filtering on the NDVI data to achieve an 8-day, 30-meter resolution.

[0034] The normalized vegetation index data and actual evapotranspiration data in the remote sensing time-series dataset were obtained in the following manner: S1101. Acquire surface reflectance data from MOD09A1 and multispectral data from Landsat-8. Perform spatiotemporal fusion on the surface reflectance data and multispectral data to obtain normalized vegetation index data at a first resolution (30m), which is used as the normalized vegetation index data in the remote sensing time series dataset.

[0035] The process of implementing spatiotemporal filtering of NDVI data using remote sensing spatiotemporal fusion method is as follows: 1. Perform data preprocessing: MOD09A1 provides 8-day periodic surface reflectance data with a spatial resolution of 250 meters. First, its NDVI calculation results are extracted.

[0036] The formula for calculating NDVI is NDVI=(NIR-R) / (NIR+R), where NIR is the reflectivity in the near-infrared band and R is the reflectivity in the red band.

[0037] Landsat-8 provides 16-day multispectral data with a spatial resolution of 30m. The NDVI value is also calculated, and the image acquisition time is recorded.

[0038] Match the NDVI data of MOD09A1 and Landsat-8 in time and space to ensure that the time range of the input data for subsequent fusion is consistent.

[0039] 2. Spatiotemporal fusion methods can include Gaussian Filtering-Savitzky-Golay Filtering (GF-SG) algorithm, Spatial and Temporal Adaptive Reflectance Fusion Model (STARFM), Enhanced STARFM (ESTARFM), etc.

[0040] Taking spatiotemporal fusion using the GF-SG algorithm as an example, the GF-SG (Gaussian Filtering and Savitzky-Golay) algorithm can effectively fuse MOD09A1 (8 days, 250 meters) and Landsat-8 (16 days, 30 meters) NDVI data into an 8-day, 30-meter resolution NDVI sequence. The fusion process is as follows: First, spatial filtering is performed using Gaussian filtering: Gaussian filtering is applied to each period of NDVI data in MOD09A1 to generate a spatially smoothed NDVI sequence (250-meter resolution). The kernel size of the Gaussian filter can be adjusted according to the data noise level and spatial resolution differences. Spatial filtering of the MOD09A1 NDVI data reduces noise and extracts large-scale spatial variation features.

[0041] Secondly, time filtering is performed using Savitzky-Golay filtering: Savitzky-Golay filtering is applied to the Landsat-8 NDVI time series to generate a time-smoothed NDVI trend (30m resolution). Savitzky-Golay filtering uses local polynomial fitting to smooth noise while preserving detailed features of the time series (such as crop growth changes). The filter window size and polynomial order can be set according to actual needs; the filter window size can be 5 or 7 periods, and the polynomial order can be 2 or 3, to balance the smoothing effect and detail preservation. By smoothing the Landsat-8 NDVI time series, the time trend is extracted while retaining high-frequency details. 5 or 7 periods refer to the size of the Savitzky-Golay filter window, i.e., the number of data points used for local polynomial fitting.

[0042] Finally, the spatially smoothed MOD09A1 NDVI sequence (250-meter resolution) and the temporally smoothed Landsat-8 NDVI trend (30-meter resolution) are spatiotemporally fused. For each spatial location (x, y) and each time point t, the spatiotemporally fused NDVI data is... fused (x,y,t) is calculated in the following way: First, spatial resampling is performed: the spatially smoothed NDVI (original resolution of 250 meters) is resampled to a resolution of 30 meters to obtain the resampled MOD09A1 NDVI. MOD,smooth,30m (x, y, t). This data provides low spatial resolution but temporally continuous spatial background information. This can be achieved through bilinear interpolation or nearest-neighbor interpolation. Bilinear interpolation calculates the value of the target pixel by taking the weighted average of the four known pixels surrounding the target pixel. Nearest-neighbor interpolation directly assigns the value of the target pixel to the value of its nearest neighbor.

[0043] Secondly, time interpolation is performed: using the time-smoothed Landsat-8 NDVI trend, the NDVI value at intermediate time points (such as the 8-day cycle of MOD09A1) is predicted, resulting in the time-interpolated Landsat-8 trend NDVI. L8,SG,interpolated (x, y, t). This data provides trend information with high temporal resolution but low spatial resolution. This can be achieved through linear interpolation or spline interpolation. Taking linear interpolation as an example, linear interpolation predicts the value at an intermediate time point between two known time points using a linear relationship.

[0044] Finally, spatiotemporal fusion is performed: the resampled MOD09A1 data is combined with the time-interpolated Landsat-8 trend data, and high spatiotemporal resolution NDVI data is generated through weighted averaging or linear regression, i.e., the spatiotemporally fused NDVI data described above. fused (x,y,t).

[0045] NDVI fused (x,y,t)=α NDVI MOD,smooth,30m (x,y,t) β NDVI L8,SG,interpolated (x,y,t); Where α and β are weighting coefficients, satisfying α β=1.

[0046] This invention generates NDVI data with 8 days and 30m resolution by combining the spatial information of MOD09A1 with the temporal information of Landsat-8.

[0047] Using the spatiotemporally fused NDVI data as the aforementioned Normalized Difference Vegetation Index (NDVI) data, the original spatiotemporal resolutions of precipitation data, actual evapotranspiration data, and the spatiotemporally fused NDVI data may differ. Before constructing the three-dimensional feature space, remote sensing spatiotemporal fusion methods are needed to ensure that these data achieve a certain degree of temporal and spatial matching. Through matching, it is ensured that each winter wheat pixel has complete precipitation, evapotranspiration, and vegetation index data at its corresponding time point, in order to facilitate subsequent feature space construction and irrigation attribute identification.

[0048] S1102. Input the normalized vegetation index data of the first resolution into the pre-trained actual evapotranspiration data prediction model, and obtain the actual evapotranspiration prediction value of the actual evapotranspiration data prediction model as the actual evapotranspiration data in the remote sensing time series dataset.

[0049] Pre-trained prediction models based on real-world evapotranspiration data are saved in specific file formats (such as pickle files in Python, or SavedModel format in TensorFlow). When using the model for prediction, it must first be loaded into memory. After loading, the model is in a ready state, waiting to receive input data and make predictions.

[0050] The first-resolution NDVI data is input into the model according to the required format. The actual evapotranspiration data prediction model automatically extracts features from the input data; these features are the NDVI values ​​of each sample point (raster cell) at each time step. For example, for a specific raster cell, the model will use its NDVI values ​​for every 8 days within a year as a feature vector, which is then input into the various decision trees of the model for processing. The actual evapotranspiration data prediction model can use a random forest model, which consists of multiple decision trees, each of which makes an independent prediction on the input feature vector. In the decision tree, starting from the root node, the decision tree is traversed downwards step by step according to the comparison results of each feature value of the feature vector with the node splitting rule until a leaf node is reached. Each leaf node corresponds to a predicted value, which is the estimated actual evapotranspiration of that sample point (raster cell) at that time step.

[0051] The final predicted value is obtained by combining the prediction results of all decision trees. The averaging method can be used for this combination. For each sample point (raster cell) and each time step, the model calculates the average of the actual evapotranspiration estimates predicted by all decision trees, which is taken as the final prediction result for that sample point at that time step. After completing the prediction, the model outputs the actual evapotranspiration prediction value for each sample point (raster cell) at each time step. These prediction values ​​can be stored in the same raster data format as the input NDVI data, with the value of each raster cell representing the actual evapotranspiration prediction value for that location at the corresponding time step. The prediction results for all time steps are then organized in chronological order to form a complete actual evapotranspiration time-series dataset. This dataset has a spatial and temporal correspondence with the first-resolution NDVI data and can be used as actual evapotranspiration data in a remote sensing time-series dataset for subsequent analysis, application, and research.

[0052] MOD09A1 data typically has relatively low spatial resolution but high temporal resolution, while Landsat-8 multispectral data has high spatial resolution but low temporal resolution. This invention combines the two data sources using spatiotemporal fusion technology to generate normalized vegetation index data with a first-resolution resolution. This approach leverages the advantages of both types of data, ensuring both high spatial detail and good temporal continuity, thus providing a more comprehensive and accurate reflection of changes in surface vegetation at different temporal and spatial scales.

[0053] Actual evapotranspiration data is often difficult to obtain directly on a large scale and at high frequency through conventional methods. This invention inputs the fused first-resolution Normalized Difference Vegetation Index (NDVI) data into a pre-trained actual evapotranspiration data prediction model, enabling rapid and efficient generation of actual evapotranspiration data matching the spatiotemporal resolution of the NDVI data. This indirect acquisition method avoids complex field measurements, significantly reducing data acquisition costs and time, while achieving high-resolution AET data coverage over a large area. The Normalized Difference Vegetation Index is closely related to vegetation growth status and physiological characteristics, and vegetation status has a significant impact on the actual evapotranspiration process. Using NDVI data as input to predict ET data fully utilizes the indicative role of vegetation information on ET, making the generated AET data more consistent with actual vegetation transpiration and soil evaporation processes, thereby improving the accuracy and reliability of AET data and better reflecting the true state of the surface water cycle. This invention utilizes a random forest algorithm on the GEE platform. It trains a random forest model by inputting NDVI time-series data at a first resolution (30m) and actual evapotranspiration time-series data at a second resolution (1km). This allows the model to learn the correspondence between the NDVI time-series data and the actual evapotranspiration time-series data at the first resolution. Specifically, before inputting the actual evapotranspiration time-series data at the second resolution (1km) into the model, the resolution is first increased to the first resolution, and the actual evapotranspiration time-series data at the first resolution is used as the true actual evapotranspiration value. After model training, the aforementioned actual evapotranspiration data prediction model is obtained. Inputting the NDVI time-series data at the first resolution into the trained actual evapotranspiration data prediction model outputs the actual evapotranspiration time-series data at the first resolution (30m). The actual evapotranspiration time-series data at the second resolution is ETMonitor time-series data.

[0054] The actual evapotranspiration data prediction model was trained in the following manner: S1001. Obtain time-series data of normalized vegetation index (NDVI) at the first resolution and actual evapotranspiration at the second resolution.

[0055] We collected time-series data of Normalized Difference Vegetation Index (NDVI) at first resolution and actual evapotranspiration (AET) at second resolution. The first resolution is higher than the second resolution; NDVI data has finer spatial details, while AET data has relatively lower spatial resolution.

[0056] S1002. The actual evaporation time series data at the second resolution is upsampled to the first resolution to obtain the actual evaporation time series data at the first resolution.

[0057] The actual evapotranspiration time-series data at the second resolution is upsampled to the first resolution to obtain the actual evapotranspiration time-series data at the first resolution, thus matching the two datasets in terms of spatial resolution. Resampling can employ interpolation methods, such as bilinear interpolation or cubic convolution interpolation, to estimate the AET value after resolution upsampling as accurately as possible.

[0058] S1003. Randomly generate multiple sample points within the study area, and extract the normalized vegetation index value of each sample point at each time step from the normalized vegetation index time series data as a feature.

[0059] Multiple sample points were randomly generated within the study area. These sample points will serve as the basic units for data extraction and analysis. The NDVI value of each sample point at each time step was extracted from the normalized vegetation index (NDVI) time series data as a feature. NDVI can reflect the growth status and cover of vegetation and is closely related to actual evapotranspiration, making it reasonable as an input feature for the model.

[0060] S1004. Based on the preset training sample ratio, randomly select multiple training samples from all sample points, and use the remaining samples as validation samples. The training samples are used for model learning and parameter tuning, while the validation samples are used to evaluate the model's performance and generalization ability.

[0061] S1005. Input the features of the training samples and the corresponding actual evapotranspiration values ​​into the random forest model for training to obtain the trained model.

[0062] We selected a random forest model as the prediction model, setting the number of trees to 50 and the number of seeds to 6. The preset training sample ratio was 0.001, with other parameters set to default. Random forest is an ensemble learning method that improves the accuracy and stability of a model by constructing multiple decision trees and combining their predictions. The features (NDVI values) of the training samples and their corresponding real evapotranspiration values ​​were input into the random forest model for training. During training, the model learns the complex relationship between features and actual evapotranspiration, adjusting its internal parameters to minimize prediction error.

[0063] S1006. Input the features of the verification sample and the corresponding actual evapotranspiration value into the trained model, and obtain the predicted actual evapotranspiration value output by the trained model.

[0064] S1007. Calculate the model performance index based on the predicted actual evapotranspiration value and the actual actual evapotranspiration value of each validation sample, and evaluate the trained model based on the model performance index. Repeat the model training and evaluation process until the maximum number of iterations is reached or the model performance index meets the preset requirements, and obtain the trained random forest model as the actual evapotranspiration data prediction model.

[0065] Model performance metrics, such as root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R²), are calculated based on the predicted and actual evapotranspiration values ​​of each validation sample. These metrics reflect the model's prediction accuracy and goodness of fit.

[0066] The trained model is evaluated based on model performance metrics. If the model performance metrics do not meet the preset requirements or the maximum number of iterations is not reached, the model training and evaluation process is repeated, adjusting model parameters or optimizing the training strategy until the model performance meets the requirements or the maximum number of iterations is reached. The finally trained random forest model is obtained as the prediction model for actual evapotranspiration data. The above preset requirements can be set according to actual needs; for example, a model validation R² value of 0.85 is considered to meet the preset requirements.

[0067] This invention uses high-resolution NDVI time-series data as model input features to capture more refined information on the dynamic changes in vegetation. This information is closely related to actual evapotranspiration, which helps improve the model's prediction accuracy. The random forest model, by integrating multiple decision trees, can handle complex nonlinear relationships, reduce the risk of overfitting, and improve the model's generalization ability and prediction stability. Simultaneously, through reasonable parameter settings and model optimization, the model's prediction accuracy can be further improved. Resampling low-resolution actual evapotranspiration data to high resolution, matching it spatially with high-resolution NDVI data, solves the compatibility problem between data of different resolutions, enhances data applicability, and allows the model to fully utilize multi-source data for prediction. Randomly generating sample points and dividing the sample within the study area ensures the representativeness and diversity of the samples, enabling the model to learn the actual evapotranspiration characteristics at different locations and times within the study area, improving the model's applicability in different scenarios. Training the model using time-series data can capture the changing patterns of actual evapotranspiration over time.

[0068] The aforementioned precipitation data, actual evapotranspiration data, and normalized vegetation index (NVI) data were integrated in a time series format to form a remote sensing time-series dataset. The dataset underwent preprocessing, including data cleaning, missing value handling, and outlier detection, to ensure data accuracy and reliability. The processed dataset was saved in a standardized data format for easy subsequent data analysis and processing. Through these steps, a remote sensing time-series dataset containing farmland precipitation data, actual evapotranspiration data, and NVI data for the target area can be obtained, providing data support for subsequent construction of a three-dimensional feature space and identification of irrigated winter wheat.

[0069] S120. Construct a three-dimensional feature space composed of precipitation data, actual evapotranspiration data, and normalized vegetation index data.

[0070] Considering that irrigation events can impact evapotranspiration, this invention constructs a Pre-AET feature space from the perspective of the water cycle (precipitation and evapotranspiration) between the atmosphere and the biosphere (including soil). Leveraging the advantage of NDVI (Natural Difference Vegetation Index) in representing crop growth, the two-dimensional Pre-AET feature space is elevated to a three-dimensional feature space to detect whether farmland is irrigated. Specifically, precipitation data, actual evapotranspiration data, and Normalized Difference Vegetation Index (NDVI) data are combined to construct a three-dimensional feature space. Precipitation data and actual evapotranspiration data form the two-dimensional feature plane, while the NDVI data serves as the third dimension. This three-dimensional feature space construction comprehensively considers the relationship between water conditions (precipitation and evapotranspiration) and vegetation growth status (NDVI), providing a more comprehensive perspective for subsequent irrigation analysis.

[0071] S130. Fit a rainfed line in a two-dimensional feature plane formed by precipitation data and actual evapotranspiration data, wherein the rainfed line represents the relationship between the actual evapotranspiration data and precipitation data of rainfed crops.

[0072] Reference Figure 2 As shown, a rainfed line is fitted within the two-dimensional feature plane formed by precipitation data and actual evapotranspiration data. The rainfed line characterizes the distribution trajectory of crops in a three-dimensional feature space under rainfed conditions (i.e., relying solely on natural precipitation without irrigation). The rainfed line represents the relationship between actual evapotranspiration data and precipitation data for rainfed crops (i.e., crops that grow entirely dependent on natural precipitation). Mathematical statistical methods, such as linear regression, can be used to fit a linear relationship curve between precipitation data and actual evapotranspiration data, i.e., the rainfed line, based on a large sample of rainfed crop data. The rainfed line is used to distinguish between rainfed crops and irrigated crops.

[0073] S140. Determine the vertical irrigation line value from any winter wheat pixel in the three-dimensional feature space to the rain-fed line, and determine the slope length index based on the vertical irrigation line value and the normalized vegetation index data.

[0074] Reference Figure 2 As shown, any winter wheat pixel in the three-dimensional feature space has a vertical distance AD ​​from the rainfed line, which is the Perpendicular Irrigation Line (PIL). Figure 2 In this context, A'(a,b,z) is a winter wheat pixel in the three-dimensional feature space, and A(a,b) is the projection of that pixel onto the two-dimensional feature plane. For any winter wheat pixel in the three-dimensional feature space, determine its vertical irrigation line value to the rainfed line. The vertical irrigation line value reflects the differences between this winter wheat pixel and rainfed crops in terms of water conditions and vegetation growth.

[0075] The slope length index (SLI) is determined based on vertical irrigation line values ​​and normalized difference in vegetation index (NDVI) data. The SLI quantifies the intensity of the irrigation signal. It is a comprehensive index that combines the length of the vertical irrigation line (reflecting the degree of water variation) with the NDVI data (reflecting vegetation growth status), providing a more accurate description of the irrigation characteristics of winter wheat pixels. (Reference) Figure 2 As shown, the color bars represent the Slope Length Index (SLI) values, ranging from 0 to 1. Different colors correspond to different SLI values; the redder the color, the larger the SLI value and the stronger the irrigation signal; the bluer the color, the smaller the SLI value and the weaker the irrigation signal.

[0076] S150. Determine the irrigation attributes of each winter wheat pixel in the three-dimensional feature space based on the slope length index, and map it to the pre-acquired winter wheat crop layer to obtain an irrigation and rainfed classification map; the irrigation attributes include irrigated winter wheat and rainfed winter wheat.

[0077] The winter wheat crop layer was determined through methods such as remote sensing image interpretation and field surveys. It identifies the planting area of ​​winter wheat within the study region. Each pixel in the layer corresponds to a specific geographical area and contains attribute information indicating whether winter wheat is planted in that area. Remote sensing image interpretation refers to using multispectral or hyperspectral remote sensing images and, based on the spectral characteristics of winter wheat at different growth stages, employing classification algorithms (such as maximum likelihood classification and support vector machine classification) to extract the winter wheat planting areas from the images, generating the winter wheat crop layer. Field surveys involve recording the actual planting location and extent of winter wheat through on-site visits and sampling, and then digitizing this information to form the winter wheat crop layer.

[0078] The winter wheat crop layer covers the geographical area of ​​the study region, with each pixel corresponding to a specific geographical region, typically possessing a certain spatial resolution (e.g., 30 meters, 1 kilometer, etc.). Each pixel in the layer contains attribute information, primarily indicating whether winter wheat is grown in that area. The winter wheat crop layer is the foundation for generating the irrigation and rainfed classification map. It determines the planting range of winter wheat within the study region, allowing subsequent irrigation attribute determination processes to be targeted within the winter wheat planting area, avoiding unnecessary calculations and analyses in non-planted areas, thus improving work efficiency and the accuracy of classification results.

[0079] By setting appropriate thresholds, winter wheat pixels with a slope length index greater than or equal to a preset threshold are classified as irrigated winter wheat, while those less than the preset threshold are classified as rainfed winter wheat. The thresholds can be determined through methods such as field surveys and statistical analysis. For example, a certain number of known irrigated and rainfed winter wheat samples can be selected, their slope length indices calculated, and then an appropriate threshold determined based on the sample distribution.

[0080] Winter wheat pixels in the 3D feature space, whose irrigation attributes (irrigated winter wheat or rainfed winter wheat) have been determined, are mapped to a pre-acquired winter wheat crop layer according to their geographical location information. Specifically, for each pixel in the winter wheat crop layer, if there is a corresponding winter wheat pixel in the 3D feature space, the irrigation attribute of that corresponding pixel is assigned to the pixel in the winter wheat crop layer. After mapping, each pixel in the winter wheat crop layer has irrigation attribute information, thus generating an irrigation-rainfed classification map. This map visually shows which areas in the winter wheat planting area within the study region are irrigated and which are rainfed.

[0081] The method for identifying irrigated winter wheat in the field provided by this invention effectively solves the problem of insufficient mechanistic rationality and interpretability caused by the reliance on a single vegetation index in existing irrigation mapping methods. This is achieved through key steps such as constructing a three-dimensional feature space, fitting a rainfed line, and calculating the slope length index (SLI). Specifically, the method utilizes not only Normalized Difference Vegetation Index (NDVI) data but also precipitation data (Pre) and Actual Evapotranspiration (AET) data to construct a three-dimensional feature space. This fusion of multi-dimensional data allows irrigation identification to go beyond a single vegetation index, comprehensively considering the water supply and demand relationship and evapotranspiration process of crop growth, thereby improving the method's mechanistic rationality and interpretability. Fitting a rainfed line within a two-dimensional feature plane formed by precipitation data and actual evapotranspiration data clarifies the relationship between actual evapotranspiration data and precipitation data for rainfed crops. This step not only provides a theoretical basis for irrigation identification but also makes the identification results more interpretable, as the different distributions of irrigated and rainfed fields on both sides of the rainfed line reflect their essential differences in water use. The slope length index combines vertical irrigation line values ​​(representing irrigation probability) and normalized vegetation index data (representing crop growth) to form a comprehensive indicator. This indicator not only reflects the impact of irrigation on crop growth but also embodies the crop's own growth status, thereby further enhancing the method's mechanistic and interpretable aspects.

[0082] This invention utilizes a remote sensing spatiotemporal fusion method to perform spatiotemporal filtering on NDVI data, achieving a high spatial resolution (e.g., 8 days, 30m). Acquiring this high-resolution data enables irrigation monitoring at the field scale. By constructing a three-dimensional feature space and calculating the slope length index, this method can accurately identify the irrigation status of different fields. This precise identification at the field scale overcomes the shortcomings of existing irrigation mapping methods, such as low spatial resolution and difficulty in capturing irrigation conditions in highly fragmented farmland. This invention has been validated in different regions and environments, demonstrating its efficiency and convenience in field-scale irrigated crop mapping. For example, referring to… Figure 3As shown, the combination of maps and curves visually illustrates the spatial distribution of winter wheat irrigated and rain-fed planting areas in a certain region in 2019, as well as the changes in the area of ​​the two planting types at different longitudes and latitudes.

[0083] Figure 3 In the map, Irri represents irrigated winter wheat, shown in blue. NIrri represents rainfed winter wheat, shown in green. Boundary indicates the region boundary.

[0084] The blue areas represent irrigated winter wheat growing areas. The denser and darker the blue, the more concentrated the distribution of irrigated winter wheat or the relatively large area it may cover. The green areas represent rainfed winter wheat growing areas, that is, winter wheat areas that mainly rely on natural rainfall for growth.

[0085] The horizontal axis of the curve above represents longitude, used to locate the region along the longitude direction. The vertical axis represents area, measured in square kilometers (km²). 2 The figure reflects the area distribution of irrigated (Irri, blue curve) and rainfed (NIrri, green curve) winter wheat at different longitudes. It can be seen that the area of ​​irrigated or rainfed winter wheat varies significantly in certain longitude ranges.

[0086] The horizontal axis of the curve on the right also represents area, with the unit being square kilometers (km). 2 The vertical axis represents latitude, used to locate the region's position in the latitudinal direction. This curve shows the area distribution of irrigated (blue) and rainfed (green) winter wheat at different latitudes, revealing the changing trends of the two types of winter wheat area as latitude changes.

[0087] In the identification of winter wheat irrigation in this region, the method of this invention demonstrated advantages of high spatial resolution and high accuracy, providing better scientific guidance for fine-scale grain planting decisions. This method achieves high-precision irrigation identification at the field scale through spatiotemporal fusion technology, possessing technical advantages of clear physical mechanisms and strong crop specificity, providing reliable decision support for precision agriculture management. The identification accuracy for rainfed wheat was (PA=0.976, UA=0.941, F1=0.958), and the accuracy for irrigated wheat was (PA=0.972, UA=0.988, F1=0.980). It can be seen that the method of this invention has greater advantages in irrigated wheat identification with higher spatial resolution and higher accuracy, providing better scientific guidance for fine-scale grain planting decisions.

[0088] PA represents the proportion of samples that were actually irrigated winter wheat but were correctly classified as irrigated winter wheat. UA represents the proportion of samples that were classified as irrigated winter wheat but were actually indeed irrigated winter wheat.

[0089] The F1 score is the harmonic mean of precision and recall, taking into account both the precision and recall of the method of this invention. Precision refers to the proportion of samples predicted as positive that are actually positive, while recall refers to the proportion of samples that are actually positive that are correctly predicted as positive. The F1 score can comprehensively evaluate the classification performance of irrigated winter wheat.

[0090] This invention comprehensively utilizes precipitation data, actual evapotranspiration data, and normalized difference vegetation index (NDVI) data, fully considering the interrelationship between water conditions and vegetation growth, thus more accurately reflecting the irrigation status of winter wheat. Compared to traditional single-data source methods, it effectively reduces errors and improves the accuracy of irrigation information acquisition. By constructing a three-dimensional feature space, the distribution of winter wheat pixels in the dimensions of water and vegetation growth can be more intuitively displayed, providing a more scientific basis for judging irrigation attributes. Utilizing remote sensing technology to acquire time-series data enables rapid and efficient coverage of large areas of farmland, achieving real-time monitoring and dynamic updates of irrigation information. Compared to traditional ground survey methods, it significantly improves work efficiency and reduces costs. This invention's method can be automated through computer programs, rapidly analyzing and classifying large amounts of remote sensing data, and acquiring irrigation attribute information for large areas of farmland in a short time. By accurately identifying irrigated and rainfed winter wheat, targeted irrigation strategies can be formulated based on farmland with different irrigation attributes, achieving precision irrigation and improving water resource utilization efficiency.

[0091] In some embodiments, the present invention, based on the differences in the ecological implications of irrigation and rainfed agriculture, fits the lower envelope of a two-dimensional feature space to construct a rainfed line. By fitting the rainfed line, winter wheat fields under rainfed conditions can be accurately identified within a two-dimensional feature plane composed of precipitation data and actual evapotranspiration data, providing a scientific basis for irrigation identification, improving the accuracy and reliability of irrigation identification, and supporting field-scale irrigation monitoring and agricultural water resource management.

[0092] The fitting of rainfed lines in the two-dimensional feature plane formed by precipitation data and actual evapotranspiration data as described in S130 above includes: S1301. Using precipitation data as the horizontal axis and actual evapotranspiration data as the vertical axis, all winter wheat pixel data pairs form a two-dimensional feature space; where the data pair is a combination of precipitation data and actual evapotranspiration data.

[0093] Precipitation data (Pre) and actual evapotranspiration data (AET) were extracted from the acquired remote sensing time-series dataset, targeting winter wheat fields within the target area. A feature space was constructed on a two-dimensional feature plane, using precipitation data as the x-axis and actual evapotranspiration data as the y-axis. Each winter wheat pixel corresponds to a data pair, namely a set of precipitation and actual evapotranspiration data (Pre, AET), which together describe the moisture status of that pixel. All Pre-AET data pairs for winter wheat pixels constitute the point set in this two-dimensional feature space, with each point representing the moisture status of a winter wheat field at a specific time point.

[0094] S1302. Fit the envelope in the two-dimensional feature space based on the data of all winter wheat pixels to obtain the rainfed line.

[0095] The envelope refers to the theoretically co-increasing trend of actual evapotranspiration (AET) and precipitation (Pre) of rainfed crops within a two-dimensional feature space; that is, as precipitation increases, actual evapotranspiration also increases accordingly (provided it does not exceed the maximum transpiration capacity of the crop). Mathematical statistical methods (such as linear regression and nonlinear fitting) are used to fit data pairs of all winter wheat pixels to find the lower envelope that encloses these data points. This envelope represents the relationship between actual evapotranspiration and precipitation in winter wheat fields under rainfed conditions and is called the rainfed line.

[0096] Rainwater-fed lines can be represented by a mathematical expression, such as a linear equation. , in, and These represent the slope and intercept of the rainfed line, respectively, which are also the slope and intercept of the envelope fitting. Indicates winter wheat as a unit The actual evapotranspiration data, where x represents a winter wheat pixel.

[0097] This invention, by fitting a rainfed line, can clearly distinguish between irrigated and rainfed fields. The actual evapotranspiration of irrigated fields is typically higher than the rainfed line because irrigation provides additional moisture. Compared to traditional methods, the introduction of the rainfed line helps reduce misclassification caused by ignoring the impact of irrigation, improving the accuracy of irrigation identification. The rainfed line fitting is based on the physical processes of crop water supply and demand mechanisms, reflecting the impact of irrigation on surface-atmosphere moisture transport, enhancing the mechanistic and interpretable nature of irrigation identification methods. The rainfed line provides a more scientific explanation of the impact of irrigation on crop growth and the water cycle, offering a solid theoretical foundation for irrigation identification. At the field scale, the moisture status of irrigated and rainfed fields differs significantly. Rainfed line fitting helps accurately identify irrigated fields at the field scale, meeting the needs of high-resolution irrigation monitoring. Rainfed lines can provide more refined irrigation guidance, helping farmers rationally adjust irrigation strategies and improve water resource utilization efficiency. In multi-season crop planting areas (such as rice-wheat or wheat-corn rotation areas), rainfed line fitting helps distinguish the irrigation distribution of different crops. By accurately identifying the irrigation status of each crop, data confusion or misclassification can be avoided, thereby improving the accuracy and reliability of irrigation mapping products.

[0098] In some embodiments, in the field of agricultural remote sensing and irrigation mapping, a key issue is how to distinguish between irrigated crops and rainfed crops using remote sensing data (such as vegetation index, surface temperature, etc.). By constructing a three-dimensional feature space, features of irrigated crops can be extracted, but further quantification of irrigation intensity is needed. The determination of the vertical irrigation line value from any winter wheat pixel in the three-dimensional feature space to the rainfed line, as described in S140 above, and the determination of the slope length index based on the vertical irrigation line value and the normalized vegetation index data, includes: S1401. Calculate the vertical irrigation line value from any winter wheat pixel in the three-dimensional feature space to the rainfed line, and normalize the vertical irrigation line values ​​of all winter wheat pixels to obtain standardized vertical irrigation line values.

[0099] In the Pre-AET-NDVI 3D feature space, the vertical irrigation line value represents the vertical distance from winter wheat pixel x to the rainfed line. This distance reflects the degree to which the NDVI of winter wheat pixel x deviates from the expected NDVI under rainfed conditions. It can be used to indicate the likelihood of irrigation; that is, the closer a pixel is to the rainfed line, the more likely it is to rely on non-irrigated rainfall for its growing season water requirements, and vice versa. The vertical irrigation line value is determined by the following formula: in, This represents the vertical irrigation line value of any winter wheat pixel in the three-dimensional feature space. and These represent the slope and intercept of the rainfed line, respectively. and Representing winter wheat pixels The actual evapotranspiration data and precipitation data.

[0100] The calculated vertical irrigation line value PIL(x) is normalized to eliminate the influence of different regions or time scales and to facilitate subsequent NDVI coupling. The normalization formula is: Wherein, NPIL(x) represents the standardized vertical irrigation line value of any winter wheat pixel in the three-dimensional feature space, with a value range of 0 to 1. PIL(x) min and PIL(x) max These represent the minimum and maximum values ​​of the vertical irrigation line values ​​for all winter wheat pixels in the three feature spaces, respectively.

[0101] S1402. Determine the slope length index based on standardized vertical irrigation line values ​​and normalized vegetation index data.

[0102] The slope length index is determined by the following formula: in, Represents any winter wheat pixel in the three-dimensional feature space The slope length index, and Representing winter wheat pixels Standardized vertical irrigation line values ​​and normalized vegetation index data.

[0103] Both the Vertical Irrigation Line Value (PIL(x)) and the Slope Length Index (SLI) can quantify the irrigation signal intensity of winter wheat pixels in the three-dimensional feature space. PIL(x) directly reflects the degree to which a pixel deviates from rainfed conditions by calculating the vertical distance from the winter wheat pixel to the rainfed line; SLI further enhances the quantification of the irrigation signal by quantifying the change in NDVI along the vertical irrigation line direction. A larger SLI value indicates a greater impact of irrigation on the pixel.

[0104] Rainfed crops typically have lower PIL(x) and SLI values ​​because their water supply relies primarily on natural precipitation, resulting in relatively gradual changes in NDVI. Irrigated crops, on the other hand, have higher PIL(x) and SLI values ​​because irrigation compensates for insufficient natural precipitation, leading to increased actual evapotranspiration and consequently higher NDVI. Therefore, these two indicators can effectively distinguish between rainfed and irrigated crops.

[0105] Using PIL(x) and SLI, irrigation zone boundaries can be mapped more accurately, especially in areas where rainfed and irrigated crops are intercropped. Traditional irrigation mapping methods may struggle to accurately distinguish between these two crop types, while vertical irrigation lines and slope length indices based on the Pre-AET-NDVI 3D feature space provide more scientific quantitative indicators, enabling clearer identification of irrigation zones and improving the accuracy of irrigation mapping.

[0106] In some embodiments, the slope length index of the present invention considers not only the crop's moisture status (represented by NPIL(x)) but also the crop's growth status (represented by NDVI(x)), thus possessing stronger comprehensive analytical capabilities. In practical applications, it can better cope with complex agricultural ecological environments and diverse planting patterns, providing more valuable decision-making basis for agricultural research and production. To more accurately identify irrigated winter wheat, NDVI is added as the normal plane of NPIL. Then, the length of the NDVI of any winter wheat pixel in the three-dimensional feature space and NPIL can form a triangle. The length of the hypotenuse of this triangle can be regarded as a slope length, i.e., the slope length index. This slope is the plane formed by point A' and the rainfed line in the three-dimensional feature space. Figure 2 The slope length index is determined by the following formula: in, Represents any winter wheat pixel in the three-dimensional feature space The slope length index, and Representing winter wheat pixels Standardized vertical irrigation line values ​​and normalized vegetation index data.

[0107] The Slope Length Index (SLI) is calculated by combining the standardized vertical irrigation line value (NPIL) (x) and the normalized difference in vegetation index (NDVI) (x). NPIL (x) reflects the degree to which a pixel deviates from the rainfed line in the three-dimensional feature space, i.e., the impact of irrigation on crop water supply; NDVI (x) directly reflects the crop growth status and vegetation cover. By combining these two indices, SLI (x) can more comprehensively and accurately quantify the intensity of the irrigation signal. For example, when both NPIL (x) and NDVI (x) are high, it indicates that the pixel not only deviates far from the rainfed line (significant irrigation impact) but also that the crop is growing vigorously, further confirming the effectiveness of irrigation. The value of SLI (x) will also be correspondingly high, thus highlighting the irrigation signal more clearly. The magnitude of SLI indicates that wheat with a higher NPIL and better growth is more likely to be irrigated, and vice versa.

[0108] This formula amplifies and integrates the variations of NPIL(x) and NDVI(x) through square, summation, and square root operations. This processing makes the differences in SLI(x) values ​​under different irrigation levels more significant, enhancing the discriminative power of irrigation signals. In areas where rainfed and irrigated crops are intercropped, it can more clearly distinguish between the two, improving the accuracy of identifying irrigated crops.

[0109] In actual farmland environments, various factors can influence crop growth and water conditions, such as soil type and topography. SLI(x) comprehensively considers both NPIL(x) and NDVI(x), enabling it to more accurately reflect crop irrigation status in complex environments, reducing misjudgments caused by single factors, and improving the accuracy of distinguishing between rainfed and irrigated crops. Furthermore, SLI(x) values ​​can be used to create more precise irrigation area maps. On the map, areas with larger SLI(x) values ​​can be marked as irrigated areas, while areas with smaller values ​​are rainfed areas. Due to the enhancement and improved discrimination of the irrigation signal by SLI(x), the boundaries of irrigated areas can be defined more clearly, reducing boundary ambiguity and making irrigation mapping more accurate. SLI(x) is a continuous numerical indicator that spatially reflects changes in irrigation levels. When creating irrigation maps, this continuity allows the maps to better display the spatial distribution characteristics of irrigated areas, avoiding the discontinuity of map information caused by discrete classification, further improving the accuracy and practicality of irrigation mapping.

[0110] Agricultural managers can understand the irrigation demand and intensity of different regions based on the value of SLI(x). When water resources are limited, irrigation resources should be prioritized for areas with higher SLI(x) values, i.e., areas with higher irrigation demand and greater dependence on irrigation for crop growth, thereby improving the efficiency of water resource utilization.

[0111] The calculation of SLI(x) relies on the standardized vertical irrigation line value NPIL(x) and the normalized difference vegetation index NDVI(x), which are derived from different data sources and processing procedures. NPIL(x) is calculated based on the rainfed line and pixel location in the three-dimensional feature space, involving multi-source data such as precipitation and evapotranspiration; NDVI(x) comes directly from remote sensing imagery. By fusing these data together through SLI(x), the advantages of multi-source data are fully utilized, providing more comprehensive and accurate crop irrigation information.

[0112] In some embodiments, rain-fed crops typically have a lower NPIL(x) because their water supply relies primarily on natural rainfall, crop growth is water-limited, and NDVI(x) is also relatively low. According to the formula for calculating SLI(x), the value of SLI(x) will be lower in this case. Irrigated crops, on the other hand, have a higher NPIL(x), and due to the supplemental water provided by irrigation, they grow well, resulting in a higher NDVI(x) and consequently a higher SLI(x) value. This invention addresses this by setting a threshold... Furthermore, by classifying the irrigation attributes of each winter wheat pixel based on the actual evapotranspiration, precipitation, and irrigation water supply and demand during the winter wheat growing season, rainfed crops and irrigated crops can be effectively distinguished. The above-mentioned step S150, which determines the irrigation attributes of each winter wheat pixel in the three-dimensional feature space based on the slope length index and maps them to a pre-acquired winter wheat crop layer to obtain an irrigation-rainfed classification map, includes: For each winter wheat pixel in the three-dimensional feature space: When the slope length index of winter wheat pixels is greater than or equal to a preset threshold In the case where the winter wheat pixel's irrigation attribute is determined to be irrigated winter wheat; when the slope length index of the winter wheat pixel is less than a preset threshold... In this case, the irrigation attribute of the winter wheat pixel is determined to be rain-fed winter wheat.

[0113] Where Irri_HNP(x) is the irrigation attribute of any winter wheat pixel in the three-dimensional feature space, where 1 represents irrigated winter wheat and 0 represents rain-fed winter wheat.

[0114] Determining the preset threshold requires comprehensive consideration of factors such as the characteristics of the study area, crop planting conditions, and practical application needs. It can be determined through statistical analysis or field verification. Statistical analysis involves statistically analyzing the slope length index of all winter wheat pixels within the study area, such as calculating its mean and standard deviation. Based on the statistical distribution characteristics, a suitable value is selected as the threshold; for example, the mean can be increased by a certain multiple of the standard deviation to ensure a good distinction between irrigated and rainfed winter wheat. Field verification involves selecting a certain number of sample points within the study area for field investigation, recording the actual irrigation conditions (irrigated or rainfed) at these points. Then, the slope length index of these sample points is calculated, and by analyzing the relationship between the actual irrigation conditions and the slope length index, a threshold that can accurately distinguish between the two is determined.

[0115] For each winter wheat pixel in the three-dimensional feature space, its calculated slope length index SLI(x) is compared with a preset threshold: If SLI(x) is greater than or equal to the preset threshold, it indicates that the irrigation signal of the cell is strong, the deviation from the rain-fed line is large, and the crop growth is good, which is consistent with the characteristics of irrigated winter wheat. Therefore, the irrigation attribute of the winter wheat cell is determined to be irrigated winter wheat.

[0116] If SLI(x) is less than the preset threshold, it indicates that the irrigation signal of the cell is weak and close to the characteristics under rainfed conditions. Therefore, the irrigation attribute of the winter wheat cell is determined to be rainfed winter wheat.

[0117] This invention effectively distinguishes between irrigated and rainfed winter wheat by combining a slope length index and a preset threshold. The slope length index comprehensively considers standardized vertical irrigation line values ​​and normalized difference vegetation index, fully reflecting the crop's water status and growth state. The preset threshold acts as a classification boundary, allowing winter wheat pixels with different irrigation attributes to be clearly distinguished, thus improving classification accuracy. When mapping the distribution of irrigated winter wheat in the field, this method can more accurately identify irrigated and rainfed areas. Because the slope length index accurately quantifies irrigation signals, and the threshold judgment can clearly define the irrigation attributes of pixels, the generated irrigation map more realistically reflects the actual situation, with clearer boundaries, reducing misjudgments and ambiguous areas, and providing a more reliable basis for agricultural resource management and planning.

[0118] In actual farmland environments, various factors influence the process, such as soil type, topographical differences, and climate fluctuations. The slope length index calculation comprehensively considers multiple factors and can adapt to these complex environmental conditions. The preset threshold can also be adjusted according to different regional characteristics, making the method of this invention applicable to different research areas and planting patterns, and able to accurately identify irrigated winter wheat.

[0119] Figure 4 The technical process for identifying winter wheat irrigated in the field is demonstrated. The technical process mainly includes three stages: data preparation, algorithm processing, and irrigation identification.

[0120] 1. In the data preparation stage, the first step is to acquire a 30m resolution layer of winter wheat crop data for each year (i.e., ...). Figure 4 Using the winter wheat layer in the image as the base data, we combined the Landsat-8 surface reflectance dataset with a temporal resolution of 30m / 16 days with the NDVI dataset that was improved to 30m / day through spatiotemporal fusion. At the same time, we integrated the daily evapotranspiration dataset and the monthly precipitation dataset with a resolution of 1 km. We then downscaled all the data to a spatial resolution of 30m.

[0121] 2. In the algorithm processing stage, the prepared data is mainly used to construct a three-dimensional feature space, delineate the rain-fed line, and calculate the slope length index, ultimately achieving the identification of irrigated winter wheat. The specific steps are as follows: Three-dimensional feature space construction: A three-dimensional feature space is constructed based on precipitation, evapotranspiration, and NDVI datasets. This space uses precipitation data, actual evapotranspiration data, and normalized difference vegetation index (NDVI) data as three dimensions to comprehensively analyze the crop growth environment and water status. The three-dimensional feature space uses precipitation data (Pre) as the X-axis, actual evapotranspiration data (AET) as the Y-axis, and NDVI as the Z-axis.

[0122] Rainfed line delineation: Rainfed lines are delineated by fitting data from the Pre-AET two-dimensional feature plane, thus distinguishing between natural precipitation and irrigation supplementation. The rainfed line represents the distribution trajectory of crops in a three-dimensional feature space under conditions relying solely on natural precipitation.

[0123] Slope length index calculation: The slope length index is calculated based on rainfed lines and data in the 3D feature space. This index quantifies the degree of NDVI variation of a pixel along the vertical irrigation line, reflecting the intensity of the irrigation signal. Specifically, it determines the vertical irrigation line value from any winter wheat pixel in the 3D feature space to the rainfed line, and then determines the slope length index based on the vertical irrigation line value and normalized difference vegetation index (NDVI) data. 3. During the irrigation identification phase, a preset threshold is used. The study area was divided into different irrigation categories. Each winter wheat pixel was classified as irrigated winter wheat (SLI(x)≥...). ) or rainfed winter wheat (SLI(x) < This generates a 30m resolution classification map of irrigation and rain-fed wheat, which visually displays the distribution of irrigated winter wheat in the field.

[0124] 4. A confusion matrix is ​​generated using measured survey data, other existing irrigation products, and the irrigation-rain-fed classification map. The accuracy of the irrigation-rain-fed classification map is then assessed using the confusion matrix.

[0125] A confusion matrix is ​​an important tool for evaluating the accuracy of irrigation rainfed classification maps. When generating a confusion matrix using measured survey data, other existing irrigation products, and irrigation rainfed classification maps, the following steps can be taken: (1) Prepare data, including obtaining data from actual survey points and existing irrigation products.

[0126] Measured survey data refers to the actual irrigation conditions (irrigated or rainfed) of each sample point within the study area, obtained through methods such as field investigation, questionnaire survey, and sensor monitoring. These sample points should have clear geographical location information, represented by latitude and longitude coordinates. The measured survey data should be compiled into a table, recording the sample point's number, geographical location, and actual irrigation attribute (irrigated winter wheat or rainfed winter wheat).

[0127] Existing irrigation product data refers to other existing irrigation products, specifically existing irrigation-related information products, such as irrigation area maps published by the agricultural sector, and irrigation datasets generated based on remote sensing or other models.

[0128] The existing irrigation product data is matched with the geographical locations of the measured survey points, and the predicted irrigation attributes corresponding to each measured sample point in the existing irrigation products are extracted.

[0129] The irrigation and rainfed classification map contains the predicted irrigation attribute for each pixel (corresponding to a specific geographic area). By matching the geographic location of the surveyed points with the irrigation and rainfed classification map, the predicted irrigation attribute (i.e., irrigation or rainfed) for each surveyed point in the map is obtained.

[0130] (2) Constructing the confusion matrix: Based on the measured survey data from each sampling point, the corresponding predicted irrigation attributes in existing irrigation products, and the predicted irrigation attributes in the irrigation-rain-fed classification map, a confusion matrix is ​​constructed. The confusion matrix is ​​a 2×2 square matrix, where rows represent the actual category and columns represent the predicted category. The specific filling method is shown in Table 1 below: Table 1

[0131] True Instance (TP): The number of sample points that are actually irrigated winter wheat in the actual survey data and are also predicted to be irrigated winter wheat in the irrigation rainfed classification map and other existing irrigation products.

[0132] False negatives (FN): The number of sample points that are actually irrigated winter wheat in the actual survey data, but are predicted to be rainfed winter wheat in the irrigation rainfed classification map or other existing irrigation products.

[0133] False positives (FP): The number of sample points that are actually rainfed winter wheat in the actual survey data, but are predicted to be irrigated winter wheat in the irrigation rainfed classification map or other existing irrigation products.

[0134] True counterexamples (TN): The number of sample points that are actually rainfed winter wheat in the actual survey data and are also predicted to be rainfed winter wheat in the irrigation rainfed classification map and other existing irrigation products.

[0135] For each measured sample point, the final prediction category is determined by comprehensively considering the prediction results of existing irrigation products and measured survey sample point data. For example, a voting mechanism can be used: if both are predicted as irrigation, the final prediction is irrigation; otherwise, it is rainfed. Alternatively, a comprehensive judgment can be made based on the weights of both. Then, the confusion matrix is ​​filled based on the measured irrigation attributes and the predicted irrigation attributes in the irrigation-rainfed classification map.

[0136] (3) Analysis and evaluation: Based on the confusion matrix, various evaluation metrics can be calculated, such as accuracy, precision, and recall.

[0137] Accuracy represents the proportion of all correctly predicted samples out of the total number of samples.

[0138] Accuracy=(TP+TN) / (TP+TN+FP+FN), Precision represents the proportion of sample points predicted to be irrigated that were actually irrigated. Precision = TP / (TP + FP) Recall represents the proportion of samples that were actually irrigated but were correctly predicted as irrigated.

[0139] Recall = TP / (TP + FN) The accuracy of the irrigation-rain-fed classification map is evaluated based on the calculated assessment indicators. High accuracy, precision, and recall indicate that the classification results closely match the actual situation; conversely, low accuracy or recall suggest a need for further analysis of error sources, such as data quality issues or classification inconsistencies, and for improvement of the methodology.

[0140] If the evaluation results show that the accuracy of the irrigation and rainfed classification map is not ideal, the problem can be analyzed based on the confusion matrix and evaluation indicators. For example, if a low recall rate is found, it indicates that the classification map may not fully identify certain irrigated or rainfed areas. In this case, the classification parameters can be adjusted, such as the preset thresholds mentioned above, to improve the recall rate of the classification map.

[0141] This invention integrates information from three different sources—field measurements, existing irrigation product data, and irrigation-rainfed classification map data—by acquiring measured survey data, existing product predictions, and classification map predictions, and then matching these data geographically. The measured survey data, with its high accuracy, serves as the benchmark for evaluation; the existing irrigation product data reflects irrigation attribute predictions based on specific technologies and methods; and the irrigation-rainfed classification map is the object to be evaluated. The fusion of multi-source data allows for evaluation of the irrigation-rainfed classification map from different perspectives, avoiding biases that may arise from a single data source, and making the evaluation results more comprehensive and objective. Each measured sample point simultaneously possesses measured survey data, the corresponding predicted irrigation attributes in existing irrigation products, and the predicted irrigation attributes in the irrigation-rainfed classification map. This multi-dimensional data structure provides rich information for constructing a confusion matrix. By comparing irrigation attributes from different sources, a deeper understanding of the performance of the irrigation-rainfed classification map under various conditions can be gained, identifying potential problems and biases.

[0142] The method for identifying irrigated winter wheat proposed in this invention combines precipitation, actual evapotranspiration data, and normalized difference vegetation index (NDE) to accurately capture the spatiotemporal distribution characteristics of irrigated crops. Through in-depth analysis of crop water supply and demand mechanisms, this method provides important reference for optimizing field irrigation volume and timing. Furthermore, this method effectively overcomes the monitoring uncertainties that may arise from using a single remote sensing vegetation index, particularly excelling in addressing the heterogeneity of irrigated crops in multi-season planting systems. The innovative index of this invention (Slope Length Index, SLI) is not only applicable to the irrigation identification of winter wheat but also has good scalability, allowing for wide application in the irrigation identification and mapping of other crop types, demonstrating broad application prospects.

[0143] The apparatus for identifying winter wheat under field irrigation provided by the present invention will be described below. The apparatus for identifying winter wheat under field irrigation described below can be referred to in correspondence with the method for identifying winter wheat under field irrigation described above.

[0144] The device for identifying winter wheat irrigated in the field provided by this invention refers to... Figure 5 As shown, it includes: The data acquisition module 210 is used to acquire remote sensing time-series datasets of farmland within the target area, including precipitation data, actual evapotranspiration data, and normalized vegetation index data. The spatial construction module 220 is used to construct a three-dimensional feature space composed of precipitation data, actual evapotranspiration data and normalized vegetation index data; wherein, precipitation data and actual evapotranspiration data constitute a two-dimensional feature plane, and normalized vegetation index data serves as the third dimension. The data fitting module 230 is used to fit a rainfed line in a two-dimensional feature plane formed by precipitation data and actual evapotranspiration data, wherein the rainfed line represents the relationship between the actual evapotranspiration data and precipitation data of rainfed crops. The index determination module 240 is used to determine the vertical irrigation line value from any winter wheat pixel in the three-dimensional feature space to the rain-fed line, and to determine the slope length index based on the vertical irrigation line value and the normalized vegetation index data. The attribute classification module 250 is used to determine the irrigation attribute of each winter wheat pixel in the three-dimensional feature space based on the slope length index, and map it to a pre-acquired winter wheat crop layer to obtain an irrigation-rainfed classification map; the irrigation attribute includes irrigated winter wheat and rainfed winter wheat. If the slope length index of a winter wheat pixel is greater than or equal to a preset threshold, the irrigation attribute of the winter wheat pixel is determined to be irrigated winter wheat; if the slope length index of a winter wheat pixel is less than the preset threshold, the irrigation attribute of the winter wheat pixel is determined to be rainfed winter wheat.

[0145] Figure 6 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 6 As shown, the electronic device may include a processor 310, a communications interface 320, a memory 330, and a communication bus 340. The processor 310, communications interface 320, and memory 330 communicate with each other via the communication bus 340. The processor 310 can call logical instructions from the memory 330 to execute a method for identifying winter wheat irrigated in the field.

[0146] Furthermore, the logical instructions in the aforementioned memory 330 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0147] On the other hand, the present invention also provides a computer program product, the computer program product including a computer program that can be stored on a non-transitory computer-readable storage medium, and when the computer program is executed by a processor, the computer is able to perform the methods for identifying winter wheat irrigated in the field provided by the above methods.

[0148] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the methods for identifying winter wheat under field irrigation provided by the methods described above.

[0149] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0150] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0151] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method of identifying irrigated winter wheat in a field, characterized in that, The application relates to a method for obtaining a winter wheat irrigation rain-fed classification map. The method comprises the following steps: acquiring remote sensing time series data of farmland in a target area, including precipitation data, actual evapotranspiration data and normalized vegetation index data; constructing a three-dimensional feature space composed of the precipitation data, the actual evapotranspiration data and the normalized vegetation index data; wherein the precipitation data and the actual evapotranspiration data constitute a two-dimensional feature plane, and the normalized vegetation index data serves as the third dimension; fitting a rain-fed line in the two-dimensional feature plane constituted by the precipitation data and the actual evapotranspiration data, wherein the rain-fed line represents the relationship between the actual evapotranspiration data and the precipitation data of rain-fed crops; determining the vertical irrigation line value of any winter wheat pixel in the three-dimensional feature space to the rain-fed line, and determining the slope length index according to the vertical irrigation line value and the normalized vegetation index data; 2. The method of identifying field-irrigated winter wheat of claim 1, wherein, determining the irrigation attribute of each winter wheat pixel in the three-dimensional feature space according to the slope length index, and mapping the irrigation attribute to a pre-acquired winter wheat crop layer to obtain an irrigation rain-fed classification map; the irrigation attribute includes irrigated winter wheat and rain-fed winter wheat; in the case that the slope length index of the winter wheat pixel is greater than or equal to a preset threshold value, the irrigation attribute of the winter wheat pixel is determined as irrigated winter wheat; in the case that the slope length index of the winter wheat pixel is less than the preset threshold value, the irrigation attribute of the winter wheat pixel is determined as rain-fed winter wheat. The normalized vegetation index data and the actual evapotranspiration data in the remote sensing time series data set are obtained by the following method: acquiring surface reflectance data from MOD09A1 and multispectral data from Landsat-8, and performing spatio-temporal fusion on the surface reflectance data and the multispectral data to obtain normalized vegetation index data of a first resolution, which serves as the normalized vegetation index data in the remote sensing time series data set; 3. The method of identifying field-irrigated winter wheat of claim 2, wherein, inputting the normalized vegetation index data of the first resolution into a pre-trained actual evapotranspiration data prediction model to obtain the actual evapotranspiration prediction value of the actual evapotranspiration data prediction model as the actual evapotranspiration data in the remote sensing time series data set. The actual evapotranspiration data prediction model is obtained by the following method: acquiring normalized vegetation index time series data of a first resolution and actual evapotranspiration time series data of a second resolution; wherein the first resolution is higher than the second resolution; resampling the actual evapotranspiration time series data of the second resolution to the first resolution to obtain actual evapotranspiration time series data of the first resolution; randomly generating a plurality of sample points in the research area, and extracting the normalized vegetation index value of each sample point at each time step from the normalized vegetation index time series data as a feature; randomly selecting a plurality of training samples from all the sample points according to a preset training sample ratio, and taking the remaining samples as verification samples; inputting the features and the corresponding actual evapotranspiration values of the training samples into a random forest model for training to obtain a trained model; wherein the number of trees of the random forest model is 50, and the seed number is 6; inputting the features and the corresponding actual evapotranspiration values of the verification samples into the trained model to obtain the predicted actual evapotranspiration values output by the trained model. The model performance indicators are calculated according to the predicted actual evapotranspiration values and the true actual evapotranspiration values of each verification sample, and the trained model is evaluated based on the model performance indicators, the model training and evaluation process is repeated until a maximum iteration number is reached or the model performance indicators meet preset requirements, and a trained random forest model is obtained as the actual evapotranspiration data prediction model.

4. The method of identifying field-irrigated winter wheat of claim 1, wherein, The method further comprises: obtaining measured survey sample data and existing irrigation product data, matching the existing irrigation product data with the geographic positions of the measured survey samples, and extracting the corresponding predicted irrigation attributes of each measured sample in the existing irrigation product; matching the geographic positions of the measured survey samples with the irrigation rain-fed classification map to obtain the predicted irrigation attributes of each measured survey sample in the irrigation rain-fed classification map; based on the measured survey sample data of each measured survey sample, the corresponding predicted irrigation attributes in the existing irrigation product, and the predicted irrigation attributes in the irrigation rain-fed classification map, a confusion matrix is constructed; based on the confusion matrix, a plurality of evaluation indicators are calculated; the plurality of evaluation indicators at least include accuracy, precision and recall rate; according to the calculated evaluation indicators, the accuracy of the irrigation rain-fed classification map is evaluated.

5. The method of identifying field-irrigated winter wheat of claim 1, wherein, The vertical irrigation line value is determined by the following formula: ； wherein, represents the vertical irrigation line value for any winter wheat pixel in the three-dimensional feature space, and respectively represent the slope and intercept of the rainfed line, and respectively represent the actual evapotranspiration data and precipitation data for the winter wheat pixel .

6. The method of identifying field-irrigated winter wheat of claim 1, wherein, The vertical irrigation line value of any winter wheat pixel in the three-dimensional feature space to the rain-fed line is determined according to the vertical irrigation line value and the normalized vegetation index data, including: calculating the vertical irrigation line value of any winter wheat pixel in the three-dimensional feature space to the rain-fed line, and normalizing the vertical irrigation line values of all winter wheat pixels to obtain standardized vertical irrigation line values; determining the slope length index based on the standardized vertical irrigation line values and the normalized vegetation index data; ； wherein, denotes the slope length exponent of any winter wheat pixel in the three-dimensional feature space , and denotes the normalized vertical irrigation line value and the normalized difference vegetation index data of the winter wheat pixel , respectively.

7. A device for identifying winter wheat fields under irrigation, characterized in that, including: The data acquisition module is configured to acquire remote sensing time series data of farmland in a target region, including precipitation data, actual evapotranspiration data and normalized vegetation index data. The space construction module is configured to construct a three-dimensional feature space composed of precipitation data, actual evapotranspiration data and normalized vegetation index data; wherein the precipitation data and the actual evapotranspiration data constitute a two-dimensional feature plane, and the normalized vegetation index data serves as the third dimension. The data fitting module is configured to fit a rain-fed line in the two-dimensional feature plane constituted by the precipitation data and the actual evapotranspiration data, wherein the rain-fed line represents the relationship between the actual evapotranspiration data and the precipitation data of rain-fed crops. The index determination module is configured to determine the vertical irrigation line value of any winter wheat pixel in the three-dimensional feature space to the rain-fed line, and determine the slope length index according to the vertical irrigation line value and the normalized vegetation index data. An attribute classification module is configured to determine irrigation attributes of each winter wheat pixel in the three-dimensional feature space according to the slope length index, and map the irrigation attributes to a pre-acquired winter wheat crop layer to obtain an irrigation and rain-fed classification map. The irrigation attributes include irrigated winter wheat and rain-fed winter wheat. In a case where the slope length index of a winter wheat pixel is greater than or equal to a preset threshold, the irrigation attribute of the winter wheat pixel is determined as irrigated winter wheat. In a case where the slope length index of a winter wheat pixel is less than the preset threshold, the irrigation attribute of the winter wheat pixel is determined as rain-fed winter wheat.

8. An electronic device comprising a memory, a processor, and a computer program stored on the memory and running on the processor, characterized in that, The computer program is executed by the processor to implement the method for identifying irrigated winter wheat in a field according to any one of claims 1 to 6. 9.A non-transitory computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the method for identifying irrigated winter wheat in a field according to any one of claims 1 to 6.

10. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to implement the method for identifying irrigated winter wheat in a field according to any one of claims 1 to 6.

Images