Method and system for accurately identifying irrigated land based on space-time two-dimensional feature fusion

Abstract

The invention discloses a method and a system for accurately identifying a irrigated land based on space-time two-dimensional feature fusion, and belongs to the technical field of remote sensing image processing and agricultural information. The method comprises the following steps: acquiring multi-source data of a target area, wherein the multi-source data comprises a remote sensing image sequence, surface water source spatial distribution data, underground water irrigation facility spatial distribution data and cultivated land pattern spot data; spatial analysis is carried out based on water source and irrigation facility data, and an irrigation condition spatial distribution diagram is generated through Euclidean distance calculation and nuclear density analysis; extracting a time sequence vegetation index of the cultivated land pattern spot data in a moisture sensitive phenological period based on the remote sensing image sequence, and calculating a stability index; constructing a deep learning model fusing the spatial semantic segmentation network and the time sequence analysis network, and performing model training by taking the spatial distribution map and the stability index as fusion input features; and identifying and outputting a irrigated land distribution map by using the trained model. The invention further discloses a corresponding system, electronic equipment and a storage medium. According to the invention, through time-space two-dimensional feature fusion, the precision of irrigated land identification is effectively improved, and the limitation of a single-dimensional method is solved.

Description

Technical Field

[0001] This invention relates to the fields of remote sensing image processing and agricultural information technology, and in particular to a method and system for accurate identification of irrigated land that combines spatial irrigation potential analysis and time-series vegetation stability monitoring. Background Technology

[0002] Irrigated land refers to arable land with a guaranteed water source and irrigation facilities, capable of normal irrigation in ordinary years. Accurate identification of irrigated land is of great significance for agricultural water management, crop yield forecasting, and agricultural policy formulation.

[0003] Existing methods for identifying irrigated land suffer from the following shortcomings: some methods extract spectral and texture features from single-temporal remote sensing images for fine-grained classification, while others focus on improving the deep learning network structure to enhance the model's feature extraction capabilities. However, most of these methods fail to effectively integrate spatial accessibility information of irrigation facilities with temporal irrigation behavior characteristics during critical crop growth periods, resulting in limited accuracy in distinguishing between complex scenarios such as "having facilities but not irrigating" and "having no facilities but sufficient rainfall."

[0004] Methods relying solely on time-series vegetation indices lack consideration for the spatial distribution of irrigation facilities, resulting in limited accuracy in areas with unevenly distributed irrigation infrastructure. Traditional machine learning methods also have limited feature extraction capabilities, making it difficult to effectively integrate multi-source, heterogeneous spatiotemporal features.

[0005] Therefore, there is an urgent need for a precise identification method that can simultaneously consider both spatial irrigation potential and temporal irrigation behavior characteristics. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method and system for accurate identification of irrigated land that can effectively integrate spatial irrigation potential and temporal vegetation index stability.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a method for accurate identification of irrigated land based on spatiotemporal dual-dimensional feature fusion, comprising the following steps: S1: Obtain multi-source data for the target area, wherein the multi-source data includes at least the remote sensing image sequence for the target year, spatial distribution data of surface water sources, spatial distribution data of groundwater irrigation facilities, and cultivated land patch data of the target area; S2: Based on the spatial distribution data of surface water sources and the spatial distribution data of groundwater irrigation facilities, perform spatial analysis to generate a spatial distribution map of irrigation conditions that characterizes irrigation potential; S3: Based on the remote sensing image sequence, extract the time series vegetation index of the cultivated land patch data during the water-sensitive phenological period, and calculate its stability index; S4: Construct a deep learning model, using the spatial distribution map of irrigation conditions and the stability index of the time series vegetation index as fusion input features to train the model and obtain an irrigated land identification model; S5: Using the trained irrigation land recognition model, identify the target area and output an irrigation land distribution map.

[0008] Furthermore, the method for obtaining the cultivated land parcel data in step S1 includes: extracting it from an existing land use database or map; or obtaining it by identifying cultivated land in the target area using a semantic segmentation model based on the remote sensing image sequence.

[0009] Further, the generation of the spatial distribution map of irrigation conditions in step S2 includes: Based on the spatial distribution data of surface water sources, the Euclidean distance from each grid cell to the nearest surface water source is calculated, and a surface water protection rate layer is generated. Based on the spatial distribution data of the groundwater irrigation facilities, kernel density analysis is performed to generate a groundwater availability layer. The surface water security rate layer and the groundwater availability layer are weighted and superimposed to generate a comprehensive spatial distribution map of irrigation conditions.

[0010] Furthermore, in step S3, the water-sensitive phenological period may include June to August; the vegetation index may be NDVI; and the stability index may be the coefficient of variation or standard deviation.

[0011] Furthermore, the deep learning model described in step S4 is a hybrid model that integrates a spatial semantic segmentation network and a time series analysis network.

[0012] Furthermore, the spatial semantic segmentation network is used to extract the spatial features of the irrigation condition spatial distribution map; the time series analysis network is used to extract the temporal features of the time series vegetation index stability index.

[0013] Furthermore, the spatial semantic segmentation network adopts the U-Net network structure, and the time series analysis network adopts the LSTM or Transformer structure.

[0014] Furthermore, in step S4, when training the deep learning model, the loss function used is a composite loss function, which includes classification cross-entropy loss and a spatial neighborhood smoothing regularization term used to constrain the spatial continuity of the recognition results.

[0015] Secondly, the present invention provides a precise identification system for irrigated land based on spatiotemporal dual-dimensional feature fusion, comprising: The data acquisition module is used to perform step S1 of claim 1; The spatial analysis module is used to perform step S2 of claim 1; The time analysis module is used to execute step S3 of claim 1; The model building and recognition module is used to perform steps S4 and S5 of claim 1.

[0016] Thirdly, the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the above-described method.

[0017] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method.

[0018] The beneficial effects of this invention are: 1. By combining spatial irrigation potential analysis with temporal vegetation stability monitoring, the limitations of single-dimensional methods are effectively addressed, and the accuracy of irrigated land identification is significantly improved. 2. By adopting a deep learning model to adaptively fuse multi-source heterogeneous features, the problem of insufficient feature extraction capability of traditional methods is overcome; 3. By introducing a spatial neighborhood smoothing regularization term, the spatial continuity of the recognition results is ensured, which is more in line with the needs of practical applications; 4. The method of the present invention has strong versatility and scalability, and can be applied to the identification of irrigated land in different geographical regions and crop types. Attached Figure Description

[0019] Figure 1 This is an overall flowchart of the method of the present invention; Figure 2 This is a detailed flowchart of the spatial analysis steps in this invention; Figure 3 This is a schematic diagram of the structure of the deep learning model in this invention; Figure 4 This is a block diagram of the system of the present invention. Detailed Implementation

[0020] The present invention will now be described in detail with reference to the accompanying drawings and embodiments. It should be understood that the following embodiments are for illustrative purposes only and are not intended to limit the scope of the invention.

[0021] Example 1: Identification of Irrigated Land in an Agricultural Area of ​​the North China Plain 1. Overview of the Study Area The North China Plain is an important grain-producing area in China. It has a temperate monsoon climate and relies on groundwater and some surface water for agricultural irrigation. Irrigated land and rain-fed farmland are interspersed, making it a typical area for verifying the method of this invention.

[0022] 2. Data Preparation and Preprocessing Remote sensing image sequence: Acquired Sentinel-2 image sequence for the 2023 crop growing season (April to September), with a spatial resolution of 10 meters and a temporal resolution of 5 days. Radiometric calibration, atmospheric correction, and cloud detection processing were performed. Surface water source data: Obtain vector data of rivers and irrigation canals at a scale of 1:50,000; Groundwater irrigation facility data: Obtain survey data on electromechanical well locations, including location information and well depth; Cultivated land patch data: Obtain the cultivated land layer from the 2022 National Land Change Survey. When using a semantic segmentation model to obtain cultivated land patch data, commonly used semantic segmentation models such as U-Net and DeepLabV3+ can be selected.

[0023] 3. Spatial Analysis Implementation Surface water protection rate calculation: The Euclidean distance from each raster cell to the nearest river or canal is calculated using ArcGIS software. The formula is: De = sqrt((x_i - x_r)² + (y_i - y_r)²), where (x_i, y_i) are the center coordinates of raster i, and (x_r, y_r) are the coordinates of the nearest water source. The calculation results are then subjected to min-max normalization to obtain a surface water protection rate value in the range of 0-1. Groundwater availability calculation: Kernel density analysis was performed on the locations of electromechanical wells, with a search radius set to 2000 meters. A quadratic kernel function was used, with the formula: KDE=(1 / (h²))×Σ[weight_j×K(distance(i,j) / h)], where h is the bandwidth parameter and K is the kernel function. The calculation results were normalized to the range of 0-1. Generation of spatial distribution map of irrigation conditions: The surface water security rate layer and the groundwater availability layer are weighted and superimposed with a weight of 0.5:0.5 to generate a comprehensive spatial distribution map of irrigation conditions.

[0024] 4. Time Analysis Implementation Determination of water-sensitive phenological period: Based on the phenological calendar of the main crops (winter wheat-summer maize rotation) in the North China Plain, the period from June 1 to August 31 is determined to be the water-sensitive phenological period; Vegetation index calculation: Based on the preprocessed Sentinel-2 image, the NDVI time series of each cultivated patch is calculated. The formula is: NDVI=(NIR-Red) / (NIR+Red), where NIR is the near-infrared band and Red is the red band. Stability index calculation: Calculate the coefficient of variation of NDVI value for each cultivated patch during the water-sensitive phenological period. The formula is: CV=σ / μ, where σ is the standard deviation and μ is the mean.

[0025] 5. Model Building and Training Construct a hybrid model of U-Net and LSTM, where: The U-Net branch input is a spatial distribution map of irrigation conditions. The encoder uses 4 downsampling layers and the decoder uses 4 upsampling layers, with skip connections. The LSTM branch input is the NDVI mutation coefficient sequence, and a two-layer LSTM network is used with a hidden layer dimension of 128. The feature fusion layer concatenates the outputs of the two branches along the channel dimension, and outputs the classification result through a fully connected layer; The loss function adopts a composite form: Loss = α × CrossEntropyLoss + β × SmoothnessLoss, where α = 0.8, β = 0.2, and SmoothnessLoss adopts a spatial smoothness constraint based on four neighborhoods. Training parameters: initial learning rate of 0.001, Adam optimizer used, batch size of 16, and training epochs of 100.

[0026] 6. Identification Results and Verification Verification showed that the method of this invention achieved an accuracy of 92.5% in identifying irrigated land in this area, with a Kappa coefficient of 0.89. Compared with single methods, the accuracy of identification using only spatial features is 85.3%, and the accuracy of identification using only temporal features is 87.1%. The method of this invention is significantly superior to single-dimensional methods.

[0027] Example 2: Identification of Irrigated Land in Uxin Banner, Inner Mongolia 1. Overview of the Study Area Uxin Banner is located in Ordos City, Inner Mongolia Autonomous Region. It is a typical arid and semi-arid region where agricultural irrigation mainly relies on groundwater. It is an ideal area to verify the adaptability of the method of this invention in different geographical environments.

[0028] 2. Data Preparation and Preprocessing Remote sensing image sequence: Sentinel-2 image sequence acquired during the 2023 growing season (April to September); Water source and irrigation facility data: collect vector data of some tributaries in the Yellow River Basin and general survey data of electromechanical well locations throughout the banner; Farmland patch data: extracted from the previous year's land change survey data.

[0029] 3. Key parameter adjustment Water-sensitive phenological period: Based on the phenological calendar of the main local crops (corn, potato), it is determined to be from June 1 to August 31; Spatial analysis weights: Given that groundwater irrigation is dominant in this region, the superposition weights of the surface water security rate layer and the groundwater availability layer are set to 0.4:0.6. Kernel density analysis search radius: Due to the relatively sparse distribution of well points, the search radius is adjusted to 3000 meters.

[0030] 4. Results and Validation Compared with field survey samples, this method achieved a user identification accuracy of 91% and a mapping accuracy of 89% in irrigated land in Uxin Banner, with an F1 score of 0.90. The results clearly show the distribution of irrigated land along the Yellow River irrigation area and in areas with dense electromechanical wells, which matches the actual situation well.

[0031] Example 3: Model Ablation Experiment To verify the contribution of each component in this invention, an ablation experiment was conducted: Using only spatial features (spatial distribution map of irrigation conditions): accuracy 85.3%; Using only time features (NDVI coefficient of variation): accuracy 87.1%; Using spatial and temporal features, but employing simple stitching and fusion: accuracy 89.2%; The complete method of this invention (spatial + temporal + hybrid model): accuracy 92.5%; Experimental results show that the spatiotemporal dual-dimensional feature fusion method and hybrid model structure proposed in this invention have a significant effect on improving recognition accuracy.

[0032] The above embodiments demonstrate that the method of the present invention can maintain high recognition accuracy under different geographical environments and irrigation modes, and has strong practicality and promotion value.

[0033] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for accurate identification of irrigated land based on spatiotemporal dual-dimensional feature fusion, characterized in that, Includes the following steps: S1: Obtain multi-source data for the target area, wherein the multi-source data includes at least the remote sensing image sequence for the target year, spatial distribution data of surface water sources, spatial distribution data of groundwater irrigation facilities, and cultivated land patch data of the target area; S2: Based on the spatial distribution data of surface water sources and the spatial distribution data of groundwater irrigation facilities, perform spatial analysis to generate a spatial distribution map of irrigation conditions that characterizes irrigation potential; S3: Based on the remote sensing image sequence, extract the time series vegetation index of the cultivated land patch data during the water-sensitive phenological period, and calculate its stability index; S4: Construct a deep learning model, using the spatial distribution map of irrigation conditions and the stability index of the time series vegetation index as fusion input features to train the model and obtain an irrigated land identification model; S5: Using the trained irrigation land recognition model, identify the target area and output an irrigation land distribution map.

2. The method according to claim 1, characterized in that, In step S1, the method for obtaining the cultivated land patch data is as follows: Extracted from existing land use databases or maps; or obtained by identifying cultivated land in the target area using a semantic segmentation model based on the remote sensing image sequence.

3. The method according to claim 1, characterized in that, The step S2 of generating the spatial distribution map of irrigation conditions includes: S21: Based on the spatial distribution data of surface water sources, calculate the Euclidean distance from each grid cell to the nearest surface water source and generate a surface water protection rate layer; S22: Based on the spatial distribution data of the groundwater irrigation facilities, perform kernel density analysis to generate a groundwater availability layer; S23: The surface water security rate layer and the groundwater availability layer are weighted and superimposed to generate a comprehensive spatial distribution map of irrigation conditions.

4. The method according to claim 1, characterized in that, In step S3: The water-sensitive phenological period includes June to August; and / or, the vegetation index is NDVI; and / or, the stability index is the coefficient of variation or standard deviation.

5. The method according to claim 1, characterized in that, In step S4, the deep learning model is a hybrid model that integrates a spatial semantic segmentation network and a time series analysis network. The spatial semantic segmentation network is used to extract the spatial features of the spatial distribution map of irrigation conditions; The time series analysis network is used to extract the temporal characteristics of the stability index of the time series vegetation index.

6. The method according to claim 5, characterized in that, The spatial semantic segmentation network adopts the U-Net network structure, and the time series analysis network adopts the LSTM or Transformer structure.

7. The method according to claim 1, characterized in that, In step S4, when training the deep learning model, the loss function used is a composite loss function, which includes classification cross-entropy loss and a spatial neighborhood smoothing regularization term used to constrain the spatial continuity of the recognition results.

8. A precise identification system for irrigated land based on spatiotemporal dual-dimensional feature fusion, characterized in that, include: The data acquisition module is used to perform step S1 of claim 1; The spatial analysis module is used to perform step S2 of claim 1; The time analysis module is used to execute step S3 of claim 1; The model building and recognition module is used to perform steps S4 and S5 of claim 1.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1 to 7.

Images