A remote sensing extraction method and device for lodging corn
By combining the remote sensing monitoring of lodged corn in the Northeast region, combining the remote sensing spectral index and texture characteristics, a decision tree model is constructed, which solves the problem of quickly and accurately obtaining the distribution and area of lodged corn, and realizes efficient lodged corn monitoring, satisfying the post-disaster diagnosis and prevention and control needs.
Patent Information
- Application Number
- CN202011441317.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2020-12-08
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2040-12-08
AI Technical Summary
It is difficult to quickly and accurately obtain the distribution and area of large-scale lodging corn, especially in high-density corn cultivation areas in the Northeast region, which affects the efficiency of post-disaster agricultural production management and agricultural insurance.
Using the method of lodged corn extraction based on remote sensing spectral index (NDVI, RVI, DVIG-B, etc.) and texture features, the multi-spectral image is obtained for pre-processing, vegetation index and texture information are calculated, and the decision tree model is constructed, combining spectral and texture features. The lodged corn information was extracted and post-processed and accuracy verification were performed.
It realizes rapid and accurate monitoring of the distribution of lodged corn, which can distinguish between lodged and non-lodged corn, and without complex multi-spectral bands, it uses only blue, green, red and near-infrared data to extract lodged corn to achieve 90% or above classification accuracy.
Smart Images

Figure CN112766036B_ABST
Abstract
Description
Technical Field
[0001] The present application belongs to the field of agricultural disaster and remote sensing monitoring application technology, and in particular, relates to a remote sensing extraction method and device for lodging corn, and in particular, to a remote sensing extraction method and device for lodging corn for the Northeast region. Background Art
[0002] In summer, there are frequent disastrous weather such as strong winds and rainstorms. Due to its high stalks, corn is prone to lodging during its growth period, especially in large-scale corn planting areas in Northeast China. The lodging area is an important indicator for assessing the extent of crop lodging disasters. Investigating the area of crop lodging is of great significance to post-disaster agricultural production management, agricultural disaster insurance, agricultural subsidies and other work. After corn lodging, its canopy structure changes, which changes the canopy radiation transmission characteristics and spectral characteristics, which makes it possible to remotely sense and monitor large-scale corn lodging.
[0003] In recent years, with the development of remote sensing technology, research on remote sensing monitoring of crop lodging has begun to attract the attention of scholars. Remote sensing can quickly obtain information such as the spectrum and texture of the target object through long-distance detection. It is a fast and effective technical method to investigate the condition of crop lodging. The current research on remote sensing monitoring of corn lodging mainly focuses on the study of changes in the spectral characteristics of lodging corn and the extraction of lodging corn using commonly used supervised and unsupervised classification methods. Texture, as a major feature of remote sensing images, plays an important role in crop classification. In terms of lodging monitoring, combining spectral and texture features has become the development direction of remote sensing applications. In order to quickly and accurately obtain the distribution and area of large-scale lodging corn, an efficient remote sensing extraction method for lodging corn was explored and invented in the Northeast Plain area as the research area. Summary of the invention
[0004] The technical problem to be solved by the present invention is to quickly and accurately obtain the distribution and area of large-scale lodging corn, thereby providing a remote sensing spectral index (NDVI, RVI, DVI) based on the G-B A remote sensing extraction method and device for lodging corn based on data from a plant and plant (such as soil, soil, and soil conditions) and texture features.
[0005] The technical solution adopted by the present invention to solve the technical problem is as follows: first, the present invention provides a remote sensing extraction method for lodging corn, comprising the following steps:
[0006] (1) Obtain remote sensing multispectral images of the study area after lodging. The images used include red, green, blue, and near-infrared bands, such as GF1 images and Sentinel-2 images. Then preprocess the images. Image preprocessing mainly includes: radiation calibration, atmospheric correction, orthorectification, image registration, cropping and other operation steps;
[0007] (2) Calculate NDVI, RVI, and DVI on the image G-B , SVI vegetation index;
[0008] Among them, NDVI is the normalized vegetation index, which is the ratio of the difference between the reflectance of the near-infrared and red bands to the sum of them, and can be used to distinguish vegetation from non-vegetation;
[0009] DVI G-B The custom vegetation index of the present invention is the difference between the reflectance of the blue and green bands, which can reflect the spectral difference between corn and rice that fall down due to typhoons at the waxy maturity stage, and can be used to remove rice from crops;
[0010] RVI is the ratio vegetation index, which is the ratio of the reflectance of the near-infrared band to the red light band, and is used to eliminate other vegetation, which refers to vegetation other than non-lodging corn and rice;
[0011] SVI is the shadow vegetation index, which is the product of NDVI and near-infrared reflectivity. It is used to eliminate the effects of vegetation shadows, field ridges, etc.
[0012] (3) The gray-level co-occurrence matrix (GLCM) based on statistical analysis was used to calculate the texture information of lodging and non-lodging corn, rice, and other crops;
[0013] GLCM is a matrix that statistically describes a certain relationship between the grayscale of adjacent pixels in a local area or the entire area of an image or two pixels within a certain distance. The matrix describes the probability that a pair of pixels separated by d pixels have grayscales i and j respectively in the direction θ. Its elements can be recorded as P(i,j|d,θ). When θ and d are determined, it is abbreviated as P i,j. Its order is determined by the grayscale quantization level of the image. The present invention selects the texture mean (Mean) index to express the texture characteristics of the image through research and analysis. The calculation formula is:
[0014]
[0015] Among them, f(i,j|d,θ) represents the probability that a pair of pixels separated by d pixels have gray levels i and j respectively in the direction θ, where M=N is the gray level, and M×N represents the scale of the gray level co-occurrence matrix; θ is generally 0°, 45°, 90°, or 135°;
[0016] (4) Construct a decision tree model to extract information on corn lodging conditions;
[0017] (5) post-processing the results extracted by the model;
[0018] (6) Accuracy verification and model parameter adjustment.
[0019] Furthermore, in the above method, the image used to extract the results in step (1) is a GF1 WFV image, and the image used for accuracy verification in step (6) is a GF-2 image with a resolution of 0.8 m.
[0020] Further preferably, the calculation formulas of the NDVI, RVI, DVI GB, and SVI vegetation indices in step (2) are:
[0021]
[0022] DVI G-B =ρ Green -ρ Blue ,
[0023]
[0024] Among them, ρ Nir , ρ Red , ρ Green ,ρ Blue are the spectral reflectances of the near-infrared, red, green, and blue bands, respectively.
[0025] Further preferably, the texture information of the image in step (3) is represented by a texture mean based on a gray level co-occurrence matrix, and the calculation formula is:
[0026]
[0027] Among them, f(i,j|d,θ) represents the probability that a pair of pixels separated by a distance of d pixels in the direction θ have gray levels i and j respectively, where M=N is the gray level, and M×N represents the scale of the gray level co-occurrence matrix; θ is generally 0°, 45°, 90°, or 135°.
[0028] Furthermore, step (4) of constructing a decision tree model to extract corn lodging status information includes the following steps:
[0029] S1: Use NDVI to remove water bodies, buildings, and some vegetation other than corn;
[0030] S2: Use B2_Mean to distinguish lodging corn, rice, non-lodging corn, and other vegetation, where B2_Mean is the texture mean of the green band.
[0031] In the above method, further preferably, step (4) of constructing a decision tree model to extract the distribution information of lodging corn comprises the following steps:
[0032] S1: Set the condition T0 < NDVI < T1. When the condition is false, the pixel is marked as other features. When the condition is true, proceed to S2 for judgment. Here, NDVI is calculated from the reflectance of the red and near-infrared bands of the image, and the thresholds T0 and T1 are determined respectively based on the distribution and statistics of the sample data.
[0033] S2: Set the condition B2_Mean > T2. If the condition is false, proceed to S3 for judgment; if the condition is true, proceed to S4 for judgment. B2_Mean is the key to distinguishing between lodged and non-lodged corn, and the threshold T2 is obtained from the distribution statistics of the sample texture data.
[0034] S3: After determining that S2 is false, use NDVI, RVI, and SVI to eliminate other vegetation, vegetation shadows, etc. Specifically, set the conditions NDVI > T3, RVI < T4, SVI > T5. If the conditions are true, the pixel is non-lodged corn; the thresholds T3, T4, and T5 are obtained respectively from the distribution and statistics of the sample data.
[0035] S4: After determining that S2 is true, use DVI G-B to eliminate rice. Specifically, set the condition DVI G-B < T6. When the condition is false, the pixel is marked as rice; when the condition is true, the pixel is marked as lodged corn. The threshold T6 is determined according to the distribution and statistics of the sample data.
[0036] Furthermore, step (5) performs post-processing on the results extracted by the model, mainly aiming at merging the speckle noise and non-target features (other features, rice, and other vegetation) in the results. The post-processing mainly includes filtering, sieving, and clumping processes to eliminate small patches and merge classes, obtaining raster result data with a more regular shape.
[0037] Furthermore, the accuracy verification and model parameter tuning in step (6) include randomly generating sample points in the study area, setting the number of sample points according to the area range and size, and determining the sample class attributes based on visual interpretation of high-resolution and medium-resolution images as verification samples. Compare the remote sensing extraction results with them, calculate the confusion matrix, and obtain the classification accuracy. If the overall classification accuracy reaches 90% or above, it meets the requirements; if it is lower than 90%, proceed to step (4) to re-adjust the model parameter thresholds and repeat steps (4) and (5) until the requirements are met.
[0038] The beneficial effects of the present invention are:
[0039] Mastering the range and severity of corn lodging is the key to disaster diagnosis, timely prevention and control, and loss assessment. The present invention creatively proposes that the green band texture mean (B2_Mean) can be used to distinguish between lodged and non-lodged corn, and by constructing DVIG-B In order to distinguish lodged corn from rice, a decision tree model for lodged corn extraction combining spectrum and texture was created. The method does not need to obtain the scope of the corn planting area in advance, and the model can be directly used to extract lodged corn from the preprocessed image. The data used does not require complex multi-spectral bands, and only image data containing four bands of blue, green, red, and near-infrared are needed to extract lodged corn, thereby realizing rapid and accurate monitoring of the distribution of lodged corn. BRIEF DESCRIPTION OF THE DRAWINGS
[0040] The technical solution of the present application is further described below in conjunction with the accompanying drawings and embodiments.
[0041] Figure 1 This is a flow chart of the remote sensing extraction technology for lodging corn provided by the present invention;
[0042] Figure 2 This is the GF1 image of corn lodging in the Northeast Plains;
[0043] Figure 3 It is the texture mean image of the green light band;
[0044] Figure 4 It is a flow chart of the extraction model of lodging corn and non-lodging corn;
[0045] Figure 5 This is a graph of the extraction results of lodging and non-lodging corn;
[0046] Figure 6 This is the result of extraction of lodging corn;
[0047] Fig. 7A and Figure 7B This is a comparison of the results of extracting fallen corn in a local area. Fig. 7A This is the GF1 image from September 6th. Figure 7B This is a diagram of the extraction of fallen corn. DETAILED DESCRIPTION
[0048] In order to better reflect the advantages and specific processes of the present invention, the present invention is further explained below with reference to examples. It should be noted that the embodiments and features in the embodiments of the present application can be combined with each other without conflict.
[0049] Example
[0050] The technical flow chart of the present invention is as follows: Figure 1 As shown in the figure, part of Ningjiang District, Songyuan City, Jilin Province was selected as the study area, and the GF1 WFV image data was used as the data source. The image time was September 6, 2020.
[0051] Step 1: Data acquisition and data preprocessing
[0052] Download GF1 WFV data with a resolution of 16 meters, perform radiometric calibration, atmospheric correction, orthorectification and registration on the image data, and use the downloaded Pre-Lodging Sentinel 2 data for registration. After atmospheric correction, convert the image DN value into reflectivity data, and resample the image to 10 meters after preprocessing. Crop the image according to the scope of the study area. After cropping, the study area is as follows Figure 2 shown.
[0053] Step 2: Calculate vegetation index
[0054] The NDVI, RVI, DVI GB, and SVI vegetation indices are calculated using the blue, green, red, and near-infrared bands of the GF1 image. The calculation formula is as follows:
[0055]
[0056]
[0057] DVI G-B =ρ Green -ρ Blue ,
[0058]
[0059] Among them, ρ Nir , ρ Red , ρ Green ,ρ Blue They are the spectral reflectance of near infrared, red, green and blue bands respectively, DVI G-B It is a custom index, which is the difference between the reflectivity of green light and blue light bands, and can distinguish between lodged corn and rice.
[0060] Step 3: Calculate the texture mean
[0061] The mean texture is calculated based on the gray-level co-occurrence matrix, the moving window size is set to 3×3, the transformation amount in the X and Y directions of the matrix is 1, and the gray level is 64. According to the research and analysis, the mean texture of the green light band is selected for subsequent model construction. The calculation formula of the texture mean (Mean) based on the gray-level co-occurrence matrix is:
[0062]
[0063] The obtained green light band texture mean image is as follows Figure 3 shown.
[0064] Step 4: Constructing a lodging corn extraction model
[0065] S1: Set the condition to 0.72>NDVI>0.47. When the condition is false, mark the pixel as other features. When the condition is true, enter step S2.
[0066] S2: Set the condition to B2_Mean>8.5. When the condition is false, enter S3 for judgment, and when the condition is true, enter S4 for judgment.
[0067] S3: Set the conditions to NDVI>0.55, RVI<5.4 and SVI>0.12. When the conditions are false, the pixels are marked as other vegetation, and when the conditions are true, the pixels are marked as non-lodging corn;
[0068] S4: Set the condition to D G-B <0.024. When the condition is false, the pixel is marked as rice, and when the condition is true, the pixel is marked as lodging corn.
[0069] The specific extraction process is as follows: Figure 4 As shown, the extraction results of lodging corn and non-lodging corn are as follows Figure 5 As shown, the extraction results of lodging corn are as follows Figure 6 shown.
[0070] Step 5: Post-processing of results and accuracy verification
[0071] The initial extraction results often have speckle noise, and in order to eliminate other objects, the results have many categories. Therefore, after using the above model for extraction, the results need to be removed from small patches and merged into categories, and the results are post-processed using filtering, small patch removal, clustering and other operations. Accuracy verification uses randomly generated samples, combines high-resolution and medium-resolution images for visual interpretation, obtains sample attribute information, and then compares it with the extraction results, calculates the confusion matrix and Kappa coefficient, and obtains the extraction accuracy.
[0072] The final model parameters were obtained by continuously adjusting the parameters, and the model was used to extract lodging and non-lodging corn and other ground objects. The accuracy of the results was verified after post-processing. The calculation showed that the overall accuracy of the results was 90.33%, and the Kappa coefficient was 0.8495. The overall accuracy was high, and 7A and Figure 7B The extraction results and image comparison of lodging corn in local areas are shown.
[0073] Based on the above ideal embodiments of this application, the relevant staff can make various changes and modifications without departing from the technical concept of this application through the above description. The technical scope of this application is not limited to the content in the specification, and its technical scope must be determined according to the scope of the claims.
Claims
1. A remote sensing extraction method for lodging corn, It is characterized in that The following steps are involved: (1) Obtain remote sensing multispectral images of the study area after lodging, extract images containing red, green, blue, and near-infrared bands, and then perform radiometric calibration, atmospheric correction, orthorectification, and geometric registration on the images; (2) Calculate NDVI, RVI, and DVI on the image G-B , SVI vegetation index; Among them, NDVI is the normalized vegetation index, which is the ratio of the difference between the reflectance of the near-infrared and red bands to their sum; DVI G-B It is the difference between the reflectivity of green and blue bands; RVI is the ratio vegetation index, which is the ratio of the reflectance of the near-infrared band to the red light band; SVI is the shadow vegetation index, which is the product of NDVI and near-infrared band reflectance; (3) The gray-level co-occurrence matrix (GLCM) based on statistical analysis was used to calculate the texture information of lodging and non-lodging corn, rice, and other crops; (4) Constructing a lodging corn extraction model to extract corn lodging distribution information; (5) post-processing the results extracted by the model; (6) Accuracy verification and model parameter adjustment; The step (4) of constructing a lodging corn extraction model to extract corn lodging distribution information comprises the following steps: S1: Use NDVI to remove water bodies, buildings, and some vegetation other than corn; S2: Use B2_Mean to distinguish lodging corn, rice, non-lodging corn, and other vegetation, where B2_Mean is the mean value of green band texture; The step (4) of constructing a lodging corn extraction model to extract corn lodging distribution information comprises the following steps: S1: Setting condition T 0 <NDVI<T 1 ,When the condition is false, the pixel is marked as other features, and when the condition is true, it enters S2 judgment; S2: Set the condition B2_Mean>T 2 ,If the condition is false, enter S3 judgment; if the condition is true, enter S4 judgment; S3: Set the condition NDVI>T 3 , RVI <T 4 , SVI>T 5 , if the condition is true, the pixel is non-lodging corn; S4: Setting Conditional DVI G-B <T 6 , when the condition is false, the pixel is marked as rice, and when the condition is true, the pixel is marked as lodging corn; Among them, T 0 、T 1 、T 6 Both are thresholds, which are determined based on the distribution of sample data and statistical calculations; T 2 is the threshold value, which is obtained from the distribution statistics of sample texture data; T 3 、T 4 、T 5 Both are thresholds, obtained from sample data distribution and statistics respectively.
2. The method according to claim 1, It is characterized in that The image used to extract the results in step (1) is the GF1WFV image, and the image used for accuracy verification in step (6) is the GF-2 image with a resolution of 0.8 m.
3. The method according to claim 1, It is characterized in that NDVI, RVI, DVI in step (2) G-B , The calculation formula of SVI vegetation index is: DVI G-B =ρ Green -r Blue , Among them, ρ Nir , ρ Red , ρ Green ,ρ Blue are the spectral reflectances of the near-infrared, red, green, and blue bands, respectively.
4. The method according to claim 1, It is characterized in that The texture information of the image in step (3) is expressed by the texture mean based on the gray level co-occurrence matrix, and its calculation formula is: Among them, f(i,j|d,θ) represents the probability that a pair of pixels separated by a distance of d pixels in the direction θ have gray levels i and j respectively, where M=N is the gray level and M×N represents the scale of the gray level co-occurrence matrix.
5. The method according to claim 1, It is characterized in that Step (5) post-processes the results extracted by the model, which refers to merging the speckle noise and non-target objects in the results. The post-processing includes filtering, sieve processing and clustering processing to remove small patches and merge categories to obtain raster result data with a more regular shape.
6. The method according to claim 1, It is characterized in that The accuracy verification and model parameter adjustment in step (6) include randomly generating sample points in the study area, comparing the remote sensing extraction results with them, calculating the confusion matrix, and obtaining the classification accuracy. If the overall classification accuracy reaches 90% or above, the requirements are met; if it is lower than 90%, the model parameter threshold is readjusted until the requirements are met.
7. A remote sensing extraction device for lodging corn, It is characterized in that include: Image acquisition and processing module: used to obtain remote sensing multispectral images after lodging in the study area, extract images containing red, green, blue and near-infrared bands, and perform radiation calibration, atmospheric correction, orthorectification and geometric registration on the images; Vegetation index calculation module: used to calculate NDVI, RVI, DVI on images G-B , SVI vegetation index; Texture information calculation module: used to calculate the texture information of lodging and non-lodging corn, rice and other crops based on the gray level co-occurrence matrix (GLCM) of statistical analysis method; Model building module for extracting corn lodging distribution information, including: S1: Setting condition T 0 <NDVI<T 1 ,When the condition is false, the pixel is marked as other features, and when the condition is true, it enters S2 judgment; S2: Set the condition B2_Mean>T 2 ,If the condition is false, enter S3 judgment; if the condition is true, enter S4 judgment; S3: Set the condition NDVI>T 3 , RVI <T 4 , SVI>T 5 , if the condition is true, the pixel is non-lodging corn; S4: Setting Conditional DVI G-B <T 6 , when the condition is false, the pixel is marked as rice, and when the condition is true, the pixel is marked as lodging corn; Among them, T 0 、T 1 、T 6 Both are thresholds, which are determined based on the distribution of sample data and statistical calculations; T 2 is the threshold value, which is obtained from the distribution statistics of sample texture data; T 3 、T 4 、T 5 are thresholds, obtained from sample data distribution and statistics respectively; Model extraction result post-processing module: used to filter, filtration and cluster the speckle noise and non-target objects in the model extraction results to remove small patches and merge categories to obtain raster result data with a more regular shape; Accuracy verification and model parameter adjustment module: used to compare the remote sensing extraction results with the randomly generated sample points in the study area, calculate the confusion matrix, and obtain the classification accuracy. If the overall classification accuracy reaches 90% or above, the requirements are met. If it is lower than 90%, the model parameter threshold is readjusted until the requirements are met.
8. The device according to claim 7, It is characterized in that The vegetation index calculation module contains NDVI, RVI, and DVI. G-B , The calculation formula of SVI vegetation index is: DVI G-B =ρ Green -r Blue , Among them, ρ Nir , ρ Red , ρ Green ,ρ Blue are the spectral reflectances of near-infrared, red, green, and blue bands, respectively; The texture information of the image in the texture information calculation module is expressed by the texture mean based on the gray level co-occurrence matrix, and the calculation formula is: Among them, f(i,j|d,θ) represents the probability that a pair of pixels separated by a distance of d pixels in the direction θ have gray levels i and j respectively, where M=N is the gray level and M×N represents the scale of the gray level co-occurrence matrix.
