A crop sample extraction method based on vegetation near-infrared reflectance vegetation index

Abstract

The present invention discloses a crop sample extraction method based on vegetation near infrared reflectance vegetation index, which belongs to the field of image data processing methods and agricultural remote sensing, and is used for extracting historical training samples of crops. The present invention uses optical remote sensing data, and realizes crop classification based on the crop samples of the initial year through a random forest algorithm, which is used as the basis for extracting training samples of adjacent years, and the ratio of the NIRv of the remote sensing images of the initial year and the adjacent year is used as a variable input. By setting the variable threshold range, the crop training sample extraction of adjacent years is realized, and the crop historical training sample extraction is realized year by year in this iteration. The present invention can ensure the accuracy of training sample extraction, solve the problem of lack of historical training samples in crop remote sensing classification, and realize crop classification of historical years in the same region.

Description

Technical Field

[0001] The invention discloses a crop sample extraction method based on vegetation near-infrared reflectance vegetation index, and belongs to the field of image data processing method and agricultural remote sensing. Background Art

[0002] When performing crop classification, a sufficiently high-quality ground truth dataset or a visually interpreted dataset is required as a training sample. The quality and quantity of training samples often determine the accuracy of classification, but the process of collecting a large number of high-quality training datasets requires a lot of time, energy, and expertise. The sample extraction method of the existing technology is limited to the years with field survey data, and the number of training samples cannot be expanded for years without field survey data. However, since the crop types between years are not constant, the training samples will change with the change of crop types. This shows that training samples are often time-sensitive, that is, this year's training samples may not be suitable for crop classification in the previous or next year. Especially for existing training samples, it is a major challenge to achieve crop classification in historical years in the same region, so how to obtain a large number of training samples from historical satellite images has received more and more attention. However, no specific method has been developed to collect historical training samples through historical satellite images, and it is urgent to develop a method for extracting historical training samples year by year. Summary of the invention

[0003] The purpose of the present invention is to provide a crop sample extraction method based on vegetation near-infrared reflectance vegetation index to solve the problems of poor quality and small quantity of historical training samples in crop training sample extraction in the prior art.

[0004] A crop sample extraction method based on vegetation near infrared reflectance vegetation index, comprising:

[0005] S1. Obtain satellite images of the crop growth period when the difference in the near infrared reflectance vegetation index NIRv between the two crops is the largest, and obtain the crop classification results of the initial year based on the random forest algorithm using the satellite images and the sampling points of the initial year;

[0006] S2. Acquire crop optical remote sensing data of the initial year, which includes spectral information of near infrared band and red band of satellite images, and calculate NIRv values ​​of different crops in the initial year based on the crop classification results of the initial year;

[0007] S3. Obtain crop optical remote sensing data of the year before or after the initial year, define the year of the crop optical remote sensing data obtained in S3 as an adjacent year, and calculate the NIRv value of the adjacent year based on the crop optical remote sensing data of the adjacent year;

[0008] S4. Calculate the ratio of the NIRv value of the initial year to the NIRv value of the adjacent year

[0009] S5. Build a training sample extraction model and set Threshold range, if the initial year and the adjacent year are in the same position If the value is within the threshold range, it is determined that the location is the same crop in two years, and the crop training samples of adjacent years can be extracted;

[0010] S6. Using the random forest algorithm based on the crop training samples of neighboring years to obtain the classification results of crops in neighboring years;

[0011] S7. Extract crop training samples for the next adjacent year based on the crop classification results of the adjacent years, extract inter-year crop training samples, and further extract historical year-by-year training samples.

[0012] In S2 and S3, the NIRv value is the product of the normalized difference vegetation index NDVI and the near infrared band reflectance NIR: NIR V =NDVI×NIR;

[0013] Where NIR and Red are the near-infrared band reflectance and the red band reflectance, respectively.

[0014] In S4, It is the ratio of NIRv of the adjacent year. The adjacent year includes the initial year and the adjacent year, a total of two years. When the adjacent year is the year after the initial year, The calculation formula is:

[0015] or

[0016] In the formula, n is the initial year; n+1 is the year after the initial year.

[0017] In S4, It is the ratio of NIRv of the adjacent year. The adjacent year includes the initial year and the adjacent year, a total of two years. When the adjacent year is the year before the initial year, The calculation formula is:

[0018] or

[0019] In the formula, n is the initial year; n-1 is the year before the initial year.

[0020] In S5, The threshold is fixed when If the crops are within the threshold range, they are determined to be of the same type and the corresponding satellite images are extracted as training samples for crop classification this year. The threshold ranges are as follows:

[0021]

[0022] In the formula, It is the ratio of NIRv of adjacent years, where adjacent years are the initial year and adjacent years.

[0023] Compared with the prior art, the present invention has the following beneficial effects: the present invention comprehensively considers agricultural knowledge and remote sensing information, constructs a training sample extraction model, realizes the extraction of crop historical training samples year by year, and solves the problem of poor quality and small number of historical training samples in crop remote sensing classification in the prior art. In addition, compared with vegetation indices such as NIRv and NDVI in the TSE model, it contains more photosynthetic physiological information, especially the photosynthetic physiological information difference between C3 crops, including wheat, rice and soybeans and C4 corn crops is greater, and can more effectively extract training samples, which can be applied to the extraction of historical training samples of various crops such as corn, soybeans, wheat, and rice; it is determined that the maximum difference between the near-infrared reflectance vegetation index NIRv of two crops is the key period of crop training samples, and the quality of crop training samples extracted during this period is the highest. The quality of crop training samples is the key factor determining the accuracy of crop classification; based on the extracted crop historical training samples, the historical crop classification is realized by the random forest algorithm, and the inter-annual crop planting type and crop planting area information are obtained, which is of great significance to the scientific formulation of import and export decisions, grain market prices and trade, and agricultural insurance assessment applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a technical flow chart of the crop training sample extraction method based on vegetation near-infrared reflectivity of the present invention;

[0025] Figure 2 This is a diagram of rice training sample extraction results of the crop training sample extraction method based on vegetation near-infrared reflectivity of the present invention;

[0026] Figure 3 This is a diagram of corn training sample extraction results of the crop training sample extraction method based on vegetation near-infrared reflectivity of the present invention;

[0027] Figure 4 This is a soybean training sample extraction result diagram of the crop training sample extraction method based on vegetation near-infrared reflectivity of the present invention; Figure 5 This is a graph of wheat training sample extraction results of the crop training sample extraction method based on vegetation near-infrared reflectivity of the present invention. DETAILED DESCRIPTION

[0028] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention is described clearly and completely below. Obviously, the described embodiments are 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.

[0029] A crop sample extraction method based on vegetation near infrared reflectance vegetation index, comprising:

[0030] S1. Obtain satellite images of the crop growth period when the difference in the near infrared reflectance vegetation index NIRv between the two crops is the largest, and obtain the crop classification results of the initial year based on the random forest algorithm using the satellite images and the sampling points of the initial year;

[0031] S2. Acquire crop optical remote sensing data of the initial year, which includes spectral information of near infrared band and red band of satellite images, and calculate NIRv values ​​of different crops in the initial year based on the crop classification results of the initial year;

[0032] S3. Obtain crop optical remote sensing data of the year before or after the initial year, define the year of the crop optical remote sensing data obtained in S3 as an adjacent year, and calculate the NIRv value of the adjacent year based on the crop optical remote sensing data of the adjacent year;

[0033] S4. Calculate the ratio of the NIRv value of the initial year to the NIRv value of the adjacent year

[0034] S5. Build a training sample extraction model and set Threshold range, if the initial year and the adjacent year are in the same position If the value is within the threshold range, it is determined that the location is the same crop in the two years, and the crop training samples of the adjacent years are extracted; S6. Based on the crop training samples of the adjacent years, the classification results of the crops of the adjacent years are obtained by using the random forest algorithm;

[0035] S7. Extract crop training samples for the next adjacent year based on the crop classification results of the adjacent years, extract inter-year crop training samples, and further extract historical year-by-year training samples.

[0036] In S2 and S3, the NIRv value is the product of the normalized difference vegetation index NDVI and the near-infrared band reflectance NIR: NIRV = NDVI × NIR; Where NIR and Red are the near-infrared band reflectance and the red band reflectance, respectively.

[0037] In S4, It is the ratio of NIRv of the adjacent year. The adjacent year includes the initial year and the adjacent year, a total of two years. When the adjacent year is the year after the initial year, The calculation formula is:

[0038] or

[0039] In the formula, n is the initial year; n+1 is the year after the initial year.

[0040] In S4, It is the ratio of NIRv of the adjacent year. The adjacent year includes the initial year and the adjacent year, a total of two years. When the adjacent year is the year before the initial year, The calculation formula is:

[0041] or

[0042] In the formula, n is the initial year; n-1 is the year before the initial year.

[0043] In S5, The threshold is fixed when If the crops are within the threshold range, they are determined to be of the same type and the corresponding satellite images are extracted as training samples for crop classification this year. The threshold ranges are as follows:

[0044]

[0045] In the formula, It is the ratio of NIRv of adjacent years, where adjacent years are the initial year and adjacent years.

[0046] S1 includes:

[0047] The period when the NIRv difference between the two crops is the key period of crop training samples. Satellite images within the corresponding period are obtained according to the crop type and the key period of crop training samples. The key period of crop training samples for different crop types is shown in the following table:

[0048] Table 1 Critical periods of crop training samples for different crop types

[0049] Crop Type Critical period Corn-Soybean Corn milk stage, soybean maturity stage Corn-Rice Within 1 month after rice transplanting Corn-Wheat Corn jointing stage, wheat maturity stage (applicable to one season per year) Soybean-Rice Within 1 month after rice transplanting soybean-wheat Soybean seedling branching stage, wheat maturity stage (applicable to one season per year) Rice-Wheat Within 1 month after rice transplanting or when wheat matures (applicable to one season per year)

[0050] The technical flow chart of the present invention is as follows: Figure 1As shown in the figure, the random forest algorithm is used to analyze the satellite images of the key period of the initial year according to the sample points of the crop type in the initial year to obtain the crop classification results of the initial year. The crop classification results of the initial year are extracted as sample points for the crop classification of the adjacent years, the NIRv values ​​of the satellite images corresponding to the same period of the adjacent years are calculated, and the training sample extraction model is constructed to determine the NIRv ratio of the adjacent years. The threshold range is used to obtain crop training samples of neighboring years, including the initial year and neighboring years. The classification results of crops in neighboring years are obtained by using the random forest algorithm, and the classification results of crops in neighboring years are used as sample points for the classification of crops in the next neighboring year to continue the extraction of crop training samples in the next neighboring year and the remote sensing classification of crops in the next neighboring year.

[0051] The training sample extraction results of the crop training sample extraction method based on vegetation near-infrared reflectance are shown in Figure 2. Figure 2 , Figure 3 , Figure 4 , Figure 5 As shown in the figure, it can be seen that the model extracts a large number of training samples with a wide distribution range. Using key period images as a prerequisite for extracting training samples means that when extracting training samples, the training samples of all crops are not extracted all at once, but extracted in steps. Table 2 shows the crop classification results based on a small number of training samples and the crop classification results of TSE model training samples. It can be seen from the table that the crop sample extraction TSE model constructed by the present invention can provide high-quality and high-quantity training samples for crop classification in historical years, while ensuring the accuracy of crop remote sensing classification.

[0052] Table 2 Comparison of crop classification results based on a small number of training samples and crop classification results based on TSE model training samples

[0053] The model of the present invention has simple input data, which solves the problem of lack of training samples in crop remote sensing classification. The classification results provide basic data for crop planting area distribution, crop growth monitoring, crop disaster remote sensing assessment and crop yield prediction.

[0054] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some or all of the technical features may be replaced by equivalents, and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A crop sample extraction method based on vegetation near-infrared reflectance vegetation index, characterized in that: include: S1. Obtain satellite images of the crop growth period when the difference in the near infrared reflectance vegetation index NIRv between the two crops in the initial year is the largest, and obtain the crop classification results of the initial year based on the random forest algorithm using the satellite images and the sampling points of the initial year; S2. Acquire crop optical remote sensing data of the initial year, which includes spectral information of near infrared band and red band, and calculate NIRv values ​​of different crops in the initial year based on the crop classification results of the initial year; S3. Obtain crop optical remote sensing data of the year before or after the initial year, define the year where the crop optical remote sensing data obtained in S3 is located as an adjacent year, and calculate the NIRv value of the adjacent year based on the crop optical remote sensing data of the adjacent year; S4. Calculate the ratio P of the NIRv value of the initial year to the NIRv value of the adjacent year NIRV ; S5. Build training sample extraction, model, and set P NIRV Threshold range, if the initial year and the adjacent year have the same position P NIRV If the value is within the threshold range, it is determined that the location is the same crop in two years, and the crop training samples of adjacent years can be extracted; S6. Using the random forest algorithm based on the crop training samples of neighboring years to obtain the classification results of crops in neighboring years; S7. Extract crop training samples for the next adjacent year based on the crop classification results of the adjacent years, extract inter-year crop training samples, and further extract historical year-by-year training samples.

2. The crop sample extraction method based on vegetation near infrared reflectance vegetation index according to claim 1, characterized in that: In S2 and S3, the NIRv value is the product of the normalized difference vegetation index NDVI and the near-infrared band reflectance NIR: Where NIR and Red are the near-infrared band reflectance and the red band reflectance, respectively.

3. The crop sample extraction method based on vegetation near infrared reflectance vegetation index according to claim 2 is characterized in that: In S4, It is the ratio of NIRv of the adjacent year. The adjacent year includes the initial year and the adjacent year, a total of two years. When the adjacent year is the year after the initial year, The calculation formula is: or In the formula, n is the initial year; n+1 is the year after the initial year.

4. The method for extracting crop samples based on vegetation near-infrared reflectance vegetation index according to claim 2, characterized in that: In S4, It is the ratio of NIRv of the adjacent year. The adjacent year includes the initial year and the adjacent year, a total of two years. When the adjacent year is the year before the initial year, The calculation formula is: or Where n is the initial year and n-1 is the year before the initial year.

5. A method for extracting crop samples based on vegetation near-infrared reflectance vegetation index according to any one of claims 3 or 4, characterized in that: In S5, The threshold is fixed when If the crops are within the threshold range, they are determined to be of the same type and the corresponding satellite images are extracted as training samples for crop classification this year. The threshold ranges are as follows: In the formula, It is the ratio of NIRv of adjacent years, where adjacent years are the initial year and adjacent years.

Images