A high-natural-value farmland identification method based on a high-resolution index combination

Abstract

The present application relates to the technical field of high natural value farmland identification, and provides a high natural value farmland identification method based on a high-resolution index combination, and gives specific steps of data acquisition, data preprocessing, obtaining a land cover map and a land use type, obtaining a vegetation index, obtaining a richness index, and marking an HNVf type farmland. Through analysis of the above high-resolution index system of the present application, it is found through comparison between six randomly selected verification areas in the obtained HNVf identification area and Google map real scene data that the HNVf type farmland elements in five areas have a high proportion, which meets the basic accuracy requirement of high natural value farmland identification. Meanwhile, the combined index identification method also has the characteristics of high accuracy, feasibility, easy acquisition, strong representation, and is suitable for HNVf type farmland identification in China, fills the gap in the research of HNVf in China, and provides a reference direction for the research of high natural value farmland in China.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of high nature value farmland identification, and particularly relates to a high nature value farmland identification method based on a high resolution index combination. BACKGROUND

[0002] High nature value farmland (HNVf), which was proposed in the early 1990s, is a low-intensity agricultural system with biodiversity protection as the core. It is divided into three types of farmland:

[0003] 1) farmland with a high proportion of semi-natural vegetation;

[0004] 2) farmland containing natural and structural elements such as shrubs, woodland, and small streams, or low-intensity agriculture;

[0005] 3) farmland supporting rare plant or animal populations.

[0006] These low-intensity HNVf are of great significance to soil carbon sequestration, farmland landscape heterogeneity, and high ecosystem carrying capacity.

[0007] At present, HNVf identification research is widely carried out in the European Union, and the technology is relatively mature. The European Union provides a special land use type (CORINE Land Cover, CLC) database and an integrated administration and control system (IACS) database for storing related data. Based on these databases, the indicators for identifying HNVf types can be divided into three categories.

[0008] The first type of land use is the basic indicator for identifying HNVf, which is considered as an important representation of the transition from human to nature. It describes the human gradient of the spatial distribution of natural and semi-natural elements in HNVf by dividing the study area into urban areas, farmland, woodland, wetland, and other land use types.

[0009] The second type of agricultural activity indicator is used to analyze the farmland with low-intensity agricultural characteristics in the region by introducing factors such as farm area and input level when the precision of land use type data cannot be improved, and then inferring HNVf.

[0010] The third type of indicator high-precision farm information mainly comes from the IACS database, which provides high-precision farm and plot information updated annually by EU member states, and overcomes the uncertainty of the identification results caused by low resolution of land use types and the lack of sample size of agricultural activity intensity.

[0011] In China, traditional farmland is facing three serious challenges of biodiversity reduction, non-point source pollution and unreasonable land use, and the development of HNVf farmland plays a key role in timely curbing and solving these problems. Therefore, the present patent intends to develop a set of high-resolution indicators suitable for the identification of HNVf farmland in China.

[0012] Although HNVf has global sustainable practical significance, the current HNVf identification technology has three shortcomings.

[0013] The first shortcoming is that the current HNVf identification indicators are limited to the European Community. At present, the indicators for identifying HNVf mainly come from the CLC database and the IACS database provided by the European Community. These databases have the defects of low sample coverage and high access limitation, making it difficult to promote their use on a large scale. In particular, these databases only have data for some regions in the European Community, and various research indicators are established based on the data of the European Community region, so they are not suitable for global use and have certain limitations.

[0014] The second shortcoming is that there is no complete and quantitative reference indicator for the definition of HNVf in China. At present, the research on HNVf identification in China is still in the exploratory stage, and there is no HNVf identification indicator standard and data set suitable for the Chinese region.

[0015] The third shortcoming is that the current land use type indicator for identifying HNVf has low spatial resolution, which can only roughly infer the spatial location of potential HNVf, and cannot further obtain other characteristics (such as agricultural activity intensity) of these areas, so the identification result has great uncertainty. SUMMARY

[0016] To solve the problems in the background art, the present application provides a high natural value farmland identification method based on a combination of high-resolution indicators, which comprises the following steps:

[0017] S1, data acquisition: acquiring PMS remote sensing image data of a satellite, at least containing one N1 resolution panchromatic band and four N2 resolution multi-spectral band data;

[0018] S2, data preprocessing: performing radiation calibration, atmospheric correction and image cropping preprocessing on the data obtained in step S1;

[0019] S3, obtaining land use type map and land use type: performing supervised classification on the preprocessed panchromatic band to obtain an N1 resolution land use type map, and identifying the land use type TP based on the same;

[0020] S4, obtaining a vegetation index: fusing the data after the preprocessing in step S2 to obtain four multispectral fusion bands of N1 resolution, and calculating a vegetation index NDVI according to the multispectral fusion bands;

[0021] S5, obtaining a richness index: calculating a richness index including Shannon diversity SH and Simpson index SI based on the land use type map of N1 resolution obtained in step S3;

[0022] S6, modeling and weighted processing: establishing a Kohonen model for the four indexes TP, NDVI, SH and SI obtained in steps S3-S5, and assigning a weight to each index;

[0023] S7, marking of HNVf farmland: layer superimposition is performed according to the weights of the indexes in step S6 to obtain a result map of HNVf in five grades; the third, fourth and fifth grade areas are marked as HNVf farmland.

[0024] In a preferred scheme, the N1 resolution is 2m and the N2 resolution is 8m.

[0025] In a preferred scheme, the specific process of steps S1 and S2 includes radiation calibration and atmospheric correction on the GF1B satellite PMS image obtained from the natural resource cloud platform; the panchromatic band is subjected to radiation calibration and orthorectification together with the PMS image; and the two are combined with the NNDiffuse Pan Sharpening algorithm for image fusion to obtain a multi-band image of 2m resolution.

[0026] In a preferred scheme, the specific process of step S3 includes supervised classification on the fused image in claim 3 combined with a machine learning algorithm; a land use type map of 2m x 2m resolution is obtained; and land use types including construction land, farmland, forest land, river and semi-natural vegetation are identified according to land classification standards.

[0027] In step S4, the formula for calculating the vegetation index NDVI is as follows:

[0028]

[0029] wherein NIR is the reflectivity of the near-infrared band and R is the reflectivity of the red band.

[0030] In step S5,

[0031] SH estimates the average uncertainty of random pixels belonging to which type of land use, and is particularly sensitive to the non-uniform distribution of various land use types, emphasizing the contribution of rare objects to information; and is calculated according to the following formula:

[0032]

[0033] Wherein, s represents the total number of land use types in the study area, Pi is the total area proportion of the i-th land use type, and the SH value of each image is obtained. When the SH value is equal to 0, it indicates that the region is a single land use type. As the value increases, the number of land use types in the region increases.

[0034] In addition, in step S5, SI is a comprehensive index describing uniformity and richness, and the SI index is less affected by rare land use types and is more inclined to represent uniformity. If the SI index is close to zero, it indicates that the distribution of the research object in a certain region is more uniform, and the diversity is higher. SI is calculated according to the following formula:

[0035]

[0036] s represents the total number of land use types in the study area, and Pi is the total area proportion of the i-th land use type.

[0037] In the preferred scheme, in step S6, the weight of the land use type is 45%, the weight of the vegetation index NDVI is 25%, and the weights of the Shannon diversity SH and the Simpson index SI are both 15%.

[0038] The present application has the following beneficial effects:

[0039] By the set of high-resolution index systems proposed in the present application, it is found that five of the six randomly selected verification regions in the obtained HNVf-identified regions have a high proportion (more than 50%) of HNVf farmland elements, meeting the basic accuracy requirements of high-natural-value farmland identification. At the same time, the combined index identification method also shows high accuracy, feasibility, ease of acquisition, strong representativeness, and is suitable for HNVf farmland identification in China, filling the gap in HNVf research in China and providing a reference direction for high-natural-value farmland research in China. BRIEF DESCRIPTION OF DRAWINGS

[0040] Figure 1 is a flowchart of a high-natural-value farmland identification method based on a set of high-resolution indexes according to the present application.

[0041] Figure 2 is a result map of the five grades of HNVf in Example 1. DETAILED DESCRIPTION

[0042] In order to facilitate those skilled in the art to understand the present application, the specific embodiments of the present application will be described below in conjunction with specific cases.

[0043] Example 1

[0044] Referring to Figure 1 In this embodiment, the required data is all derived from the PMS remote sensing image of GF-1 satellite, including one panchromatic band with 2m resolution and four multispectral bands with 8m resolution. After obtaining the data, the remote sensing image is preprocessed, including radiation calibration, atmospheric correction, image cropping and the like. The radiation calibration and atmospheric correction are performed on the PMS image of GF-1 satellite obtained from the natural resource cloud platform, and the panchromatic band after radiation calibration is subjected to orthorectification together with the PMS image. Then, the two are combined with the NNDiffuse Pan Sharpening algorithm to perform image fusion, so as to obtain a multi-band image with 2m resolution. The NNDiffuse Pan Sharpening algorithm is commonly used for band fusion of high-resolution remote sensing images, and it can achieve minimum mutual overlap of spectral response functions of multiple spectral bands, and all combinations of multispectral bands can better cover the spectral range of the panchromatic band. The following requirements are required for spectral grid when running the algorithm: the pixel size of the low-resolution image grid must be an integral multiple of the pixel size of the high-resolution grid; when the grid projection information of the high-resolution band and the low-resolution band needs to be the same.

[0045] Obtaining high-resolution land use type map: supervised classification is performed on the preprocessed panchromatic band to obtain a land use type map with 2m resolution. The specific process includes that the fused image is combined with a machine learning algorithm to perform supervised classification. The ground object categories refer to the Chinese land use status classification standard. Finally, a land use type map (resolution 2m x 2m) is obtained, and the types include construction land, farmland, forest land, river and semi-natural vegetation. The five categories correspond to 1, 2, 3, 4 and 5 levels. The quantitative index is a land use type map of five levels.

[0046] Obtaining high-resolution vegetation index: one panchromatic band and four multispectral bands after preprocessing are fused to obtain four 2m resolution multispectral fusion bands, and then a high-resolution vegetation index is calculated. The formula for calculating the vegetation index NDVI is as follows:

[0047]

[0048] Among them, NIR is the reflectivity of the near-infrared band, and R is the reflectivity of the red light band.

[0049] Obtaining and analyzing richness index: based on the land use type map with 2m resolution, the landscape index is calculated to obtain the richness index. The index is calculated based on the land use type map, including two indexes of Shannon diversity (SH) and Simpson index (SI). Among them:

[0050] Shannon diversity SH:

[0051] As one of the most commonly used indicators to measure diversity, SH has the advantage of simple concept. SH estimates the average uncertainty of random pixels belonging to which type of land use, and is more sensitive to the non-unbalanced distribution of each land use type, emphasizing the contribution of rare objects to information.

[0052] According to the following formula:

[0053]

[0054] Where s represents the total number of land use types in the study area, Pi is the proportion of the total area occupied by the i-th land use type, and the SH value of each image is obtained. When SH value is equal to 0, it indicates that the regional land use type is single. With the increase of value, the regional land use type increases.

[0055] Pielou index SI:

[0056] SI can be used as a comprehensive index to describe uniformity and richness, and SI index is less affected by rare land use types, and is more inclined to represent uniformity. SI index close to zero indicates that the distribution of the study object in a certain region is more uniform, and its diversity is higher. SI is calculated according to the following formula:

[0057]

[0058] s represents the total number of land use types in the study area, and Pi is the proportion of the total area occupied by the i-th land use type.

[0059] Modeling and weighting processing: Kohonen model is established for the four indicators. Kohonen neural network is a self-organizing neural feature mapping model proposed in 1980, which is based on the research results of physiology and brain science. The neural cells of human brain are orderly arranged in two-dimensional space, and there is lateral interaction between adjacent neural cells, which leads to the emergence of competitive cells. Kohonen neural network is to simulate these characteristics and working mechanism, and to classify and calculate data based on clustering algorithm.

[0060] The flow of Kohonen algorithm is: 1) input sample data to the neural network, calculate the distance between input nodes and output nodes; 2) through the competition between neurons in the competition layer, the winning neuron is obtained; 3) adjust the weight, repeat the training according to the above process until all samples have corresponding winning neurons, and the competition ends; 4) associate all data according to the maximum weight value, and divide into different categories. In this study, Kohonen algorithm is used for weight calculation of recognition indicators, which avoids the subjectivity of prior knowledge weight assignment, and introduces a competition mechanism, which can effectively improve the clustering accuracy. The weight of each indicator is obtained. The weight of land use type is 45%, the weight of vegetation index is 25%, and the weight of SH and SI index is 15%.

[0061] According to the weight of each high-resolution indicator, layer superposition is carried out, and five grades of HNVf result map are obtained. As shown in Figure 2 The third grade (including the third grade) above is considered as HNVf farmland. The division basis is the definition of HNVf: farmland with high proportion of semi-natural vegetation, or low-intensity agriculture dominated farmland containing shrubs, forest land, small rivers and other natural and structural elements, all belong to HNVf. Therefore, the division of five levels is based on the five levels of land use classification, and the five levels of NDVI, SI and SH three indicators are realized by normalizing the interval 1-5.

[0062] The third grade above is recognized as HNVf, high natural value farmland, which contains part of farmland (grade 2), forest land (grade 3), river (grade 4) and semi-natural vegetation (grade 5) except construction land (grade 1), so it is defaulted that the area above the third grade after superposition belongs to HNVf.

[0063] The recognition result of the above HNVf farmland is compared with Google real scene data to determine the recognition accuracy of the application. According to the analysis of the above high-resolution indicator system proposed by the application, six verification areas are randomly selected from the obtained HNVf recognition area and compared with Google map real scene data. It is found that five areas have high proportion of HNVf farmland elements, which meet the basic accuracy requirement of high natural value farmland recognition.

[0064] The above embodiments of the application do not constitute a limitation on the protection scope of the application. Any modifications, equivalent replacements and improvements made within the spirit and principles of the application shall be included in the protection scope of the claims of the application.

Claims

1. A method for identifying high-natural-value farmland based on a combination of high-resolution indicators, characterized in that, The package includes the following steps: S1. Data Acquisition: Acquire PMS remote sensing image data from the satellite, including at least one panchromatic band at N1 resolution and four multispectral bands at N2 resolution; S2. Data preprocessing: Perform radiometric calibration, atmospheric correction, and image cropping preprocessing on the data obtained in step S1. S3. Obtain land cover map and land use type: Supervised classification is performed on the preprocessed panchromatic bands to obtain a land cover map with N1 resolution, and the land use type TP is identified accordingly. S4. Obtain the vegetation index: The data preprocessed in step S2 is fused to obtain four multispectral fusion bands with N1 resolution. The vegetation index NDVI is calculated based on the multispectral fusion bands. S5. Obtain the richness index: Based on the land cover map with N1 resolution obtained in step S3, calculate the richness index, which includes Shannon diversity (SH) and Simpson index (SI). S6. Modeling and weighting: Establish a Kohonen model for the four indicators TP, NDVI, SH, and SI obtained from steps S3-S5, and assign weights to each indicator. S7, HNVf type farmland marking: Layers are overlaid according to the weights of the indicators in step S6 to obtain five levels of HNVf result maps; the third, fourth and fifth level areas are marked as HNVf type farmland.

2. The method for identifying high-natural-value farmland based on a combination of high-resolution indicators according to claim 1, characterized in that: The resolution of N1 is 2m, and the resolution of N2 is 8m.

3. The method for identifying high-natural-value farmland based on a combination of high-resolution indicators according to claim 1, characterized in that: The specific processes of steps S1 and S2 include radiometric calibration and atmospheric correction of the GF1B satellite PMS image acquired by the Natural Resources Cloud Platform, and orthorectification of the panchromatic band image after radiometric calibration together with the PMS image. The two are then combined with the NNDiffuse Pan Sharpening algorithm to perform image fusion, resulting in a 2m resolution multi-band image.

4. The method for identifying high-natural-value farmland based on a combination of high-resolution indicators according to claim 3, characterized in that: The specific process of step S3 includes: performing supervised classification on the fused image from claim 3 using a machine learning algorithm; obtaining a land cover type map with a resolution of 2m×2m; and identifying land use types according to land classification standards, including five types: construction land, farmland, forest land, river, and semi-natural vegetation.

5. The method for identifying high-natural-value farmland based on a combination of high-resolution indicators according to claim 1, characterized in that: In step S4, the formula for calculating the vegetation index NDVI is as follows: Wherein, NIR is the reflectance in the near-infrared band, and R is the reflectance in the red band.

6. The method for identifying high-natural-value farmland based on a combination of high-resolution indicators according to claim 1, characterized in that: In step S5, SH estimates the average uncertainty of which land use category a random pixel belongs to, and is particularly sensitive to the uneven distribution of land cover types, emphasizing the contribution of rare objects to the information; it is calculated using the following formula: Where s represents the total number of land cover types in the study area, and Pi is the proportion of the total area occupied by the i-th land cover type, the SH value of each image is obtained; when the SH value is equal to 0, it indicates that the region has a single land cover type; as the value increases, the number of land cover types in the region increases.

7. The method for identifying high natural value farmland based on a combination of high-resolution indicators according to claim 1, characterized in that: In step S5, SI is a comprehensive index describing evenness and richness. The SI index is less affected by rare land cover types and tends to represent evenness. If the SI index is close to zero, it indicates that the distribution of research objects in a certain area is more even and the diversity is higher. SI is calculated according to the following formula: s represents the total number of land cover types in the study area, and Pi is the proportion of the total area occupied by the i-th land cover type.

8. The method for identifying high natural value farmland based on a combination of high-resolution indicators according to claim 1, characterized in that: In step S6, the weight of land use type TP is 45%, the weight of vegetation index NDVI is 25%, and the weights of Shannon diversity SH and Simpson index SI are both 15%.

Images