Rice lodging remote sensing identification method fusing multiple spectral characteristics

Abstract

The invention relates to a rice lodging remote sensing identification method fusing multiple spectral characteristics. According to the method, a plurality of vegetation indexes and image texture features are creatively adopted to distinguish lodging rice and normal rice, and a lodging rice extraction decision tree model fusing the plurality of vegetation indexes and the texture features is created by constructing a plurality of remote sensing indexes representing the growth state of the rice. According to the method, the lodging rice area is extracted based on the Sentinel-2 image, efficient monitoring of large-range rice lodging information is achieved, and accurate information can be provided for agricultural production management and rice lodging claim settlement. Therefore, the enhancement of the monitoring and identification of the lodging area and position of the rice is of great significance for estimating the lodging loss of the rice and timely taking related rescue and compensation measures.

Description

Technical Field

[0001] The present invention relates to the field of agricultural disasters and remote sensing monitoring technology, and in particular to a rice lodging remote sensing identification method integrating multiple spectral features. Background Art

[0002] As one of the main food crops in my country, rice is often affected by lodging disasters during its growing season. After rice lodging, the orderly structure of the canopy is destroyed and the photosynthesis efficiency is greatly reduced. In addition, the bending and breaking of the stems hinder the transmission path of nutrients inside the crop, affecting the normal growth and development of rice, resulting in a decrease in rice yield and deterioration of quality. Therefore, strengthening the monitoring and identification of the area and location of rice lodging is of great significance for estimating the losses caused by rice lodging and taking relevant rescue and compensation measures in a timely manner.

[0003] Traditional lodging monitoring methods require investigators to conduct field surveys to obtain information such as the location and area of ​​lodging crops, which is inefficient, especially for irregular lodging areas where accurate measurements cannot be made. Remote sensing technology can quickly obtain image information and spatial information of large-scale farmland, and has been widely used in crop lodging monitoring in recent years. In existing crop lodging monitoring studies based on remote sensing data, the image spectral features used are relatively simple and the extraction model is relatively simple, making it difficult to accurately identify rice lodging, and there is a lack of fusion of multiple spectral features to identify rice lodging. Based on Sentinel-2 multispectral image data, this study extracted and analyzed the spectral reflectance characteristics, vegetation index characteristics, and texture characteristics of rice lodging fields. According to the characteristic differences of different fields, the hierarchical decision tree method was used to identify rice lodging, and a rice lodging recognition model was established. In terms of rice lodging monitoring, the fusion of multi-index features has become the development direction of remote sensing applications, providing data support for the spatial distribution and area extraction of rice lodging, disaster assessment, and field management. Summary of the invention

[0004] In view of the shortcomings of the existing rice lodging area extraction, the present invention provides a rice lodging remote sensing identification method that integrates multiple spectral features, deeply mines the spectral features of remote sensing images, calculates and integrates multiple spectral indexes, and quickly and accurately identifies the spatial distribution range and area size of lodging rice.

[0005] In order to achieve the above object, the technical solution of the present invention is as follows:

[0006] A rice lodging remote sensing identification method integrating multiple spectral features comprises the following steps:

[0007] Step 1: Prepare remote sensing image data of the study area: obtain Sentinel-2 multispectral remote sensing images of rice lodging in the study area, and then crop them using vector data after image preprocessing to obtain remote sensing image data covering only the study area;

[0008] Step 2, calculate the spectral index of remote sensing images: calculate the NDVI, LSWI, RVI, DVI spectral index and MEAN_RED texture features on Sentinel-2 images;

[0009] Among them, NDVI stands for Normalized Difference Vegetation Index, which is the ratio of the difference and sum of the near-infrared band and the red band of Sentinel-2 image;

[0010] LSWI stands for Vegetation Water Content Index, which is the ratio of the difference and sum of the near-infrared and short-wave infrared of the Sentinel-2 image;

[0011] RVI stands for Ratio Vegetation Index, which is the ratio calculation between the near infrared and red bands of Sentinel-2 images;

[0012] DVI stands for Difference Vegetation Index, which is the difference between the red and blue bands of the Sentinel-2 image;

[0013] MEAN_RED represents the mean feature of the red light band in the texture feature;

[0014] Step 3, calculating the attribute information of the fallen rice: according to the sample points of the fallen rice, the attribute information of the fallen rice is calculated by using a statistical analysis method;

[0015] Step 4, constructing a rice lodging recognition model: extracting the spatial distribution information of rice lodging based on a hierarchical decision tree method;

[0016] Step 5, post-recognition processing: post-processing the extraction results;

[0017] Step 6: Accuracy verification and model parameter adjustment: Based on the verification sample points, the confusion matrix is ​​used to perform accuracy verification and model parameter adjustment.

[0018] Furthermore, the image used to extract the results in step 1 was the Sentinel-2 image where rice lodging occurred, the vector data was the administrative boundary of the study area, the vector boundary of the cultivated land plot was used in the data post-processing in step 5, and the data used for accuracy verification in step 6 were the field survey points of rice lodging.

[0019] Furthermore, the preprocessing in step 1 includes radiation correction, atmospheric correction, orthorectification, cloud removal and geometric registration processing.

[0020] Furthermore, the calculation formulas for the NDVI, LSWI, RVI, DVI spectral indexes and MEAN_RED texture features in step 2 are:

[0021]

[0022] DVI=ρ Red -ρ Blue

[0023] MEAN_RED=∑ i ∑ j P(i,j)*i

[0024] Among them, ρ Red , Blue , NIR , SWIR are the surface reflectances in the red, blue, near-infrared and short-wave infrared bands respectively; P(i, j) represents the element in the i-th row and j-th column in the grayscale joint matrix.

[0025] Furthermore, the step 4 of constructing a rice lodging recognition model and extracting the spatial distribution information of rice lodging comprises the following steps:

[0026] S1: NDVI can effectively eliminate non-green vegetation and artificial buildings;

[0027] S2: Using LSWI and RVI, which are sensitive to vegetation moisture content, to set thresholds can extract rice lodging areas; combining DVI and MEAN_RED index can further distinguish non-lodging rice from lodging rice.

[0028] Furthermore, the step 4 of constructing a hierarchical decision tree model to extract the spatial distribution of rice lodging area includes the following steps:

[0029] S1: Set the condition NDVI>T0. When the condition is false, the pixel is marked as non-green vegetation. When the condition is true, enter S2 judgment;

[0030] S2: Set the condition LSWI<T1&RVI>T2. If the condition is false, the pixel is marked as other vegetation. If the condition is true, enter S3 for judgment.

[0031] S3: Set the condition DVI>T3&MEAN_RED>T4. If the condition is false, the pixel is marked as rice; if the condition is true, it is lodging rice.

[0032] Among them, T0, T1, T2, T3, and T4 are thresholds, which are obtained by statistical analysis of the lodging rice sample points.

[0033] Furthermore, step 5 performs post-processing on the results extracted by the model, mainly removing speckle noise and non-target objects in the results. The post-processing mainly includes filtering, erosion and expansion, and cultivated land vector masking to obtain raster data with a relatively regular shape.

[0034] Furthermore, the accuracy verification and model parameter adjustment described in step 6 include comparing the remote sensing extraction results with the field survey sample points of fallen rice in the study area, calculating the confusion matrix, and calculating the classification accuracy. If the overall classification accuracy reaches 85% or above, the requirements are met; if it is lower than 85%, proceed to step 3 to readjust the model parameter threshold, and repeat steps 3 and 4 until the requirements are met.

[0035] Beneficial effects:

[0036] Compared with the prior art: the present invention creatively proposes to use multiple vegetation indexes and image texture features to distinguish between lodging rice and normal rice. By constructing multiple remote sensing indexes that characterize the growth status of rice, a decision tree model for extracting lodging rice that integrates multiple vegetation indexes and texture features is created. The method extracts the area of ​​lodging rice based on Sentinel-2 images, realizes efficient monitoring of large-scale rice lodging information, and can provide accurate information for agricultural production management and rice lodging claims. Therefore, strengthening the monitoring and identification of the area and location of rice lodging is of great significance for estimating the losses caused by rice lodging and taking relevant rescue and compensation measures in a timely manner. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a flow chart of the rice lodging remote sensing extraction technology provided by the present invention;

[0038] Figure 2 This is a Sentinel-2 image of the period of rice lodging in Susong County, Anqing City, Anhui Province;

[0039] Figure 3 It is a flow chart of rice lodging extraction model;

[0040] Figure 4 It is the spatial distribution map of rice lodging area;

[0041] Figure 5 This is a comparison of the features of fallen rice in the field and in the Sentinel-2 image. DETAILED DESCRIPTION

[0042] The following will be combined with the embodiments of the present invention and the accompanying drawings to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0043] The present invention provides a rice lodging remote sensing identification method integrating multiple spectral features, comprising the following steps:

[0044] Step 1, make remote sensing image data of the study area: obtain Sentinel-2 multispectral remote sensing images of rice lodging in the study area, and then crop them with vector data after image preprocessing to obtain remote sensing image data covering only the study area; the image used for extraction results is the Sentinel-2 image of rice lodging, and the vector data is the administrative boundary of the study area. Preprocessing includes radiation correction, atmospheric correction, orthorectification, cloud removal and geometric registration processing.

[0045] Step 2, calculate the spectral index of remote sensing images: calculate the NDVI, LSWI, RVI, DVI spectral index and MEAN_RED texture features on Sentinel-2 images;

[0046] Among them, NDVI stands for Normalized Difference Vegetation Index, which is the ratio of the difference and sum of the near-infrared band and the red band of Sentinel-2 image;

[0047] LSWI stands for Vegetation Water Content Index, which is the ratio of the difference and sum of the near-infrared and short-wave infrared of the Sentinel-2 image;

[0048] RVI stands for Ratio Vegetation Index, which is the ratio calculation between the near infrared and red bands of Sentinel-2 images;

[0049] DVI stands for Difference Vegetation Index, which is the difference between the red and blue bands of the Sentinel-2 image;

[0050] MEAN_RED represents the mean feature of the red light band in the texture feature;

[0051] The calculation formulas for the NDVI, LSWI, RVI, DVI spectral indexes and MEAN_RED texture features are:

[0052]

[0053] DVI=ρ Red -ρ Blue

[0054] MEAN_RED=∑i ∑ j P(i,j)*i

[0055] Among them, ρ Red , Blue , NIR , SWIR are the surface reflectances in the red, blue, near-infrared and short-wave infrared bands respectively; P(i, j) represents the element in the i-th row and j-th column in the grayscale joint matrix.

[0056] Step 3, calculating the attribute information of the fallen rice: according to the sample points of the fallen rice, the attribute information of the fallen rice is calculated by using a statistical analysis method;

[0057] Step 4, constructing a rice lodging recognition model: extracting the spatial distribution information of rice lodging based on a hierarchical decision tree method;

[0058] Constructing a rice lodging recognition model and extracting rice lodging spatial distribution information includes the following steps:

[0059] S1: NDVI can effectively eliminate non-green vegetation and artificial buildings;

[0060] S2: Using LSWI and RVI, which are sensitive to vegetation moisture content, to set thresholds can extract rice lodging areas; combining DVI and MEAN_RED index can further distinguish non-lodging rice from lodging rice.

[0061] The method of constructing a hierarchical decision tree model to extract the spatial distribution of rice lodging area comprises the following steps:

[0062] S1: Set the condition NDVI>T0. When the condition is false, the pixel is marked as non-green vegetation. When the condition is true, enter S2 judgment;

[0063] S2: Set the condition LSWI<T1&RVI>T2. If the condition is false, the pixel is marked as other vegetation. If the condition is true, enter S3 for judgment.

[0064] S3: Set the condition DVI>T3&MEAN_RED>T4. If the condition is false, the pixel is marked as rice; if the condition is true, it is lodging rice.

[0065] Among them, T0, T1, T2, T3, and T4 are thresholds, which are obtained by statistical analysis of the lodging rice sample points.

[0066] Step 5, identification post-processing: post-process the extraction results; the vector boundaries of cultivated land plots are used in data post-processing. The data used for accuracy verification in step 6 are the field survey points of rice lodging.

[0067] Step 5: Post-process the results of model extraction, mainly to remove speckle noise and non-target objects in the results. Post-processing mainly includes filtering, erosion and expansion, and cultivated land vector masking to obtain raster data with relatively regular shapes.

[0068] Step 6: Accuracy verification and model parameter adjustment: Based on the verification sample points, the confusion matrix is ​​used to perform accuracy verification and model parameter adjustment.

[0069] The accuracy verification and model parameter adjustment include comparing the remote sensing extraction results with the field survey sample points of fallen rice in the study area, calculating the confusion matrix, and calculating the classification accuracy. If the overall classification accuracy reaches 85% or above, the requirements are met; if it is lower than 85%, proceed to step 3 to readjust the model parameter threshold, and repeat steps 3 and 4 until the requirements are met.

[0070] Example 1

[0071] The technical flow chart of this embodiment is as follows Figure 1 As shown, Susong County, Anqing City, Anhui Province was selected as the study area, and Sentinel-2 images were used as the data source. The image time was September 18, 2021 and September 26, 2021.

[0072] Step 1: Data acquisition and data preprocessing

[0073] This embodiment mainly selects Susong County, Anqing City, Anhui Province as the study area, and obtains Sentinel-2 remote sensing images of the rice growth period and lodging phenomenon in the study area on September 18, 2021 and September 26, 2021; at the same time, in order to reduce the impact of cloud pollution in remote sensing images, this embodiment screens the remote sensing images based on the condition of cloud content ≤30%.

[0074] Specific operations: Call the COPERNICUS / S2_SR image set on the GEE platform to obtain the 2A-level surface reflectance products in the sentienl-2 remote sensing images of Susong County, Anqing City, Anhui Province on September 18, 2021 and September 26, 2021. The 2A-level surface reflectance products have been radiometrically corrected, atmospherically corrected, and orthorectified, and then the quality control band QA mark is used to remove clouds; Use the administrative boundary vector data and digital terrain model data of the study area to crop the processed Sentinel-2 remote sensing images to obtain a remote sensing image dataset covering only the study area, such as Figure 2 shown.

[0075] Step 2: Calculate vegetation index

[0076] The 12 bands of Sentinel-2 images, including blue, green, red, near infrared, and short-wave infrared, are used to calculate the NDVI, LSWI, RVI, and DVI spectral indices, and the gray-level co-occurrence matrix is ​​used to calculate the statistical attribute MEAN_RED texture feature. The calculation formula is as follows:

[0077]

[0078] DVI=ρ Red -ρ Blue

[0079] MEAN_RED=∑ i ∑ j P(i,j)*i

[0080] Among them, ρ Red , Blue , NIR , SWIR are the surface reflectances in the red, blue, near infrared, and short-wave infrared bands, respectively. P(i, j) represents the element in the i-th row and j-th column of the grayscale joint matrix.

[0081] Step 3: Constructing a rice lodging extraction model

[0082] S1: Set the condition to NDVI>0.4. When the condition is false, mark the pixel as non-green vegetation, and when the condition is true, enter step S2.

[0083] S2: Set the condition LSWI<0.4&RVI>2.2. When the condition is false, mark the pixel as other vegetation, and when the condition is true, go to step S3.

[0084] S3: Set the condition DVI>0.004&MEAN_RED>0.045. When the condition is false, the pixel is marked as rice, and when the condition is true, the pixel is marked as lodging rice.

[0085] The specific extraction model process is as follows Figure 3 As shown in Figure 2, the spatial distribution of the extraction results of lodging rice is as follows: Figure 4 shown.

[0086] Step 4: Post-processing of results and accuracy verification

[0087] The model extraction results are affected by other types of land objects and have more noise, so the model results need to be post-processed, including small patch removal and hole fusion, using filtering and erosion and expansion and other image operations. The cultivated land vector data is used to eliminate the influence of non-cultivated land.

[0088] By adjusting the parameters of the decision tree in the model, calculating the confusion matrix and Kappa coefficient, the overall accuracy of rice lodging extraction was obtained through precision verification, and the optimal precision model parameters were maintained. Among them, the optimal overall accuracy was 94.75%, the Kappa coefficient was 0.89, and the overall accuracy was relatively high. Figure 5 It shows the field inspection and image comparison of rice lodging in local areas.

[0089] At the same time, the remote sensing monitoring results were verified based on field surveys and statistical data on reported rice lodging. It was learned that the Huayanghe Farm reported an area of ​​about 16,000 acres of rice lodging in 2021, and the area of ​​rice lodging identified by remote sensing was about 14,865 acres, with an identification accuracy of 92.91%.

[0090] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions only describe the principles of the present invention. The present invention may be subject to various changes and improvements without departing from the spirit and scope of the present invention. These changes and improvements fall within the scope of the present invention. The scope of protection claimed by the present invention is defined by the attached claims and their equivalents.

Claims

1. A rice lodging remote sensing identification method integrating multiple spectral features, characterized in that: The following steps are involved: Step 1: Prepare remote sensing image data of the study area: obtain Sentinel-2 multispectral remote sensing images of rice lodging in the study area, and then crop them using vector data after image preprocessing to obtain remote sensing image data covering only the study area; Step 2, calculate the spectral index of remote sensing images: calculate the NDVI, LSWI, RVI, DVI spectral index and MEAN_RED texture features on Sentinel-2 images; Among them, NDVI stands for Normalized Difference Vegetation Index, which is the ratio of the difference and sum of the near-infrared band and the red band of Sentinel-2 image; LSWI stands for Vegetation Water Content Index, which is the ratio of the difference and sum of the near-infrared and short-wave infrared of the Sentinel-2 image; RVI stands for Ratio Vegetation Index, which is the ratio calculation between the near infrared and red bands of Sentinel-2 images; DVI stands for Difference Vegetation Index, which is the difference between the red and blue bands of the Sentinel-2 image; MEAN_RED represents the mean feature of the red light band in the texture feature; Step 3, calculating the attribute information of the fallen rice: according to the sample points of the fallen rice, the attribute information of the fallen rice is calculated by using a statistical analysis method; Step 4, constructing a rice lodging recognition model: extracting the spatial distribution information of rice lodging based on a hierarchical decision tree method; Step 5, post-recognition processing: post-processing the extraction results; Step 6: Accuracy verification and model parameter adjustment: Based on the verification sample points, the confusion matrix is ​​used to perform accuracy verification and model parameter adjustment.

2. According to claim 1, a rice lodging remote sensing identification method integrating multiple spectral features is characterized in that: The image used to extract the results in step 1 is the Sentinel-2 image where rice lodging occurred, the vector data is the administrative boundary of the study area, the vector boundary of the cultivated land plot is used in the data post-processing in step 5, and the data used for accuracy verification in step 6 is the field investigation points of rice lodging.

3. The rice lodging remote sensing identification method integrating multiple spectral features according to claim 1, characterized in that: Step 1 preprocessing includes radiation correction, atmospheric correction, orthorectification, cloud removal and geometric registration.

4. The rice lodging remote sensing identification method integrating multiple spectral features according to claim 1, characterized in that: The calculation formulas for the NDVI, LSWI, RVI, DVI spectral indexes and MEAN_RED texture features in step 2 are: DVI=ρ Red -r Blue MEAN_RED=∑ i ∑ j P(i,j)*i Among them, ρ Red , Blue , NIR , SWIR are the surface reflectances in the red, blue, near-infrared and short-wave infrared bands respectively; P(i, j) represents the element in the i-th row and j-th column in the grayscale joint matrix.

5. The rice lodging remote sensing identification method integrating multiple spectral features according to claim 1, characterized in that: Step 4 constructs a rice lodging recognition model to extract the spatial distribution information of rice lodging. The following steps are involved: S1: NDVI can effectively eliminate non-green vegetation and artificial buildings; S2: Using LSWI and RVI, which are sensitive to vegetation moisture content, to set thresholds can extract rice lodging areas; combining DVI and MEAN_RED index can further distinguish non-lodging rice from lodging rice.

6. The rice lodging remote sensing identification method integrating multiple spectral features according to claim 1, characterized in that: The step 4 of constructing a hierarchical decision tree model to extract the spatial distribution of rice lodging area includes the following steps: S1: Set the condition NDVI>T0. When the condition is false, the pixel is marked as non-green vegetation. When the condition is true, enter S2 judgment; S2: Set the condition LSWI<T1&RVI>T2. If the condition is false, the pixel is marked as other vegetation. If the condition is true, enter S3 for judgment. S3: Set the condition DVI>T3&MEAN_RED>T4. If the condition is false, the pixel is marked as rice; if the condition is true, it is lodging rice. Among them, T0, T1, T2, T3, and T4 are thresholds, which are obtained by statistical analysis of the lodging rice sample points.

7. The rice lodging remote sensing identification method integrating multiple spectral features according to claim 1, characterized in that: Step 5 post-processes the results extracted by the model, mainly removing speckle noise and non-target objects in the results. The post-processing mainly includes filtering, erosion and expansion, and cultivated land vector masking to obtain raster data with relatively regular shapes.

8. The rice lodging remote sensing identification method integrating multiple spectral features according to claim 1, characterized in that: The accuracy verification and model parameter adjustment described in step 6 include comparing the remote sensing extraction results with the field survey sample points of rice lodging in the study area, calculating the confusion matrix, and calculating the classification accuracy. If the overall classification accuracy reaches 85% or above, the requirements are met; if it is lower than 85%, proceed to step 3 to readjust the model parameter threshold, and repeat steps 3 and 4 until the requirements are met.

Images