Corn and rice lodging area identification method

Abstract

The invention provides a corn and rice lodging area identification method, which comprises the following steps: preprocessing an acquired L2A-level remote sensing image of a target area to obtain target Sentinel-2 data containing a first characteristic variable; based on target Sentinel-2 data containing a first characteristic variable, utilizing a crop identification model to identify a crop type of the target area; for the identified crop type of the target area, obtaining a second characteristic variable which can influence lodging degree identification of the crop type before and after lodging; and carrying out lodging region extraction and lodging degree division on the target region by utilizing an LVQ model based on the second characteristic variables of the crop type before and after lodging. According to the method, the crop lodging area with poor illumination conditions and serious cloud interference can be identified, and meanwhile, the problem of serious dependence on real sample data in the prior art can be solved.

Description

Technical Field

[0001] The present application belongs to the technical field of lodging area detection, and in particular to a method for identifying lodging areas of corn and rice. Background Art

[0002] Lodging is one of the main causes of crop yield loss. The frequent occurrence of extreme weather events in recent years has had a significant impact on agricultural production. When crops such as corn and rice lodge, yield reductions can reach over 25%, or even total failure. Researching methods to efficiently identify areas of crop lodging after meteorological disasters will not only facilitate targeted disaster prevention and mitigation, but also provide important insights for yield assessment and appropriate post-disaster compensation.

[0003] Traditional methods for determining the area and extent of crop lodging rely primarily on manual field surveys, which are inefficient and costly, making them unsuitable for large-scale, precise monitoring of farmland. Remote sensing technology can efficiently and in real time monitor large areas of crop lodging. However, research based on optical remote sensing satellite data requires high-quality, large-scale, cloud-free imagery, which places high demands on weather conditions in the study area. Furthermore, field data collection of real crop samples with varying degrees of lodging is required, increasing monitoring costs.

[0004] In recent years, with the rapid development of drone technology, research on crop lodging monitoring using low-altitude drone remote sensing, which is unaffected by cloud cover, has emerged. Feature fusion based on drone remote sensing data can effectively improve the accuracy of extracting the extent of wheat lodging. While low-altitude drone remote sensing technology is unaffected by cloud cover, its monitoring range is much smaller than that of satellite remote sensing, and most studies are limited to the field scale.

[0005] In addition to optical remote sensing data, SAR technology can penetrate clouds to detect surface objects, making it ideal for monitoring crop lodging in areas with poor lighting conditions and severe cloud cover. However, SAR image data is often noisy and complex to process. Consequently, research using SAR imagery to identify crop lodging areas is relatively limited. Furthermore, the vast majority of current crop lodging detection research focuses on a single crop. Summary of the Invention

[0006] In view of this, the purpose of this application is to provide a method for identifying corn and rice lodging areas, which can identify crop lodging areas with poor lighting conditions and severe cloud interference, and at the same time solve the problem of heavy reliance on real sample data in the existing technology.

[0007] The present application provides a method for identifying lodging areas of corn and rice, the method comprising:

[0008] Preprocessing the collected L2A-level remote sensing image of the target area to obtain target Sentinel-2 data including a first feature variable; wherein the first feature variable is used to characterize the texture features and spatial features of the target Sentinel-2 data;

[0009] Identifying the crop type of the target area using a crop identification model based on target Sentinel-2 data including the first feature variable;

[0010] For the identified crop type in the target area, obtaining a second characteristic variable that can affect the identification of the lodging degree of the crop type before and after lodging;

[0011] Based on the second characteristic variable of the crop type before and after lodging, the LVQ model is used to extract the lodging area and identify the lodging degree of the target area; wherein the lodging degree includes: no lodging, mild lodging, moderate lodging and severe lodging.

[0012] Furthermore, the crop recognition model is obtained by the following method:

[0013] Collect historical L2A-level remote sensing images throughout the year and perform atmospheric correction to obtain historical Sentinel-2 data for the entire year;

[0014] For historical Sentinel-2 data, cloud removal, monthly median synthesis and linear interpolation are performed in sequence to obtain historical Sentinel-2 data covering the crop growing period;

[0015] Extract historical Sentinel-2 data covering the crop maturity period from historical Sentinel-2 data covering the crop growth period;

[0016] Obtaining a first characteristic variable based on historical Sentinel-2 data covering the crop maturity period, and adding the first characteristic variable to the historical Sentinel-2 data covering the crop maturity period to obtain historical Sentinel-2 data including the first characteristic variable;

[0017] A random forest classifier is trained using historical Sentinel-2 data containing the first feature variable and label information of each sample point used to characterize the crop type to obtain the crop recognition model.

[0018] Furthermore, the label information of each sample point used to characterize the crop type is obtained by the following method:

[0019] Binarization processing is performed on historical Sentinel-2 data covering the crop maturity period to obtain a binary image of the time series;

[0020] Extracting the pixel value of each sample point from the binary image of the time series and merging them to generate a merged pixel value of each sample point;

[0021] Based on the combined pixel values ​​of each sample point, the binary image of the time series is divided into non-planted areas and planted areas;

[0022] The sample points in the non-planting area are marked as other, and the sample points in the planting area are marked as rice or corn.

[0023] Furthermore, the sample points in the planting area are marked as rice or corn in the following way:

[0024] For each sample point in the planting area, a buffer zone with a radius of 100m is established with the sample point as the center point;

[0025] For each center point, count the proportion of sample points in the buffer zone that have the same original crop type as the center point;

[0026] If the proportion of sample points with the same crop type as the center point is greater than or equal to 90%, the center point is marked according to the original crop type; otherwise, the center point is removed.

[0027] The original crop type is obtained through a crop distribution dataset.

[0028] Furthermore, the crop types include: corn and rice;

[0029] The step of obtaining, for the identified crop type in the target area, a second characteristic variable that can affect the identification of the lodging degree of the crop type before and after lodging, includes:

[0030] When the identified crop type of the target area is corn, Sentinel-1 data with VH polarization characteristics before and after lodging are used as the second characteristic variable of the crop type;

[0031] When the identified crop type of the target area is rice, Sentinel-1 data with VH, VV+VH and VV / VH polarization characteristics before and after lodging are collectively used as the second characteristic variable of the crop type.

[0032] Furthermore, Sentinel-1 data with VH, VV+VH, and VV / VH polarization characteristics before and after lodging were obtained by the following method:

[0033] Sentinel-1 data were collected for specific time periods before and after the lodging, and topographic and radiometric corrections were performed.

[0034] From the Sentinel-1 data before and after lodging correction, the Sentinel-1 data with VV polarization characteristics and VH polarization characteristics were screened out respectively;

[0035] Based on the Sentinel-1 data with VV polarization characteristics and VH polarization characteristics, Sentinel-1 data with VV+VH and VV / VH polarization characteristics before and after lodging were obtained respectively.

[0036] Furthermore, before identifying the crop type of the target area, the method further includes:

[0037] Based on the target Sentinel-2 data containing the first characteristic variable, the SNIC algorithm is used to obtain the superpixel object collection corresponding to the target Sentinel-2 data.

[0038] Furthermore, before extracting the lodging area and dividing the lodging degree of the target area using the LVQ model, the method further includes:

[0039] Based on the second characteristic variable of the crop type before and after lodging, the SNIC algorithm is used to obtain the superpixel object collection corresponding to the second characteristic variable; among them, the superpixel object collection corresponding to the second characteristic variable has the same superpixel object distribution as the superpixel object collection corresponding to the target Sentinel-2 data.

[0040] The corn and rice lodging area identification method provided in this application can identify crop lodging areas with poor lighting conditions and severe cloud interference, and at the same time can solve the problem of heavy reliance on real sample data in the existing technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 A flow chart of the method for identifying corn and rice lodging areas provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0042] In order to make the purpose, technical solution and advantages of this technical solution more clear, the following technical solution is further described in detail in conjunction with specific implementation methods. It should be understood that these descriptions are only exemplary and are not intended to limit the scope of this technical solution.

[0043] See also Figure 1 , Figure 1 The flowchart of the method for identifying corn and rice lodging areas provided by the embodiment of the present application is shown. Figure 1 As shown, the method includes:

[0044] S101 , preprocessing the collected L2A-level remote sensing image of the target area to obtain target Sentinel-2 data containing a first characteristic variable.

[0045] The first feature variable is used to characterize the texture features and spatial features of the target Sentinel-2 data.

[0046] In this step, Sentinel-2 data is multispectral data. Based on the spectral data of a specific band, a commonly used vegetation index as shown in Table 1 can be obtained. The vegetation index is used as a spectral characteristic variable to generate the first characteristic variable.

[0047] Table 1. Commonly used vegetation indices

[0048]

[0049] Note: ρGREEN, ρRED, ρRED1, ρRED2, ρRED3, ρNIR, ρSWIR1, and ρSWIR2 are the surface reflectance values ​​of the B3, B4, B5, B6, B7, B8, B11, and B12 bands of the Sentinel-2 data, respectively.

[0050] In specific implementation, the first characteristic variable can be generated in the following way:

[0051] Based on the spectral feature variables, the gray level co-occurrence matrix is ​​used to generate the texture feature variables of NDVI and EVI, and the second-order autocorrelation angle matrix (ASM), autocorrelation matrix correlation (CORR) and autocorrelation matrix entropy (ENT) are selected as spatial feature variables; the first feature variable is composed of the texture feature variables and the spatial feature variables.

[0052] Furthermore, the first characteristic variable is added to the target Sentinel-2 data to obtain the target Sentinel-2 data containing the first characteristic variable.

[0053] In specific implementation, the target Sentinel-2 data is obtained through the following methods:

[0054] L2A-level remote sensing images of the target area are collected and atmospheric correction is performed to obtain the target Sentinel-2 data.

[0055] As an example, the target area of ​​the embodiment of the present application is selected in Tieling County, Liaoning Province. First, the basic situation of the county is introduced. The main types of crops in the county are rice and corn. One night, heavy rain and hail occurred in the county, resulting in severe convective weather, which caused the farmland to be damaged and lodged. Moreover, the meteorological conditions in the target area were poor and there were more clouds and fog before and after the crops lodged. Based on this background, the collected L2A-level remote sensing images of the target area are pre-processed to obtain target Sentinel-2 data containing the first characteristic variable. Here, the acquisition time of the L2A-level remote sensing image should be set before the lodging occurs, so that the first characteristic variable will not be affected by the lodging, and can more accurately characterize the texture features and spatial features of the target Sentinel-2 data, so that the subsequent crop type recognition results are more accurate.

[0056] S102: Based on the target Sentinel-2 data including the first characteristic variable, using a crop recognition model, identify the crop type of the target area.

[0057] In specific implementation, the crop recognition model is obtained by the following method:

[0058] Step 201: Collect historical L2A-level remote sensing images for the entire year and perform atmospheric correction to obtain historical Sentinel-2 data for the entire year.

[0059] Step 202: For the historical Sentinel-2 data of the whole year, cloud removal, monthly median synthesis, and linear interpolation are sequentially performed to obtain historical Sentinel-2 data covering the crop growing period.

[0060] Step 203: extract historical Sentinel-2 data covering the crop maturity period from the historical Sentinel-2 data covering the crop growth period.

[0061] Step 204: Obtain a first characteristic variable based on the historical Sentinel-2 data covering the crop maturity period, and add the first characteristic variable to the historical Sentinel-2 data covering the crop maturity period to obtain the historical Sentinel-2 data containing the first characteristic variable.

[0062] Step 205 : Use the historical Sentinel-2 data containing the first feature variable and the label information of each sample point used to characterize the crop type to train a random forest classifier to obtain the crop recognition model.

[0063] In specific implementation, the label information of each sample point used to characterize the crop type is obtained in the following manner:

[0064] Step 301: Binarize historical Sentinel-2 data covering the crop maturity period to obtain a binary image of the time series.

[0065] Step 302: extract the pixel value of each sample point from the binary image of the time series and merge them to generate a merged pixel value of each sample point.

[0066] Step 303: Divide the binary image of the time series into non-planting areas and planting areas based on the combined pixel values ​​of each sample point.

[0067] Step 304: Mark the sample points in the non-planting area as "others" and mark the sample points in the planting area as "rice" or "corn".

[0068] In specific implementation, sample points in the planting area are marked as rice or corn in the following way:

[0069] Step 3041: For each sample point in the planting area, a buffer zone with a radius of 100 m is established with the sample point as the center point.

[0070] Step 3042: For each center point, count the proportion of sample points in the buffer zone that have the same original crop type as the center point.

[0071] When the proportion of sample points with the same crop type as the center point is greater than or equal to 90%, step 3043 is executed to mark the center point according to the original crop type.

[0072] The original crop type is obtained through a crop distribution dataset.

[0073] Otherwise, step 3044: remove the center point.

[0074] In addition, before using the crop recognition model to identify the crop type of the target area, the method further includes:

[0075] Based on the target Sentinel-2 data containing the first characteristic variable, the SNIC algorithm is used to obtain the superpixel object collection corresponding to the target Sentinel-2 data.

[0076] Specifically: Based on the target Sentinel-2 data containing the first eigenvalue, a set of uniformly distributed seed points are generated as initial superpixels; based on the spectral similarity and spatial distance of the target Sentinel-2 data, the initial superpixels with similar color and position characteristics are integrated to achieve the segmentation of the target Sentinel-2 data, thereby obtaining a collection of superpixel objects corresponding to the target Sentinel-2 data.

[0077] As an example, this application sets the seed number to 15, the segmentation scale to 20, the compactness to 0, the connectivity to 8, the neighborhood size to 256, and uses the SNIC-Source full band to segment the target Sentinel-2 data to obtain a superpixel object collection SNIC-Result.

[0078] S103 . For the identified crop type in the target area, obtain a second characteristic variable that can affect the identification of the lodging degree of the crop type before and after lodging.

[0079] Among them, the crop types include: corn and rice.

[0080] In specific implementation, the second characteristic variable that can affect the identification of the lodging degree of the crop type is obtained by the following method:

[0081] When the identified crop type of the target area is corn, step 1031 is executed to use the Sentinel-1 data with VH polarization characteristics before and after lodging as the second characteristic variable of the crop type.

[0082] When the identified crop type of the target area is rice, step 1032 is executed to use the Sentinel-1 data with VH, VV+VH and VV / VH polarization characteristics before and after lodging as the second feature variable of the crop type.

[0083] Here, the reason why different crop types correspond to different second characteristic variables is that after a large number of studies, it was found that the VV polarization feature is more sensitive to changes in moisture content than the VH polarization feature. Considering that heavy rain and hail weather will significantly affect the environmental moisture, and the rice planting environment maintains a high moisture content for a long time during the rainstorm period, the moisture content in the planting area changes relatively little after extreme weather such as heavy rain, and the VV polarization feature that is sensitive to humidity is less affected. Corn is a dryland crop, and the humidity changes in the planting area are large, resulting in greater interference with the VV polarization feature. Therefore, both VV+VH and VV / VH polarization features are not suitable for identifying fallen corn after extreme weather.

[0084] In a specific implementation, Sentinel-1 data with VH, VV+VH, and VV / VH polarization characteristics before and after lodging were obtained by the following method:

[0085] Step 401: Collect Sentinel-1 data in specific time periods before and after lodging, and perform topographic and radiation correction.

[0086] Step 402: Filter out Sentinel-1 data with VV polarization characteristics and VH polarization characteristics from the Sentinel-1 data before and after lodging correction.

[0087] Step 403: Based on the Sentinel-1 data with VV polarization characteristics and VH polarization characteristics, obtain Sentinel-1 data with VV+VH and VV / VH polarization characteristics before and after lodging, respectively.

[0088] S104. Based on the second characteristic variable of the crop type before and after lodging, use the LVQ model to extract the lodging area and divide the lodging degree of the target area.

[0089] The degree of lodging includes: no lodging, mild lodging, moderate lodging and severe lodging.

[0090] In addition, before using the LVQ model to extract the lodging area and classify the lodging degree of the target area, the method further includes:

[0091] Based on the second characteristic variable of the crop type before and after lodging, the SNIC algorithm is used to obtain the superpixel object collection corresponding to the second characteristic variable; among them, the superpixel object collection corresponding to the second characteristic variable has the same superpixel object distribution as the superpixel object collection corresponding to the target Sentinel-2 data.

[0092] Specifically: for the second characteristic variable of the crop type before and after lodging, the superpixel object collection corresponding to the target Sentinel-2 data is used to average all characteristic variable values ​​of each pixel within the corresponding superpixel object to obtain a synthetic image with the same superpixel object distribution before and after lodging; by calculating the difference in characteristic variable values ​​between the same superpixel objects in the synthetic image before and after lodging, a superpixel object collection corresponding to the second characteristic variable with the same superpixel object distribution as the superpixel object collection corresponding to the target Sentinel-2 data is obtained.

[0093] Furthermore, the LVQ model assumes that data samples are labeled with categories, and uses this supervised information to assist clustering. In the superpixel-oriented image clustering process, several objects are randomly selected from the superpixel objects as initial prototype vectors. Each category can have one or more prototype vectors. For each superpixel object, the Euclidean distance between it and all prototype vectors is calculated based on the feature variable value, and the closest prototype vector is found. If the prototype vector's category matches the object, the prototype vector is updated to move closer to the object, otherwise it is moved away, until the prototype vectors converge.

[0094] The above content is only a preferred embodiment of the present invention. For ordinary technicians in this field, many changes can be made in the specific implementation methods and application scope based on the ideas of the present technical content. As long as these changes do not deviate from the concept of the present invention, they all fall within the scope of protection of this patent.

Claims

1. A method for identifying corn and rice lodging areas, characterized in that: The method comprises: Preprocessing the collected L2A-level remote sensing image of the target area to obtain target Sentinel-2 data containing a first feature variable; wherein the first feature variable is used to characterize the texture feature and spatial feature of the target Sentinel-2 data; Based on the target Sentinel-2 data including the first characteristic variable, using a crop recognition model, identifying the crop type of the target area; For the identified crop type in the target area, obtaining a second characteristic variable that can affect the identification of the lodging degree of the crop type before and after lodging; Based on the second characteristic variable of the crop type before and after lodging, the LVQ model is used to extract the lodging area and identify the lodging degree of the target area; wherein the lodging degree includes: no lodging, mild lodging, moderate lodging and severe lodging.

2. The method according to claim 1, characterized in that The crop recognition model is obtained by: Collect historical L2A-level remote sensing images throughout the year and perform atmospheric correction to obtain historical Sentinel-2 data for the whole year; For the historical Sentinel-2 data of the whole year, cloud removal, monthly median synthesis and linear interpolation processing are performed in sequence to obtain the historical Sentinel-2 data covering the crop growth period; Extract historical Sentinel-2 data covering the crop maturity period from historical Sentinel-2 data covering the crop growth period; Obtaining a first characteristic variable based on historical Sentinel-2 data covering the maturity period of the crop, and adding the first characteristic variable to the historical Sentinel-2 data covering the maturity period of the crop, to obtain historical Sentinel-2 data containing the first characteristic variable; A random forest classifier is trained using historical Sentinel-2 data containing the first feature variable and label information of each sample point used to characterize the crop type to obtain the crop recognition model.

3. The method according to claim 2, characterized in that The label information of each sample point used to characterize the crop type is obtained in the following way: Binarization processing is performed on historical Sentinel-2 data covering the crop maturity period to obtain a binary image of the time series; Extracting the pixel value of each sample point from the binary image of the time series and merging them to generate a merged pixel value of each sample point; Based on the combined pixel values ​​of each sample point, the binary image of the time series is divided into non-planted areas and planted areas; The sample points in the non-planting area are marked as others, and the sample points in the planting area are marked as rice or corn.

4. The method according to claim 3, characterized in that Sample points in the planting area are marked as rice or corn in the following way: For each sample point in the planting area, a buffer zone with a radius of 100m was established with the sample point as the center point; For each center point, count the percentage of sample points in the buffer zone that have the same original crop type as the center point; When the proportion of sample points with the same crop type as the center point is greater than or equal to 90%, the center point is marked according to the original crop type, otherwise, the center point is removed; The original crop type is obtained through a crop distribution dataset.

5. The method according to claim 1, characterized in that The crop types include: corn and rice; The step of obtaining, for the identified crop type in the target area, a second characteristic variable that can affect the identification of the lodging degree of the crop type before and after lodging comprises: When the identified crop type of the target area is corn, Sentinel-1 data with VH polarization characteristics before and after lodging are used as the second characteristic variable of the crop type; When the identified crop type of the target area is rice, the Sentinel-1 data with VH, VV+VH and VV / VH polarization characteristics before and after lodging are taken together as the second characteristic variable of the crop type.

6. The method according to claim 5, characterized in that Sentinel-1 data with VH, VV+VH and VV / VH polarization characteristics before and after lodging were obtained by the following method: Sentinel-1 data were collected in specific time periods before and after lodging, and topographic and radiation corrections were performed; From the Sentinel-1 data before and after lodging correction, the Sentinel-1 data with VV polarization characteristics and VH polarization characteristics were screened out respectively; Based on the Sentinel-1 data with VV polarization characteristics and VH polarization characteristics, Sentinel-1 data with VV+VH and VV / VH polarization characteristics before and after lodging were obtained respectively.

7. The method according to claim 1, characterized in that Before using the crop recognition model to identify the crop type of the target area, the method further includes: Based on the target Sentinel-2 data containing the first characteristic variable, the SNIC algorithm is used to obtain a collection of superpixel objects corresponding to the target Sentinel-2 data.

8. The method according to claim 1, characterized in that Before extracting the lodging area and dividing the lodging degree of the target area using the LVQ model, the method further includes: Based on the second characteristic variable of the crop type before and after lodging, the SNIC algorithm is used to obtain a superpixel object collection corresponding to the second characteristic variable; wherein the superpixel object collection corresponding to the second characteristic variable and the superpixel object collection corresponding to the target Sentinel-2 data have the same superpixel object distribution.

Images