A method for acquiring water level of rice field in multiple scales

Abstract

The application discloses a kind of rice field water level multiscale acquisition methods, comprising: collecting the image data of rice field, telemetry data, weather data, soil entropy data and radar water level data, based on image data to the relationship extraction of rice field obtains rice field atlas;Using rice field atlas through crop coverage data and rice field area is Gaussian initialization and optimization, obtains Gaussian map;According to Gaussian map and telemetry water level obtains first water level area and evaporation, through weather data and crop coverage data determine crop growth state, according to crop growth state and soil entropy data determine crop live water, using crop live water and evaporation through bayesian network to radar water level data is corrected, obtains the actual water level of rice field.The method is through collaborative plot level and regional level water level monitoring, combined with probability reasoning, improve the rice field water level monitoring precision and anti-interference ability under complex scene.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of rice field water level monitoring, and particularly relates to a rice field water level multi-scale acquisition method. BACKGROUND

[0002] Rice field water level monitoring is a core link of agricultural production and ecological management, and is extremely important for regulating, improving rice field water condition and regional water conservancy condition, improving the ability to resist natural disasters, promoting the benign circulation of the ecological environment, and being conducive to crop production.

[0003] In order to overcome the deficiencies of the traditional method and meet the demand of modern rice field multi-scale water level information acquisition, the rice field water level multi-scale acquisition method is used to cooperate with the monitoring of the water level of the plot and the region, to meet the accurate irrigation of the single plot and support the optimization of the regional water resources, to integrate the image, telemetry, weather, soil moisture, irrigation and other data, to solve the data islandization, and to improve the monitoring accuracy and anti-interference ability in the complex scene by combining the Bayesian probability reasoning. SUMMARY

[0004] The application aims to provide a rice field water level multi-scale acquisition method.

[0005] To achieve the above-mentioned purpose, the application is implemented according to the following technical scheme:

[0006] The application provides a rice field water level multi-scale acquisition method, which comprises the following steps:

[0007] Collecting rice field data and radar water level data for preprocessing, wherein the rice field data comprises image data, telemetry data, weather data and soil moisture data;

[0008] Dividing the rice field area based on the image data, extracting the rice field area, telemetry water level and crop coverage data according to the rice field area, and obtaining a rice field graph by relationship extraction of the rice field;

[0009] Using the rice field graph to perform Gaussian initialization and optimization through the crop height, crop density, crop rice field coverage and rice field area of the crop coverage data, and obtaining a Gaussian map;

[0010] According to the Gaussian map and the remote sensing water level of the corresponding paddy field, a first water level area is obtained, and based on the first water level area, an evaporation amount is obtained through weather data and crop coverage data;

[0011] The first water level area is corrected using the evaporation amount and radar water level data to obtain a second water level area, and based on the second water level area, a crop growth state is determined through weather data and crop coverage data;

[0012] According to the crop growth state and soil entropy data, a crop live water amount is determined, and the radar water level data is corrected through a Bayesian network using the crop live water amount and the evaporation amount to obtain the actual water level of the paddy field.

[0013] As a further method, the preprocessing method comprises:

[0014] The multispectral image and infrared spectrum data of the paddy field are collected by a drone, the multispectral image is subjected to grayscale conversion and median filtering to obtain image data; remote sensing information is collected, time series interpolation is used to complete missing data of the remote sensing information, abnormal data is removed, and data formats are unified to obtain remote sensing data; temperature and humidity, rainfall, wind speed, and air pressure data are collected and time stamp alignment is performed to obtain weather data; radar water level data is collected, wavelet denoising algorithm is used to remove high-frequency interference noise in the radar water level data, and single-point soil volume moisture content and irrigation data are collected to obtain soil entropy data.

[0015] As a further method, the method for obtaining the paddy field map comprises:

[0016] Based on the image data, superpixel segmentation is performed, the paddy field image is divided into small block areas with similar colors and textures, the small block areas are used as initial labels, gradient calculation is performed on the paddy field image to obtain a gradient image;

[0017] According to the gradient image and the initial labels, a watershed algorithm is used to merge low-gradient-value internal areas of the paddy field and separate boundary areas with high-gradient values, contour extraction is performed on the separated areas and converted into a polygon vector data format to obtain polygon paddy field vector areas;

[0018] Based on the polygon paddy field vector areas, a vegetation index is extracted through reflectivity of the infrared band, the crop coverage of the paddy field is calculated according to the vegetation index, the paddy field area is obtained using the crop coverage combined with geographic coordinates, the crop density is obtained according to the crop coverage and the paddy field area, the crop height and the remote sensing water level are extracted from the remote sensing data of the paddy field, and the crop height, the crop density, and the crop coverage are used as crop coverage data;

[0019] The paddy field plot is taken as a node, the paddy field area, remote sensing water level, crop coverage data and geographical coordinates are taken as node features, the spatial adjacency relationship of adjacent paddy fields and the connection relationship of the irrigation and drainage network are taken as edges, and a graph database is modeled according to the node, node feature and edge, so that a paddy field graph containing spatial topological relationship and attribute association is obtained.

[0020] As a further method, the Gaussian initialization and optimization method comprises:

[0021] Based on the node and spatial adjacency relationship of the paddy field graph, the paddy field area is extracted through GIS topological analysis, the crop height, crop density and crop coverage of the paddy field plot are extracted according to the paddy field area, and the mean and variance are calculated, and the crop height, crop density and crop coverage are standardized by using the mean and variance, so that a three-dimensional feature vector is obtained;

[0022] According to the variance of each feature in the three-dimensional feature vector, a three-dimensional covariance matrix is constructed, the crop height and crop density in the three-dimensional feature vector are respectively mapped to the longitudinal axis and transverse axis in a two-dimensional space, and a two-dimensional covariance matrix is obtained according to 0.2 times the crop density of the longitudinal coordinate and 0.1 times the crop height of the transverse coordinate in the two-dimensional space;

[0023] The area proportion of the area of the paddy field plot to the total area of the paddy field area is taken as the plot weight, the two-dimensional mean, the two-dimensional covariance matrix and the plot weight are taken as the two-dimensional Gaussian component, and the optimized two-dimensional Gaussian component is obtained by using the two-dimensional Gaussian component through the Gaussian parameter optimization formula, and the Gaussian parameter optimization formula is:

[0024]

[0025] wherein is the optimized two-dimensional Gaussian component, including the mean , the covariance and the weight , is the number of observation data points, obtained by the number of pixels after rasterization of the paddy field area, is the number of paddy field plots, is the plot index, is the plot index, is the weight of the th plot, is the two-dimensional feature mean of the plot, is the covariance matrix, is the adjacent plot obtained by the paddy field graph.

[0026] As a further method, the method for obtaining the Gaussian map comprises:

[0027] The Gaussian map parameters are updated based on the optimized two-dimensional Gaussian distribution using the Gaussian map parameter optimization formula.

[0028] Paddy field plots are sorted in descending order of area based on the paddy field region. Crop cover is used as the opacity of each plot. The optimized Gaussian map parameters and opacity are then used to generate a Gaussian map using a transparency blending formula:

[0029]

[0030] in For the first Pixels within a rice paddy area Gaussian probability value, For indexing rice paddy areas, Pixel coordinates in two-dimensional space For the first The opacity of individual paddy field plots For the first The weight of each paddy field plot, For the first The optimized two-dimensional Gaussian distribution of each paddy field plot. The number of paddy field plots. Indicates prior to The rice paddies that have been treated.

[0031] As a further method, the method for optimizing the parameters of the Gaussian map includes:

[0032] The formula for optimizing Gaussian map parameters is:

[0033]

[0034] in, The optimized Gaussian map parameters within the current field of view are obtained by minimizing... Photometric error, iterative adjustment , to generate the image Better distinguish observation images The Adam optimizer is used to efficiently solve this problem and update the Gaussian parameters. For field of view index, To predict the image, The optimized two-dimensional Gaussian components within the current field of view. This is a weighted, transparent combination.

[0035] As a further method, the method for obtaining the evaporation amount includes:

[0036] The part with a Gaussian probability value greater than 0.6 in the image is taken as a crop feature plot based on a Gaussian map, crop coverage data is extracted according to a rice field map, and a water level is remotely sensed, a region belonging to the crop feature plot and having a water level lower than an average water level by 15% is taken as a first water level region, a plot-level evaporation amount is obtained based on the first water level region according to weather data, crop coverage data and the remotely sensed water level of the rice field plot by using a Penman-Monteith evaporation model, and a regional-level evaporation amount is obtained by using a plot weight to aggregate the plot-level evaporation amount.

[0037] As a further method, a method for determining the crop growth state comprises:

[0038] Radar water level data and plot-level evaporation amounts of each point are extracted based on the first water level region, point positions with a plot-level evaporation amount greater than 2 mm within 6 hours are marked according to the radar water level data, and a second water level region is obtained by aggregating rice field plots of the marked point positions by using a plot weight of a Gaussian map.

[0039] A water stress index of a rice field plot is determined by using a water stress index model based on radar water level data, weather data and crop coverage data of the second water level region, a plot with a water stress index greater than 70% is divided into a healthy state, a plot with a water stress index less than 70% and greater than 40% is divided into a mild stress state, and a plot with a water stress index less than 40% is divided into a severe stress state.

[0040] As a further method, a method for correcting radar water level data by using a Bayesian network based on the crop live water amount and the evaporation amount comprises:

[0041] Soil entropy data is extracted based on the second water level region, and a soil volume water content change amount of a single point in the soil entropy data is obtained according to point position coordinates of the soil entropy data.

[0042] A water stress index of a crop growth state, a product of the water stress index and a corresponding plot-level evaporation amount is taken as an actual evaporation amount, soil entropy data in the same time period is extracted according to the actual evaporation amount, a sum of a single-point soil volume water content change amount in the soil entropy data and a change amount of irrigation data is added to rainfall data and subtracted from the actual evaporation amount, and a crop live water amount is obtained.

[0043] A corresponding rice field plot is taken as a plot particle based on the crop live water amount, radar water level data is obtained as time series data according to the plot particle, time series data of a current time of the plot particle is taken as an observation node, the crop live water amount, the actual evaporation amount, the irrigation data, the rainfall data and time series data of a previous time are taken as state nodes according to the observation node, a Bayesian network is obtained according to the observation node and the state node, and a rice field actual water level is obtained by using the Bayesian network through a rice field water level correction formula, the rice field water level correction formula being:

[0044]

[0045] in The corrected radar water level data. The total number of particles in the landmass. For a moment Radar water level data, for Time of the first Water level prediction value for each particle. The standard deviation of radar measurement error For the first The irrigation amount for each particle, with 0 indicating no irrigation, is obtained from rice paddy maps. For the first Evaporation rate per particle For the first The actual crop water volume for each particle. For the first The rainfall amount for each particle, with 0 indicating no rainfall, is obtained from weather data.

[0046] A second aspect of the present invention provides a multi-scale system for obtaining paddy field water levels, comprising:

[0047] Data acquisition module: used to collect paddy field data and radar water level data for preprocessing. The paddy field data includes image data, telemetry data, weather data, and soil entropy data.

[0048] Paddy field mapping module: used to divide paddy field areas based on the image data, extract paddy field area, remote sensing water level and crop cover data according to the paddy field areas, and extract relationships from the paddy fields to obtain paddy field maps;

[0049] Gaussian map module: used to perform Gaussian initialization and optimization using the rice paddy map data based on crop height, crop density, crop rice paddy coverage, and rice paddy area to obtain a Gaussian map;

[0050] First water level area module: used to obtain the first water level area based on the Gaussian map and the telemetry water level of the corresponding paddy field, and to obtain the evaporation rate based on the first water level area through weather data and crop cover data;

[0051] Second water level region module: used to correct the first water level region using the evaporation and radar water level data to obtain a second water level region, and to determine the crop growth status based on the second water level region using weather data and crop cover data;

[0052] Actual water level acquisition module: It is used to determine the actual water volume of crops based on crop growth status and soil entropy data. The actual water volume of crops and evaporation are used to correct the radar water level data through a Bayesian network to obtain the actual water level of the paddy field.

[0053] Compared with the prior art, the embodiments of the present application have at least the following advantages or beneficial effects:

[0054] The present application solves the scale fragmentation problem by cooperatively monitoring the water level at the block level and the regional level, realizes multi-scale water level acquisition, deepens multi-source data fusion, integrates data such as crop coverage of images, water level time series of remote sensing, evaporation and precipitation dynamics of weather, water storage of soil moisture, and human intervention of irrigation, solves the problems of data islandization and missing physical mechanism of model, combines physical water balance and Bayesian network probability reasoning, and improves the monitoring accuracy and anti-interference ability of radar water level in complex scenarios. BRIEF DESCRIPTION OF DRAWINGS

[0055] Figure 1 A flowchart of the steps of the rice field water level multi-scale acquisition method in the embodiments of the present application. DETAILED DESCRIPTION

[0056] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0057] Referring to Figure 1 The present application provides a rice field water level multi-scale acquisition method, which comprises:

[0058] Collect and preprocess rice field data and radar water level data, wherein the rice field data includes image data, remote sensing data, weather data, and soil moisture data.

[0059] In the actual evaluation, the unmanned aerial vehicle is used to fly at an altitude of 100 meters on a flight line, to collect blue, green and red multispectral images with a resolution of 0.15 meters at the rice tillering stage in May and the booting stage in July, to synchronously collect infrared 650 nm and near-infrared 840 nm channel data, to collect L1 LiDAR data with a point cloud density of 200 points per m2, to perform gray scale conversion on the multispectral images through ENVI software, to perform median filtering through Matlab with a 3*3 kernel, to generate 512*512 pixel image data, and to cover 1000 mu of paddy fields; the LiDAR data is supplemented by a time series interpolation method, abnormal data is removed, and data formats are unified to generate a ground surface model DSM and a terrain model DEM, to obtain remote sensing data; the temperature, humidity, rainfall, wind speed and air pressure of the paddy field area are collected every 5 minutes, and the data are aligned with the remote sensing data through a time stamp; Sentinel-1 C band radar water level data with a resolution of 10 meters are collected, wavelet noise reduction is performed, and radar water level grids in the WGS84 standard are generated; Decagon 5TM sensors with a buried depth of 15 cm are used to collect soil volume water content every hour through 100 mu of 1 group; irrigation amount and irrigation points are recorded through an irrigation system, soil moisture data are formed by associating the irrigation points with the soil moisture data.

[0060] The paddy field area is divided based on the image data, the paddy field area is extracted according to the paddy field area, remote sensing water level and crop coverage data, and the paddy field is related to extract a paddy field atlas.

[0061] In the actual evaluation, according to the image data, the SLIC algorithm is used for segmentation to obtain superpixel blocks with similar texture and an average area of 0.25 m2, the Sobel operator with a 3*3 kernel is used to calculate the gradient of the image data, the low gradient region with a gradient less than 10 is merged as the inside of the rice field by using the watershed algorithm, and the high gradient boundary with a gradient greater than 30 is separated as the dike, and the polygon vector containing the area and geographical coordinates is extracted from the gradient region by QGIS; the vegetation index NDVI is calculated by the near-infrared data NIR and the infrared data R of the collected infrared spectrum data, where NDVI greater than 0.4 is effective coverage, and the crop area ratio is 85%; the crop height is obtained by subtracting the DSM data from the DEM data in the telemetry data, and the precision is ± 5 cm, which is calibrated by ground stakes, and the DEM data of the telemetry data and the green spectrum data G and the near-infrared data NIR are used to calculate the NDWI index of water body reflection, where the NDWI index is (G-NIR) / (G+NIR), the telemetry water level is obtained, the rice field area is obtained by using the crop coverage combined with the geographical coordinates, and the crop density is obtained according to the crop coverage and the rice field area; the rice field plot is taken as a node, the rice field area, the telemetry water level, the crop coverage data, and the geographical coordinates are taken as node features, the spatial adjacency relationship of adjacent rice fields and the connection relationship of irrigation and drainage network are taken as edges, and the graph database modeling is performed according to the nodes, node features and edges, and a rice field graph containing 120 plots is formed.

[0062] The rice field graph is used to perform Gaussian initialization and optimization on the crop height, crop density, crop rice field coverage and rice field area of the crop coverage data, and a Gaussian map is obtained.

[0063] In the actual evaluation, based on the nodes and spatial adjacency relationship of the rice field graph, the rice field area is extracted by GIS topological analysis, the crop height, crop density and crop coverage of the plots in the rice field graph are extracted according to the rice field area, and the Z-score standardization is performed, wherein the average of the crop height is 80 cm, the standard deviation is 10 cm, the average of the crop density is 47,000 plants per mu, the standard deviation is 5,000, and the average of the crop coverage is 0.85, the standard deviation is 0.1, and a three-dimensional feature vector is obtained; according to the variance of each feature in the three-dimensional feature vector, a three-dimensional covariance matrix is constructed, the crop height and crop density in the three-dimensional feature vector are respectively mapped to the vertical axis and horizontal axis in the two-dimensional space, and a two-dimensional covariance matrix is obtained according to 0.2 times the crop density of the vertical coordinate and 0.1 times the crop height of the horizontal coordinate in the two-dimensional space;

[0064] The area proportion of the area of the rice field plot to the total area of the rice field region is taken as the plot weight, the two-dimensional mean, the two-dimensional covariance matrix and the plot weight are taken as the two-dimensional Gaussian component, the optimized two-dimensional Gaussian component is obtained by using the Gaussian parameter optimization formula using the Adam optimizer, the learning rate is set to 0.001, and the iteration is 200 times; the Gaussian map parameters are updated based on the optimized two-dimensional Gaussian distribution through the Gaussian map parameter optimization formula, the areas of the rice field regions are sorted in descending order, wherein the largest plot is 10 mu and the smallest is 2 mu, the crop coverage of the rice field plot is 0.85 as the opacity, and the optimized Gaussian map parameters and the opacity are generated into a Gaussian map through the transparency mixing formula in descending order, and a 10-meter resolution Gaussian map is generated, wherein the Gaussian probability value corresponds to a rice dense plot in the crop high probability area of 0.65, and the proportion is 75%.

[0065] A first water level region is obtained according to the Gaussian map and the remote sensing water level corresponding to the rice field, and the evaporation amount is obtained based on the first water level region through weather data and crop coverage data.

[0066] In actual evaluation, the part with a Gaussian probability value greater than 0.6 in the image is taken as a crop feature plot based on the Gaussian map, the crop coverage data and the remote sensing water level are extracted according to the rice field map, and the region belonging to the crop feature plot and having a remote sensing water level lower than the average water level by 15% is taken as the first water level region, wherein the region with a remote sensing water level lower than the average water level of 10.2 cm in the tillering period accounts for 15%, i.e. 150 mu; the plot-level evaporation amount is obtained based on the first water level region through the Penman-Monteith evaporation model according to the weather data, the crop coverage data and the remote sensing water level of the rice field plot, for example, the plot-level evaporation amount of a region in the tillering period is 3.2 mm per day, and that in the booting period is 4.5 mm per day, and the plot-level evaporation amount is aggregated according to the plot weight to obtain the first water level region-level evaporation amount.

[0067] The first water level region is corrected using the evaporation amount and the radar water level data to obtain a second water level region, and the crop growth state is determined based on the second water level region through weather data and crop coverage data.

[0068] In actual evaluation, the radar water level data and the plot-level evaporation amount of each point are extracted based on the first water level region, the point with a plot-level evaporation amount greater than 2 mm within 6 hours is marked according to the radar water level data, for example, the point with a plot-level evaporation amount greater than 2 mm within 6 hours accounts for 20% of the first water level region, i.e. 30 mu, and the second water level region is obtained by aggregating the rice field plots of the marked points through the plot weight of the Gaussian map, i.e. 50 mu.

[0069] The water stress index WSI of the rice field plot is determined by a water stress index model FAO-56 based on the radar water level data, weather data and crop coverage data in the second water level area, and the plot with a water stress index greater than 70% is divided into a healthy state, the plot with a water stress index less than 70% and greater than 40% is divided into a mild stress state, and the plot with a water stress index less than 40% is divided into a severe stress state.

[0070] The crop live water amount is determined according to the crop growth state and the soil entropy data, the radar water level data is corrected by using the crop live water amount and the evaporation amount through the Bayesian network, and the actual water level of the rice field is obtained.

[0071] In the actual evaluation, the soil entropy data is extracted based on the second water level area, the water stress index of the crop growth state is obtained according to the point coordinates of the soil entropy data, the product of the water stress index and the corresponding plot level evaporation amount is used as the actual evaporation amount, the soil entropy data in the same time period is extracted according to the actual evaporation amount, the sum of the change amount of the single-point soil volume water content in the soil entropy data and the change amount of the irrigation data is added to the rainfall data and subtracted from the actual evaporation amount, and the crop live water amount is obtained, for example, the soil volume water content increases by 1.5mm, the irrigation amount increases by 5mm, the rainfall is 2mm, and the actual evaporation amount is 3mm, and the crop live water amount is 5.5mm; the corresponding rice field plot is taken as a plot particle based on the crop live water amount, the radar water level data is taken as time series data according to the plot particle, the 11cm water level data of the time series data of the current time of the plot particle is taken as an observation node, the crop live water amount, the actual evaporation amount, the irrigation data, the rainfall data and the time series data of the last time are taken as state nodes according to the observation node, the Bayesian network is obtained according to the observation node and the state node, and the actual water level of the rice field is obtained by using the Bayesian network through the rice field water level correction formula.

[0072] In this embodiment, the method for preprocessing includes:

[0073] The multispectral image and infrared spectrum data of the rice field are collected by the unmanned aerial vehicle, the multispectral image is subjected to gray scale conversion and median filtering to obtain image data; the telemetry information is collected, the time series interpolation method is used to complete the missing data of the telemetry information, the abnormal data is removed, and the data format is unified to obtain telemetry data; the temperature and humidity, rainfall, wind speed and air pressure data are collected and time stamp alignment is performed to obtain weather data; the radar water level data is collected, the wavelet denoising algorithm is used to remove high-frequency interference noise in the radar water level data, and the single-point soil volume water content and irrigation data are collected to obtain soil entropy data.

[0074] In this embodiment, the method for obtaining the rice field atlas includes:

[0075] Superpixel segmentation is performed based on image data, and the paddy field image is divided into small areas with similar colors and textures, and the small areas are taken as initial labels, and gradient calculation is performed on the paddy field image to obtain a gradient image;

[0076] According to the gradient image and the initial labels, a watershed algorithm is used to merge the internal areas of the paddy field with low gradient values and separate the boundary areas with high gradient values, contour extraction is performed on the separated areas, and the contour extraction is converted into a polygon vector data format to obtain a polygon paddy field vector area;

[0077] Based on the polygon paddy field vector area, a vegetation index is extracted based on the reflectivity of the infrared band, the crop coverage of the paddy field is calculated based on the vegetation index, the crop coverage is combined with the geographic coordinates to obtain the area of the paddy field, and the crop density is obtained based on the crop coverage and the area of the paddy field. The crop height and the remote sensing water level are extracted from the remote sensing data of the paddy field, and the crop height, the crop density and the crop coverage are taken as crop coverage data;

[0078] The paddy field plots are taken as nodes, the area of the paddy field, the remote sensing water level, the crop coverage data and the geographic coordinates are taken as node features, the spatial adjacency relationship of adjacent paddy fields and the connection relationship of the irrigation and drainage network are taken as edges, and a graph database modeling is performed based on the nodes, the node features and the edges to obtain a paddy field graph containing spatial topological relationships and attribute associations.

[0079] In this embodiment, the method of Gaussian initialization and optimization includes:

[0080] Based on the nodes and the spatial adjacency relationship of the paddy field graph, a paddy field area is extracted through GIS topological analysis, the crop height, the crop density and the crop coverage of the paddy field plots are extracted from the paddy field area, and the mean and the variance are calculated, the crop height, the crop density and the crop coverage are standardized by using the mean and the variance, and a three-dimensional feature vector is obtained;

[0081] A three-dimensional covariance matrix is constructed according to the variance of each feature in the three-dimensional feature vector, the crop height and the crop density in the three-dimensional feature vector are respectively mapped to the vertical axis and the horizontal axis in a two-dimensional space, and a two-dimensional covariance matrix is obtained according to the 0.2 times the crop density of the vertical coordinate and the 0.1 times the crop height of the horizontal coordinate in the two-dimensional space;

[0082] The area proportion of the area of the paddy field plot to the total area of the paddy field area is taken as the plot weight, the two-dimensional mean, the two-dimensional covariance matrix and the plot weight are taken as two-dimensional Gaussian components, and the optimized two-dimensional Gaussian components are obtained by using the two-dimensional Gaussian components through a Gaussian parameter optimization formula, and the Gaussian parameter optimization formula is:

[0083]

[0084] wherein For the optimized two-dimensional Gaussian component, including mean , covariance and weight , is the number of observation data points, obtained by the number of pixels after rasterizing the paddy field area, is the number of paddy field plots, is the plot index, is the plot index, is the weight of the th plot, is the mean of the two-dimensional feature of the plot, is the covariance matrix, adjacent plots obtained from the paddy field map.

[0085] In this embodiment, the method for obtaining the Gaussian map comprises:

[0086] updating the Gaussian map parameters based on the optimized two-dimensional Gaussian distribution through a Gaussian map parameter optimization formula;

[0087] According to the descending order of the area of the paddy field plot in the paddy field area, the crop coverage of the paddy field plot is taken as the opacity, and the optimized Gaussian map parameters and the opacity are mixed in descending order through a transparency mixing formula to generate a Gaussian map, and the transparency mixing formula is:

[0088]

[0089] wherein is the Gaussian probability value of the pixel in the th paddy field area, is the paddy field area index, is the pixel coordinate in the two-dimensional space, is the opacity of the th paddy field plot, is the weight of the th paddy field plot, is the optimized two-dimensional Gaussian distribution of the th paddy field plot, is the number of paddy field plots, indicates the paddy field plot processed before .

[0090] In this embodiment, the method for optimizing the parameters of the Gaussian map comprises:

[0091] The Gaussian map parameter optimization formula is:

[0092]

[0093] wherein, The optimized Gaussian map parameters within the current field of view are obtained by minimizing... Photometric error, iterative adjustment , to generate the image Better distinguish observation images The Adam optimizer is used to efficiently solve this problem and update the Gaussian parameters. For field of view index, To predict the image, The optimized two-dimensional Gaussian components within the current field of view. This is a weighted, transparent combination.

[0094] In this embodiment, the method for obtaining the evaporation amount includes:

[0095] Based on the Gaussian map, the parts of the image with a Gaussian probability value greater than 0.6 are regarded as crop feature plots. Crop cover data and telemetry water level are extracted from the paddy field map. The area that belongs to the crop feature plot and whose telemetry water level is 15% lower than the average water level is regarded as the first water level area. Based on the first water level area, the plot-level evaporation is obtained by using the Penman Montes evaporation model according to the weather data, crop cover data and telemetry water level of the paddy field plots. The plot-level evaporation is aggregated using plot weights to obtain the regional evaporation.

[0096] In this embodiment, the method for determining the crop growth status includes:

[0097] Based on the first water level area, radar water level data and plot-level evaporation are extracted from each point. Points with plot-level evaporation greater than 2 mm within 6 hours are marked according to the radar water level data. The paddy fields at the marked points are aggregated using Gaussian map plot weights to obtain the second water level area.

[0098] Based on radar water level data, weather data, and crop cover data of the second water level area, the water stress index of paddy fields was determined by the water stress index model. Fields with a water stress index greater than 70% were classified as healthy, fields with a water stress index less than 70% but greater than 40% were classified as mildly stressed, and fields with a water stress index less than 40% were classified as severely stressed.

[0099] In this embodiment, the method for correcting radar water level data using the actual crop water volume and evaporation rate via a Bayesian network includes:

[0100] Soil entropy data was extracted based on the second water level area, and the location coordinates of the soil entropy data points were obtained.

[0101] The water stress index of crop growth status is used. The product of the water stress index and the corresponding plot-level evaporation is used as the actual evaporation. Soil entropy data within the same time period is extracted based on the actual evaporation. The actual crop water content is obtained by adding the sum of the single-point soil volume water content change and irrigation data change in the soil entropy data, plus the rainfall data minus the actual evaporation.

[0102] Based on the actual crop water volume, the corresponding paddy field plots are treated as plot particles. Radar water level data is obtained from these plot particles as time-series data. The current time-series data of each plot particle is used as an observation node. Based on these observation nodes, the actual crop water volume, actual evaporation, irrigation data, rainfall data, and the previous time-series data are used as state nodes. A Bayesian network is obtained based on the observation nodes and state nodes. The actual paddy field water level is then obtained using the Bayesian network and the paddy field water level correction formula. The paddy field water level correction formula is as follows:

[0103]

[0104] in The corrected radar water level data. The total number of particles in the landmass. For a moment Radar water level data, for Time of the first Predicted water level values ​​for each particle. The standard deviation of radar measurement error For the first The irrigation amount for each particle, with 0 indicating no irrigation, is obtained from rice paddy maps. For the first Evaporation rate per particle For the first The actual crop water volume for each particle. For the first The rainfall amount for each particle, with 0 indicating no rainfall, is obtained from weather data.

[0105] A second aspect of the present invention also provides a multi-scale system for obtaining paddy field water levels, comprising:

[0106] Data acquisition module: used to collect paddy field data and radar water level data for preprocessing. The paddy field data includes image data, telemetry data, weather data, and soil entropy data.

[0107] Paddy Field Mapping Module: Used to divide paddy field areas based on the image data, extract paddy field area, remote sensing water level and crop cover data according to the paddy field areas, and extract relationships from the paddy fields to obtain paddy field maps;

[0108] a Gaussian map module for Gaussian initialization and optimization using the paddy field map by crop height, crop density, crop paddy field coverage and paddy field area of crop coverage data, to obtain a Gaussian map;

[0109] a first water level area module for obtaining a first water level area according to the Gaussian map and remote sensing water level of the corresponding paddy field, and obtaining evaporation amount by weather data and crop coverage data based on the first water level area;

[0110] a second water level area module for correcting the first water level area using the evaporation amount and radar water level data to obtain a second water level area, and determining crop growth state by weather data and crop coverage data based on the second water level area;

[0111] an actual water level obtaining module for determining crop actual water amount according to the crop growth state and soil entropy data, correcting radar water level data by Bayesian network using the crop actual water amount and the evaporation amount, and obtaining paddy field actual water level.

[0112] The above is only an example and description of the structure of the present application. Those skilled in the art can make various modifications or supplements to the described specific embodiments or use similar ways to replace them, as long as they do not deviate from the structure of the present application or exceed the scope defined by the present claims.

Claims

1. A method for multi-scale acquisition of water level in a rice field, characterized in that, The method comprises the following steps: Collect and preprocess paddy field data and radar water level data, wherein the paddy field data comprises image data, remote sensing data, weather data and soil entropy data; Divide the paddy field area based on the image data, extract the paddy field area, remote sensing water level and crop coverage data according to the paddy field area, and perform relation extraction on the paddy field to obtain a paddy field graph; Use the paddy field graph to perform Gaussian initialization and optimization on the crop height, crop density, crop coverage and paddy field area of the crop coverage data to obtain a Gaussian map; Obtain a first water level area according to the Gaussian map and the remote sensing water level of the corresponding paddy field, and obtain the evaporation amount based on the first water level area and the weather data and crop coverage data; Use the evaporation amount and radar water level data to correct the first water level area to obtain a second water level area, and determine the crop growth state based on the second water level area and the weather data and crop coverage data; Determine the crop actual water amount according to the crop growth state and the soil entropy data, correct the radar water level data by using the crop actual water amount and the evaporation amount through a Bayesian network, and obtain the actual water level of the paddy field.

2. The method according to claim 1, wherein, The preprocessing method comprises: Collect multispectral images and infrared spectrum data of the paddy field by using a drone, perform gray scale conversion and median filtering on the multispectral images to obtain image data, collect remote sensing information, complete missing data of the remote sensing information by using a time series interpolation method, remove abnormal data, and unify the data format to obtain remote sensing data, collect temperature and humidity, rainfall, wind speed and air pressure data, and perform time stamp alignment to obtain weather data, collect radar water level data, remove high-frequency interference noise in the radar water level data by using a wavelet denoising algorithm, collect single-point soil volume water content and irrigation data, and obtain soil entropy data.

3. The method according to claim 1, wherein, The method for obtaining the paddy field graph comprises: Perform superpixel segmentation based on the image data, divide the paddy field image into small block areas with similar colors and textures, take the small block areas as initial labels, perform gradient calculation on the paddy field image to obtain a gradient image; Merge the internal areas of the paddy field with low gradient values and separate the boundary areas with high gradient values by using a watershed algorithm according to the gradient image and the initial labels, perform contour extraction on the separated areas and convert them into a polygon vector data format to obtain polygon paddy field vector areas; Extract a vegetation index based on the polygon paddy field vector areas by using the reflectivity of the infrared band, calculate the crop coverage of the paddy field according to the vegetation index, obtain the paddy field area by using the crop coverage in combination with geographic coordinates, obtain the crop density according to the crop coverage and the paddy field area, extract the crop height and the remote sensing water level by using the remote sensing data of the paddy field, and take the crop height, the crop density and the crop coverage as crop coverage data; Take the paddy field plots as nodes, take the paddy field area, the remote sensing water level, the crop coverage data and the geographic coordinates as node features, take the spatial adjacency relationship of adjacent paddy fields and the connection relationship of irrigation and drainage networks as edges, use the node features and the edges to model a graph database, and obtain a paddy field graph comprising spatial topological relationships and attribute associations.

4. The method according to claim 1, wherein, The Gaussian initialization and optimization method comprises: The rice field region is extracted by GIS topological analysis based on the node and spatial adjacency relationship of the rice field map, the crop height, crop density and crop coverage of the rice field block are extracted according to the rice field region, and the mean and variance are calculated, and the crop height, crop density and crop coverage are standardized by using the mean and variance, and a three-dimensional feature vector is obtained; A three-dimensional covariance matrix is constructed according to the variance of each feature in the three-dimensional feature vector, the crop height and crop density in the three-dimensional feature vector are mapped to the longitudinal axis and transverse axis of the two-dimensional space respectively, and a two-dimensional covariance matrix is obtained according to 0.2 times the crop density of the longitudinal coordinate and 0.1 times the crop height of the transverse coordinate in the two-dimensional space; The area proportion of the area of the rice field block to the total area of the rice field region is taken as the block weight, the two-dimensional mean, the two-dimensional covariance matrix and the block weight are taken as the two-dimensional Gaussian component, and the optimized two-dimensional Gaussian component is obtained by using the two-dimensional Gaussian component through the Gaussian parameter optimization formula, and the Gaussian parameter optimization formula is: wherein is the optimized two-dimensional Gaussian component, including mean , covariance and weight , is the number of observation data points, obtained by the number of pixels after rasterization of the paddy field area, is the number of paddy field plots, is the plot index, is the plot index, is the weight of the th plot, is the two-dimensional feature mean of the plot, is the covariance matrix, adjacent plots obtained from the paddy field map.

5. The method according to claim 1, wherein, The method for obtaining the Gaussian map comprises: The Gaussian map parameters are updated based on the optimized two-dimensional Gaussian distribution through the Gaussian map parameter optimization formula; The areas of the rice field blocks are sorted in descending order according to the rice field region, the crop coverage of the rice field block is taken as the opacity, and the optimized Gaussian map parameters and the opacity are generated into the Gaussian map in descending order through the transparency mixing formula, and the transparency mixing formula is: wherein is the Gaussian probability value of the pixel in the th rice field region, is the rice field region index, is the pixel coordinate on the two-dimensional space, is the opacity of the th rice field patch, is the weight of the th rice field patch, is the optimized two-dimensional Gaussian distribution of the th rice field patch, is the number of rice field patches, denotes the rice field patch that precedes the processing.

6. The method according to claim 5, wherein, The method for optimizing the parameters of the Gaussian map comprises: The Gaussian map parameter optimization formula is: wherein, is the optimized Gaussian map parameter in the current field of view, which is iteratively adjusted by minimizing the photometric error, so that the generated image better matches the observed image This problem is efficiently solved using the Adam optimizer to update the Gaussian parameters, is the field of view index, is the predicted image, is the optimized two-dimensional Gaussian component in the current field of view, is the weighted transparent combination.

7. The method according to claim 1, wherein, The method for obtaining the evaporation amount comprises: The part with a Gaussian probability value greater than 0.6 in the image is taken as a crop feature block based on the Gaussian map, the crop coverage data and the remote water level are extracted according to the rice field map, the region belonging to the crop feature block and having a remote water level lower than the average water level by 15% is taken as a first water level region, the block-level evaporation amount is obtained based on the first water level region according to the weather data, the crop coverage data and the remote water level of the rice field block through the Penman-Monteith evaporation model, and the regional-level evaporation amount is obtained by aggregating the block-level evaporation amount using the block weight.

8. The method according to claim 1, wherein, The method for determining the crop growth state comprises: The radar water level data and the block-level evaporation amount of each point are extracted based on the first water level region, the point positions with a block-level evaporation amount greater than 2 mm within 6 hours are marked according to the radar water level data, and the rice field blocks of the marked point positions are aggregated through the block weight of the Gaussian map to obtain a second water level region; The water stress index of the rice field block is determined based on the radar water level data, the weather data and the crop coverage data of the second water level region through a water stress index model, the block with a water stress index greater than 70% is divided into a healthy state, the block with a water stress index less than 70% and greater than 40% is divided into a mild stress state, and the block with a water stress index less than 40% is divided into a severe stress state.

9. The method according to claim 1, wherein, The method for correcting the radar water level data through the Bayesian network using the crop live water amount and the evaporation amount comprises: The soil entropy data is extracted based on the second water level region, the point coordinates of the soil entropy data are obtained, and the soil entropy data is corrected through the Bayesian network. The water stress index of the crop growth state, the product of the water stress index and the corresponding plot evaporation amount is used as the actual evaporation amount, the soil entropy data in the same time period is extracted according to the actual evaporation amount, the sum of the change amount of the single-point soil volume water content in the soil entropy data and the change amount of the irrigation data is added to the rainfall data and subtracted from the actual evaporation amount, and the crop actual water amount is obtained; Based on the crop actual water amount, the corresponding paddy field plot is taken as a plot particle, the radar water level data is obtained as time series data according to the plot particle, the time series data of the current moment of the plot particle is taken as an observation node, the crop actual water amount, the actual evaporation amount, the irrigation data, the rainfall data and the time series data of the last moment are taken as state nodes according to the observation node, the Bayesian network is obtained according to the observation node and the state node, and the actual water level of the paddy field is obtained by using the Bayesian network through the paddy field water level correction formula, and the paddy field water level correction formula is: wherein is the corrected radar water level data, is the total number of plots, is the radar water level data at time is the radar water level data at time is is the water level prediction value of the th particle at time is the standard deviation of the radar measurement error, is the irrigation amount of the th particle, 0 means no irrigation, obtained by paddy field map, is the evaporation amount of the th particle, is the crop live water amount of the th particle, is the rainfall amount of the th particle, 0 means no rainfall, obtained by weather data.

10. A system for acquiring water level in rice field at multiple scales, for performing the method of any one of claims 1-9, wherein The system comprises: A data acquisition module is configured to acquire and preprocess paddy field data and radar water level data, wherein the paddy field data comprises image data, telemetry data, weather data and soil entropy data; A paddy field atlas module is configured to divide a paddy field area based on the image data, extract a paddy field area, a telemetry water level and crop coverage data from the paddy field area, and obtain a paddy field atlas by performing relationship extraction on the paddy field; A Gaussian map module is configured to perform Gaussian initialization and optimization by using the paddy field atlas, the crop height, the crop density, the crop paddy field coverage and the paddy field area of the crop coverage data, and obtain a Gaussian map; A first water level area module is configured to obtain a first water level area based on the Gaussian map and the telemetry water level of the corresponding paddy field, and obtain an evaporation amount based on the first water level area, the weather data and the crop coverage data; A second water level area module is configured to correct the first water level area based on the evaporation amount and the radar water level data to obtain a second water level area, and determine a crop growth state based on the second water level area, the weather data and the crop coverage data; An actual water level acquisition module is configured to determine a crop actual water amount based on the crop growth state and the soil entropy data, correct the radar water level data based on the crop actual water amount and the evaporation amount by using a Bayesian network, and obtain an actual water level of the paddy field.

Images