Cultivated land mapping method and cultivated land mapping equipment

Abstract

The invention discloses a method and equipment for mapping cultivated land, which includes: obtaining a data set of land use products in a target area; selecting cultivated land and non-cultivated land adjacent to a preset selected point from the target area as verification samples; determining a confusion matrix according to the verification samples and the data set; obtaining a remote sensing image corresponding to the target area to determine the vegetation index, water index and terrain data corresponding to the target area; determining a fusion score layer according to the confusion matrix and the vegetation index; determining a land product distribution map according to the area of ​​cultivated land of administrative units of the same level in the target area and the fusion score layer; selecting land that meets the set requirements from the land product distribution map as a training sample; determining a map of land use products according to the vegetation index, water index and terrain data, and the training sample. The invention can improve the accuracy of mapping land use products.

Description

Technical Field

[0001] The embodiments of the present invention relate to the technical field of digital mapping, and in particular to a method and equipment for mapping cultivated land. Background Art

[0002] Remote sensing technology has been used as an important means to monitor the distribution of land products due to its wide coverage, short revisit period and easy data acquisition. It has long made great contributions to regional, national and global farmland mapping. With the maturity of cloud computing and machine learning, many high-resolution land use / cover (LULC) products and cultivated land products have been produced and made available to the public free of charge, such as CLCD, GlobeLand30, GLAD and GLC_FCS. These products provide valuable data support for crop growth monitoring, yield estimation, food security assessment and other fields.

[0003] However, due to the differences in remote sensing data, mapping methods and definitions of cultivated land used in these products, there will be a certain degree of difference in the mapping results in the same region. For example, remote sensing data with different resolutions and bands may provide different information, which in turn affects the classification results of cultivated land; and the selection of different mapping methods may affect the division of cultivated land spatial distribution. Some mapping methods may rely more on land features, such as vegetation index, time series changes, etc., while others may rely more on machine learning algorithms or expert experience; different definitions of cultivated land may contain different land use types. This difference may lead to differences in cultivated land classification results in the same region. These differences affect the accuracy of cultivated land spatial distribution and area estimation, which may increase the uncertainty in assessing agricultural sustainability and national food security.

[0004] In this context, how to provide a technical solution to improve the accuracy of land use product mapping has become a technical problem that technical personnel in this field urgently need to solve. Summary of the invention

[0005] The problem solved by the embodiments of the present invention is to provide a method and device for mapping cultivated land, which can improve the mapping accuracy of land use products.

[0006] An embodiment of the present invention provides a method for mapping cultivated land, including:

[0007] Acquire a data set of land use products of a target area, wherein the data set of land use products includes data information of cultivated land and data information of non-cultivated land;

[0008] Selecting cultivated land and non-cultivated land adjacent to the preset selected point from the target area as verification samples;

[0009] Determining a confusion matrix based on the validation sample and the data set of the land use product;

[0010] Acquire a remote sensing image corresponding to the target area, and perform interpolation processing on the remote sensing image to determine a vegetation index, a water body index, and terrain data corresponding to the target area;

[0011] Determining a fusion score layer of the land use product according to the confusion matrix and the vegetation index;

[0012] Determine a land product distribution map of the target area according to the area of ​​cultivated land of each administrative unit in the target area and the fusion score map of the land use product;

[0013] Selecting land that meets set requirements from the land product distribution map of the target area as a training sample;

[0014] A classifier is constructed based on the vegetation index, water body index and terrain data corresponding to the target area, as well as the training samples, to determine a map of land use products of the target area.

[0015] Optionally, the step of obtaining a verification sample includes:

[0016] Generate a preset number of preset selection points according to the SHP file of the target area, and the spacing between the preset selection points is greater than the set spacing;

[0017] According to the image data corresponding to the target area, the type of land use product adjacent to each preset selection point is determined, and when it is determined that the image data corresponding to the target area is the first data, the selected land use product adjacent to each preset selection point is regarded as cultivated land; otherwise, it is regarded as non-cultivated land.

[0018] Optionally, determining a confusion matrix according to the validation sample and the data set of the land use product includes:

[0019] Performing a matching and overlaying operation on the verification sample and the data set of the land use product;

[0020] According to the result of the matching and overlaying operation, the number of matches between the cultivated land contained in the verification sample and the cultivated land and non-cultivated land in the data set of the land use product is determined, and the numbers are used as the first matching parameter and the second matching parameter respectively; and the number of matches between the non-cultivated land contained in the verification sample and the cultivated land and non-cultivated land in the data set of the land use product is determined, and the numbers are used as the third matching parameter and the fourth matching parameter respectively;

[0021] The first matching parameter, the second matching parameter, the third matching parameter and the fourth matching parameter are used as elements of the confusion matrix.

[0022] Optionally, the acquiring a remote sensing image corresponding to the target area and performing interpolation processing on the remote sensing image to determine a vegetation index, a water index and terrain data corresponding to the target area includes:

[0023] Using preset image data and using a computing platform corresponding to the preset image data, the preset image data is screened and cropped, and image pixels with clouds, snow and cloud shadows are masked;

[0024] Using linear interpolation processing to fill in the covered or missing data in the preset image data;

[0025] Based on the processed image data, the vegetation index, water index and terrain data corresponding to the target area are determined.

[0026] Optionally, determining a fusion score layer of the land use product according to the confusion matrix and the vegetation index includes:

[0027] When determining that the land use product is cultivated land, determining the overall accuracy corresponding to each cultivated land according to the elements in the confusion matrix;

[0028] The overall accuracy corresponding to each cultivated land is accumulated to determine the synergy index;

[0029] Determining phenological parameters according to the normalized difference vegetation index in the vegetation index;

[0030] According to the phenological parameters and the synergy index, a fusion score layer of the land use product is determined.

[0031] Optionally, the synergy index is determined in the following manner:

[0032]

[0033] Among them, Score crop represents the synergy index, k represents the type of land use product. When the land use product is cultivated land, k is 1; when the land use product is non-cultivated land, k is 0; n represents the total number of land use products, OA i Represents the overall accuracy of the ith land use product.

[0034] Optionally, determining phenological parameters according to the normalized difference vegetation index includes:

[0035] The 95% percentile value of the normalized difference vegetation index in a year is taken as the maximum NDVI value of NDVI in a year. max , and the quantile value at 20% of the first half of the normalized difference vegetation index peak is taken as the starting value of the crop growing season NDVI start , and the quantile value at 30% of the second half of the normalized difference vegetation index peak is taken as the end value of the crop growing season NDVI end ;

[0036] According to the maximum NDVI max , NDVI at the beginning of crop growing season start and NDVI at the end of the crop growing season end , determine the phenological parameters, wherein the phenological parameters are determined in the following manner:

[0037] Score ndvi =(NDVI max -NDVI start )+(NDVI max -NDVI end )

[0038] Among them, Score ndvi Represents phenological parameters.

[0039] Optionally, determining the land product distribution map of the target area according to the area of ​​cultivated land of each administrative unit in the target area and the fusion score map of the land use product includes:

[0040] According to the size of the fused score layer of the land use product, the pixels of the administrative units corresponding to the fused score layer of the land use product are sorted;

[0041] The area of ​​the cultivated land pixels of each administrative unit is selected in sequence and added up. When it is determined that the accumulated value of the cultivated land pixels is the same as or tends to be the same as the total area of ​​the target area, the cultivated land of each administrative unit selected is integrated to determine the land product distribution map of the target area.

[0042] Optionally, selecting land that meets set requirements from the land product distribution map of the target area as a training sample includes:

[0043] The fused score layer in the land product distribution map of the target area is divided into grades according to percentage;

[0044] Selecting a land product distribution map located in a preset range, and sampling from a set proportion of the area in the land product distribution map to obtain candidate training samples;

[0045] Perform spatial filtering processing on the candidate training samples, and use the processed candidate training samples as the training samples.

[0046] Correspondingly, an embodiment of the present invention also provides a farmland mapping device, comprising: a memory and a processor, wherein the memory is suitable for storing one or more computer instructions, and when the processor runs the computer instructions, it executes the farmland mapping method described in any of the above embodiments.

[0047] Compared with the prior art, the technical solution of the embodiment of the present invention has the following advantages:

[0048] In the cultivated land mapping method provided by the embodiment of the present invention, by obtaining a verification sample including cultivated land and its quantity and non-cultivated land and its quantity, a confusion matrix can be determined based on the verification sample and the data set of land use products; at the same time, according to the remote sensing image corresponding to the target area obtained, the vegetation index, water body index and terrain data corresponding to the target area can be determined, and then according to the confusion matrix and the vegetation index, the fusion score layer of the land use product can be determined. By determining the land product distribution map of the target area according to the area of ​​cultivated land of each administrative unit in the target area and the fusion score layer of the land use product, the land that meets the set requirements can be selected from the land product distribution map as a training sample, and the map of the land use product of the target area can be determined based on the obtained vegetation index, water body index and terrain data. By adopting the above mapping method, by combining the data set of land use products and the data value of remote sensing images, the accuracy and consistency of land use product classification can be improved, and then the accuracy of land use product mapping can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 is a flow chart of a method for mapping cultivated land in an embodiment of the present invention;

[0050] Figure 2 is a flow chart of obtaining a verification sample in an embodiment of the present invention;

[0051] Figure 3 is a flow chart for acquiring remote sensing data of a remote sensing image corresponding to a target area in an embodiment of the present invention;

[0052] Figure 4 The present invention is a flowchart for determining a fused score layer in an embodiment of the present invention. DETAILED DESCRIPTION

[0053] As mentioned in the background, there are some urgent problems in the field of mapping. For example, the consistency and accuracy between remote sensing data and statistical data are poor, which can reduce the uncertainty of land spatiotemporal evolution. Although fusing multiple data sets and combining authoritative statistical data is a potential solution, there are still limitations. For example, existing methods have limitations in collecting training samples and may not cover all farming conditions, resulting in insufficient accuracy and consistency in land classification.

[0054] In order to solve the above technical problems, the present invention provides a method for mapping cultivated land, including: by obtaining a verification sample including cultivated land and its quantity and non-cultivated land and its quantity, a confusion matrix can be determined based on the verification sample and a data set of land use products; at the same time, according to the remote sensing image corresponding to the acquired target area, the vegetation index, water body index and terrain data corresponding to the target area can be determined, and then according to the confusion matrix and the vegetation index, a fusion score layer of the land use product can be determined. According to the area of ​​cultivated land of each administrative unit in the target area and the fusion score layer of the land use product, a land product distribution map of the target area can be determined, and then land that meets the set requirements can be selected from the land product distribution map as a training sample, and based on the acquired vegetation index, water body index and terrain data, a map of the land use product of the target area can be determined.

[0055] The present invention can improve the accuracy and consistency of land use product classification by combining the data set of land use products with the data values ​​of remote sensing images, thereby improving the accuracy of land use product mapping.

[0056] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, specific embodiments of the present invention are exemplarily described below in conjunction with the accompanying drawings.

[0057] See also Figure 1 FIG. 1 is a flowchart of a method for mapping cultivated land in an embodiment of the present invention. In some embodiments of the present invention, see Figure 1 , you can use the following steps to draw a map:

[0058] S11, obtaining a data set of land use products in the target area.

[0059] Specifically, the data set of land use products includes parameters such as the type of land use products, the quantity of each type of land use products, and the distribution status of land use products. By obtaining the data set of land use products in the target area, the map of the target area can be determined.

[0060] In this embodiment, the data set of the land use product includes data information of cultivated land and data information of non-cultivated land, wherein the cultivated land and non-cultivated land are lands in different usage states under the same land use product.

[0061] In some other examples, the land use product may also be at least one of forest land, garden land, grassland, water body, wetland and urban area.

[0062] For example, when the land use product is an urban area use product, the land use product may include urban areas and non-urban areas. It can be an urban area; non-cultivated land is a non-urban area.

[0063] In this embodiment, when the data set of land use products of the target area is obtained, splicing, cropping, reclassification and reprojection can be performed according to the scope of the target area to obtain a raster file containing only cultivated land and in a consistent format.

[0064] In this embodiment, publicly released global or Chinese land use or cultivated land products can be selected to download the corresponding original raster files.

[0065] S12, selecting cultivated land and non-cultivated land adjacent to the preset selection point from the target area, and the cultivated land and non-cultivated land are used as verification samples.

[0066] In this embodiment, the preset selection points can be set according to the size of the selected target area, and in the process of obtaining verification samples, in the same selected area, cultivated land and non-cultivated land do not exist at the same time, that is, there are no mixed pixels.

[0067] In some examples, the validation sample should ensure that all pixels within the range of 30m×30m are cultivated land or non-cultivated land, and there are no mixed pixels.

[0068] S13, determining a confusion matrix according to the validation sample and the data set of the land use product.

[0069] Specifically, the cultivated land and non-cultivated land included in the verification sample correspond to the cultivated land and non-cultivated land in the land use product. When there is a matching relationship between the data of the two, the data with the matching relationship can be used as an element in the confusion matrix.

[0070] In some examples, the confusion matrix may include a first matching parameter, a second matching parameter, a third matching parameter, and a fourth matching parameter, wherein the first matching parameter, the second matching parameter, the third matching parameter, and the fourth matching parameter respectively represent the land types and corresponding quantities for which there is a matching relationship between the verification sample and the data set of the land use product.

[0071] S14, acquiring a remote sensing image corresponding to the target area, and performing interpolation processing on the remote sensing image to determine a vegetation index, a water body index and terrain data corresponding to the target area.

[0072] Among them, remote sensing images can represent the image data of the target area. Remote sensing images can be images of different bands such as visible light, infrared, microwave, etc., which can be used to analyze and explain the characteristics and changes of the target area.

[0073] Specifically, by interpolating remote sensing images, the integrity and authenticity of remote sensing image data can be ensured, and based on the remote sensing image data, the accuracy of vegetation index, water index and terrain data corresponding to the target area can be improved.

[0074] In this embodiment, the vegetation index may include at least one of a normalized vegetation index, an enhanced vegetation index, and a soil adjusted vegetation index.

[0075] In a specific embodiment, the vegetation index may include a normalized vegetation index.

[0076] In this embodiment, the water body index may include a normalized water body index and a surface water body index.

[0077] In this embodiment, the terrain data may include altitude and slope data.

[0078] S15, determining a fusion score layer of the land use product according to the confusion matrix and the vegetation index.

[0079] Specifically, the confusion matrix is ​​related to the dataset of land use products, while the normalized difference vegetation index is related to remote sensing images. By using the confusion matrix and the vegetation index at the same time, it is possible to avoid errors in sample point selection caused by the single use of the dataset of land use products or the vegetation index, thereby improving the accuracy of the obtained fusion score layer.

[0080] In this embodiment, in the process of determining the fusion score layer of the land use product, the vegetation index may be the normalized difference vegetation index.

[0081] S16, determining a land product distribution map of the target area according to the area of ​​cultivated land of each administrative unit in the target area and the fusion score map of the land use product.

[0082] S17, selecting land that meets the set requirements from the land product distribution map of the target area as a training sample.

[0083] Specifically, the land product distribution map of the target area includes multiple land products, and the characteristics of land products in different locations are different. In this case, in order to improve the accuracy of the subsequent land use product map, land that meets the set requirements can be selected as training samples.

[0084] S18, constructing a classifier according to the vegetation index, water index and terrain data corresponding to the target area and the training samples, and determining a map of the land use product of the target area.

[0085] Specifically, vegetation index, water index and terrain data are all parameters related to image data, while training samples can characterize the type of land use and related parameters. Using a classifier (such as a random forest classifier), a map of land use products in the target area can be determined.

[0086] Therefore, by adopting the mapping method in the above example, by combining the data set of land use products and the data values ​​of remote sensing images, the accuracy and consistency of land use product classification can be improved, thereby improving the accuracy of land use product mapping.

[0087] In this embodiment, in the step of obtaining the verification sample, the format of the image data obtained when imaging the target area can be selected first, and then based on the value of the image data, it can be determined whether the land use product adjacent to the preset selection point is cultivated land or non-cultivated land.

[0088] As an example, refer to Figure 2 A flowchart of obtaining a verification sample in an embodiment of the present invention is shown in FIG. Figure 2 As shown, the following steps can be performed:

[0089] S21, generating a preset number of preset selection points according to the SHP file of the target area, and the interval between each preset selection point is greater than the set interval.

[0090] Among them, the SHP file refers to a geospatial vector data format used to store geographic elements (such as points, lines and polygons) and their attributes. Based on the SHP file, the size of the target area can be determined, and then the preset selection points that match the area of ​​the target area can be determined.

[0091] In the process of selecting the preset selection points, in order to improve the randomness and discreteness of the preset selection points, the spacing between the preset selection points is greater than the set spacing.

[0092] In a specific embodiment, 1000 preset selection points may be generated based on the SHP file of the target area, and the minimum spacing between the preset selection points is 1000 m.

[0093] It can be understood that the number and spacing of the preset selection points in the above examples are only examples, and the specific values ​​selected are different for different application scenarios.

[0094] S22, determining the type of land use product adjacent to each preset selection point according to the image data corresponding to the target area.

[0095] Specifically, the image data is derived from land use products, and based on the image data corresponding to the target area, the type of land use products adjacent to each preset selection point can be determined.

[0096] In this embodiment, when it is determined that the image data corresponding to the target area is the first data, the selected land use products adjacent to each preset selection point are regarded as cultivated land; otherwise, they are regarded as non-cultivated land.

[0097] For example, if the pixel value corresponding to the first data is 1, the land use product adjacent to each preset selected point is regarded as cultivated land; if it is other values, the land use product adjacent to each preset selected point is regarded as non-cultivated land.

[0098] For example, if the pixel value corresponding to the first data is 10, the land use product adjacent to each preset selected point is regarded as cultivated land; if it is other values, the land use product adjacent to each preset selected point is regarded as non-cultivated land.

[0099] That is, based on the division rules corresponding to different image data, the pixel values ​​corresponding to the first data are different.

[0100] By making the spacing between the preset selected points larger than the set spacing, the randomness and discreteness of the data in the obtained verification samples can be improved, thereby being able to more realistically reflect the distribution status of cultivated land and non-cultivated land in real space.

[0101] In this embodiment, after the verification sample is obtained, the corresponding confusion matrix can be determined according to the verification sample and the data set of the land use product.

[0102] In a specific embodiment, the confusion matrix is ​​obtained in the following manner:

[0103] A matching and overlay operation is performed on the verification sample and the data set of the land use product; based on the result of the matching and overlay operation, the number of matches between the cultivated land contained in the verification sample and the cultivated land and non-cultivated land in the data set of the land use product is determined, and the numbers are used as the first matching parameter and the second matching parameter, respectively; and the number of matches between the non-cultivated land contained in the verification sample and the cultivated land and non-cultivated land in the data set of the land use product is determined, and the numbers are used as the third matching parameter and the fourth matching parameter, respectively.

[0104] Specifically, the validation samples come from the data set of land use products, and then the two can be overlaid to determine whether there is matching data between them.

[0105] For example, when determining that the cultivated land included in the verification sample matches the cultivated land in the data set of the land use product, the number of matches is counted to obtain a first matching parameter.

[0106] When determining whether the cultivated land included in the validation sample matches the non-cultivated land in the data set of the land use product, the number of matches is counted to obtain a second matching parameter.

[0107] When determining whether the non-cultivated land included in the validation sample matches the cultivated land in the data set of the land use product, the number of matches is counted to obtain the third matching parameter.

[0108] When it is determined that the non-cultivated land included in the verification sample matches the non-cultivated land in the data set of the land use product, the number of matches is counted to obtain a fourth matching parameter.

[0109] The first matching parameter, the second matching parameter, the third matching parameter and the fourth matching parameter are used as elements of the confusion matrix.

[0110] As a specific example, the confusion matrix can be in the form of:

[0111]

[0112] Among them, x i，j It represents the number of matches between the i-th land type in the validation sample set and the j-th cultivated land type in the land use product, where i=1,2; j=1,2, where 1 represents cultivated land and 2 represents non-cultivated land.

[0113] That is: x 1，1 Indicates the first matching parameter, x 1，2 Indicates the second matching parameter, x 2，1 Indicates the third matching parameter, x 2，2 Indicates the fourth matching parameter.

[0114] For ease of understanding, the land use products are taken as arable land products as an example.

[0115] Refer to the correspondence between cultivated land types and cultivated land products in the confusion matrix shown in Table 1, wherein, by performing a matching operation, the number of cultivated land and non-cultivated land with a matching relationship in the verification samples and cultivated land products, as well as the final total number N (i.e., the number of preset selection points) can be determined.

[0116] Next, referring to Table 1, in this embodiment, after obtaining the confusion matrix, further processing can be performed to obtain the overall accuracy OA, omission error OE, commission error CE and Matthews correlation coefficient MCC, where:

[0117]

[0118]

[0119] Table 1: Correspondence between cultivated land types and cultivated land products in the confusion matrix

[0120] In this embodiment, the confusion matrix obtained by the above example can be used as a parameter for determining the fused score layer, and subsequent examples can determine another parameter for calculating the fused score layer based on the image data of the target area.

[0121] As an example, see Figure 3 The flowchart of the embodiment of the present invention for obtaining remote sensing data of a remote sensing image corresponding to a target area is shown in FIG. Figure 3 As shown, the following steps are included:

[0122] S31, using preset image data and a computing platform corresponding to the preset image data, screening and cropping the preset image data, and performing mask processing on image pixels having clouds, snow and cloud shadows.

[0123] Specifically, in the process of acquiring the image of the target area, different sampling methods are used, and the corresponding image data formats are different. Therefore, one of the sampling methods can be selected to obtain image data with a preset format type, and then the corresponding computing platform can be used to process the preset image data.

[0124] In this embodiment, the remote sensing image uses Landsat SR images. Alternatively, the Google EarthEngine platform can be used to screen and crop the remote sensing images and mask the pixels with clouds, snow, and cloud shadows.

[0125] Specifically, GEE is used to perform cloud masking on the QA band in the Landsat SR image, where the QA band provides quality information about each pixel, including whether it is covered by clouds, shadows, snow cover, etc.; then, based on the pixel values ​​in the QA band, pixels representing cloud cover are screened out. Generally speaking, a specific range of pixel values ​​in the QA band can be used to identify clouds, for example, pixels above a certain threshold may indicate cloud cover; then, the information of these screened pixels can be used to mark them as cloud cover or mask them in the original image to achieve the effect of cloud removal.

[0126] S32, using linear interpolation processing to fill in the covered or missing data in the preset image data.

[0127] Specifically, the inventors have discovered that in the process of acquiring image data, some data may be lost or masked, and thus linear interpolation can be used to fill in the missing data caused by cloud cover or other factors.

[0128] Among them, linear interpolation assumes that the missing data point is between two known data points, and then estimates the value of the missing data point according to the linear relationship based on the values ​​of the two known data points and the distance between them.

[0129] Specifically, the Landsat SR images are screened to detect missing values ​​in the data. For each missing data point, two known data points adjacent to it in time are found. The weights of the two known data points are calculated based on the time length between the missing data point and its two adjacent known data points. Then, the values ​​of the known data points and the corresponding weights are interpolated according to the linear relationship to obtain the estimated values ​​of the missing data points, and the calculated estimated values ​​are filled into the positions of the missing data points in the original image data.

[0130] S33, determining the vegetation index, water body index and terrain data corresponding to the target area according to the processed image data.

[0131] Among them, the vegetation index can include the normalized difference vegetation index, the enhanced vegetation index and the soil adjusted vegetation index.

[0132] In this embodiment, the water body index may include a normalized water body index and a surface water body index.

[0133] In this embodiment, the terrain data may include altitude and slope data.

[0134] Accordingly, the parameters listed in the above example can be obtained in the following way:

[0135] Normalized Difference Vegetation Index: NDVI = (Nir–Red) / (Nir+Red)

[0136] Enhanced Vegetation Index: EVI = 2.5 × (Nir – Red) / (Nir + 6 × Red - 7.5 × Blue + 1)

[0137] Soil Adjusted Vegetation Index: SAVI = (Nir–Red) / ((Nir+Red+0.5)×1.5

[0138] Normalized water index: NDWI = (Green-Nir / (Green-Nir)

[0139] Surface water index: LSWI = (Nir – Swir1) / (Nir + Swir1)

[0140] Among them, Blue, Green, Red, Nir and Swir1 are band 2-blue, band 3-green, band 4-red, band 5-near infrared, and band 6-shortwave infrared 1 of the image respectively.

[0141] In this embodiment, when the confusion matrix and the vegetation index are obtained, a fusion score layer of the land use product in the target area can be determined.

[0142] In one example, see Figure 4 A flowchart of determining a fused score layer in an embodiment of the present invention is shown in FIG. Figure 4 As shown, the following steps can be performed:

[0143] S41, when it is determined that the land use product is cultivated land, determining the overall accuracy corresponding to each cultivated land according to the elements in the confusion matrix.

[0144] Specifically, from the foregoing content, it can be seen that the elements in the confusion matrix may include a first matching parameter, a second matching parameter, a third matching parameter and a fourth matching parameter. Based on the first matching parameter, the second matching parameter, the third matching parameter and the fourth matching parameter, when determining that the land use product is cultivated land, the overall accuracy corresponding to each cultivated land can be determined.

[0145] In some examples, the overall accuracy corresponding to the cultivated land can refer to the calculation formula of OA in the above examples.

[0146] In this embodiment, the type of the land use product can be determined according to the ground feature pixel value corresponding to the land use product.

[0147] S42, accumulating the overall accuracies corresponding to the various cultivated lands to determine the synergy index.

[0148] In a specific embodiment, the synergy index may be determined in the following manner:

[0149]

[0150] Among them, Score crop represents the synergy index, k represents the type of land use product. When the land use product is cultivated land, k is 1; when the land use product is non-cultivated land, k is 0; n represents the total number of land use products, OA i Represents the overall accuracy of the ith land use product.

[0151] S43, determining phenological parameters according to the normalized vegetation index in the vegetation index.

[0152] Specifically, the normalized difference vegetation index can reflect the greenness of land use products. Therefore, through the normalized difference vegetation index, the response of organisms to seasonal environmental changes can be determined, and then the phenological parameters can be determined.

[0153] In a specific example, it specifically includes: taking the quantile value of the normalized difference vegetation index at 95% in a year as the maximum NDVI of NDVI in a yearmax , and the quantile value at 20% of the first half of the normalized difference vegetation index peak is taken as the starting value of the crop growing season NDVI start , and the quantile value at 30% of the second half of the normalized difference vegetation index peak is taken as the end value of the crop growing season NDVI end ; According to the maximum NDVI max , NDVI at the beginning of crop growing season start and NDVI at the end of the crop growing season end , determine the phenological parameters.

[0154] In a specific embodiment, the phenological parameters are determined in the following manner:

[0155] Score ndvi =(NDVI max -NDVI start )+(NDVI max -NDVI end )

[0156] Among them, Score ndvi Represents phenological parameters.

[0157] S44, determining a fusion score layer of the land use product according to the phenological parameters and the synergy index.

[0158] In one example, the product of the phenological parameter and the synergy index can be used as a fused score layer of the land use product.

[0159] By adopting the method in the above example, by simultaneously considering the confusion matrix and the normalized vegetation index in the vegetation index, that is, considering both the surface characteristics and the vegetation information, the distribution of land use products can be more comprehensively described, thereby improving the reliability and comprehensiveness of the mapping results.

[0160] In this embodiment, considering that the target region is relatively large, errors may occur if mapping is performed directly based on the target region as a whole.

[0161] Based on this, the target area can be divided according to the area of ​​the target area. For example, the target area can be divided according to administrative units with the same administrative level in the target area, so that multiple administrative units with the same administrative level can be obtained.

[0162] In this case, the step of determining the land product distribution map of the target area may include: sorting the pixels of the administrative units corresponding to the fusion score layer of the land use product according to the size of the fusion score layer of the land use product; selecting the area of ​​arable land of each administrative unit in sequence, and performing an accumulation operation, and when it is determined that the area accumulation value of the arable land pixel is the same or tends to be the same as the total area of ​​the target area (for example, when the difference is within 1%), integrating the selected arable land of the administrative units of the same level to determine the land product distribution map of the target area.

[0163] Specifically, determining the land product distribution map of the target area is an iterative process. The pixels in each administrative unit can be sorted based on the value of the determined fusion score layer. The cultivated land in each administrative unit is determined to be the same or close to the total cultivated land area in its statistical data when the total area of ​​the cultivated land pixels in the selected administrative units is the same or close to the total cultivated land area in its statistical data (for example, the total area of ​​the target area can be obtained from relevant websites), indicating that the cultivated land in each administrative unit currently selected can truly reflect the distribution of land products.

[0164] In a specific embodiment, the administrative unit can be a county-level administrative district, on which basis, the following steps can be performed for each county-level administrative district: count starting from the pixel with the highest score, and compare the total area of ​​the selected county-level administrative district with the statistical area; if the total area is less than the statistical area, the next scored pixel is calculated together until the total area reaches or approaches the statistical area. After completing the threshold processing of each statistical area, all results are integrated to determine the land product distribution map of the target area.

[0165] It is understandable that administrative units can also be town-level administrative districts and city-level administrative districts.

[0166] In this embodiment, in order to further improve the accuracy of mapping, after determining the land product distribution map of the target area, land that meets the set requirements can also be selected as training samples.

[0167] As an example, the fused score layer in the land product distribution map of the target area is divided into grades according to percentage; the land product distribution map located in a preset grade is selected, and sampling is performed from the area of ​​a set proportion in the land product distribution map to obtain candidate training samples; the candidate training samples are spatially filtered, and the processed candidate training samples are used as the training samples.

[0168] Specifically, for areas with different fused score layers in the target area, their corresponding weights are different. Therefore, the areas corresponding to each fused score layer can be divided into grades according to the fused score layers in the land product distribution map, and training samples for training the classifier are selected from the areas in the first few grades. Furthermore, by performing spatial filtering on the candidate training samples, the confidence of the training samples can be further improved.

[0169] In a specific embodiment, the fusion scores of the pixels where the collaborative land use product pixels are located can be divided into 5 levels according to the percentage, and stratified sampling is performed in the first 3 levels, and sampling is performed in the land use product pixels that cover at least 40% to 100% of the area to obtain candidate training samples. The candidate training samples will only be retained when it is determined that the candidate training samples are of the same type as the eight surrounding pixels, otherwise they will be discarded. That is, the sample points after spatial filtering are used as training samples in this embodiment, thereby improving the confidence of the training samples.

[0170] In this embodiment, after determining the training samples, the vegetation index, the water index and the terrain data corresponding to the target area, the land map of the target area can be determined.

[0171] In one example, a random forest algorithm can be used to construct a classifier for each administrative unit based on the vegetation index, water index and terrain data corresponding to the target area; the land map corresponding to each administrative unit is determined based on the classifier and training samples corresponding to each administrative unit, and then the land maps of each administrative unit are mosaicked and merged to generate a land map of the target area (such as a cultivated land map).

[0172] In this embodiment, vegetation index and water index corresponding to Landsat SR image data, including normalized vegetation index, enhanced vegetation index, soil adjusted vegetation index, normalized water index, pixel values ​​at 95%, 75%, 50%, 25% and 10% of the surface water index in one year, and terrain data can be used as parameters for constructing a classifier.

[0173] In this embodiment, in order to balance the computing power of the computing platform and the accuracy of the classification results, a random forest algorithm can be used to build a classifier for each administrative unit separately, where the classifier parameter number of Trees can be set to 500, and the number of nodes split each tree each time is set to the square root of the number of variables used.

[0174] By using automatic sample collection and random forest algorithm to build a classifier, the automation and efficiency of the cultivated land mapping process are achieved, saving human resources and time costs.

[0175] This embodiment also provides a farmland mapping device, including a memory and a processor, and the memory and the processor can communicate through a communication bus; the memory stores computer instructions that can be run on the processor, and when the processor runs the computer instructions, it can execute the farmland mapping method described in any of the above embodiments. For details, please refer to the above-mentioned related content, which will not be repeated here.

[0176] In a specific implementation, the processor may include a central processing unit (CPU), a field programmable gate array (FPGA), etc. The memory may include a random access memory (RAM), a read-only memory (ROM), a non-volatile memory (NVM), etc.

[0177] In a specific implementation, the computer instructions may include any suitable type of code implemented using any suitable high-level, low-level, object-oriented, visual, compiled and / or interpreted programming language, such as source code, compiled code, interpreted code, executable code, static code, dynamic code, encrypted code, etc.

[0178] In a specific implementation, the farmland mapping device may further include a display interface and a display connected through the display interface. The display interface may communicate with the memory and the processor through a communication bus. The display may display the result obtained by the processor executing the pre-alignment method provided in the embodiment of this specification.

[0179] In a specific implementation, the farmland mapping device may further include a data output interface. The data output interface may communicate with the memory and the processor via a communication bus to output a land map.

[0180] The present invention also provides a computer-readable storage medium on which computer instructions are stored. When the computer instructions are run, the cultivated land mapping method described in any of the above embodiments of the present invention can be executed. For details, please refer to the above-mentioned related content and will not be repeated here.

[0181] The computer-readable storage medium may include any suitable type of memory unit, memory device, memory article, memory medium, storage device, storage article, storage medium and / or storage unit, such as memory, removable or non-removable medium, erasable or non-erasable medium, writable or rewritable medium, digital or analog medium, hard disk, floppy disk, compact disk read-only memory (CDROM), compact disk recordable (CD-R), compact disk rewritable (CD-RW), optical disk, magnetic medium, magneto-optical medium, removable memory card or disk, various types of digital versatile disks (DVD), magnetic tape, cassette, etc.

[0182] Furthermore, the computer instructions may include any suitable type of code implemented using any suitable high-level, low-level, object-oriented, visual, compiled and / or interpreted programming language, such as source code, compiled code, interpreted code, executable code, static code, dynamic code, encrypted code, etc.

[0183] The present invention also provides a computer program product, which may include computer instructions. When the computer instructions are executed by a processor, the method for mapping cultivated land described in any of the above embodiments of the present invention is implemented. For details, please refer to the above-mentioned related content and will not be repeated here.

[0184] Although the embodiments of the present invention are disclosed above, the present invention is not limited thereto. Any person skilled in the art can make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the scope defined by the claims.

Claims

1. A method for mapping cultivated land, characterized in that: include: Acquire a data set of land use products of a target area, wherein the data set of land use products includes data information of cultivated land and data information of non-cultivated land; Selecting cultivated land and non-cultivated land adjacent to the preset selected point from the target area as verification samples; Determining a confusion matrix based on the validation sample and the data set of the land use product; Acquire a remote sensing image corresponding to the target area, and perform interpolation processing on the remote sensing image to determine a vegetation index, a water body index, and terrain data corresponding to the target area; Determining a fusion score layer of the land use product according to the confusion matrix and the vegetation index; Determine a land product distribution map of the target area according to the area of ​​cultivated land of each administrative unit in the target area and the fusion score map of the land use product; Selecting land that meets set requirements from the land product distribution map of the target area as training samples; According to the vegetation index, water index and terrain data corresponding to the target area and the training samples, a classifier is constructed to determine a map of land use products of the target area, including: according to the vegetation index, water index and terrain data corresponding to the target area, a classifier is constructed separately for each administrative unit using a random forest algorithm; According to the classifiers and training samples corresponding to each administrative unit, the land map corresponding to each administrative unit is determined, and the land maps of each administrative unit are mosaicked and merged to generate a land map of the target area.

2. The method for mapping cultivated land according to claim 1, characterized in that: The steps to obtain a validation sample include: Generate a preset number of preset selection points according to the SHP file of the target area, and the spacing between the preset selection points is greater than the set spacing; According to the image data corresponding to the target area, the type of land use product adjacent to each preset selection point is determined, and when it is determined that the image data corresponding to the target area is the first data, the selected land use product adjacent to each preset selection point is regarded as cultivated land; otherwise, it is regarded as non-cultivated land.

3. The method for mapping cultivated land according to claim 1, characterized in that: Determining a confusion matrix according to the validation sample and the data set of the land use product includes: Performing a matching and overlaying operation on the verification sample and the data set of the land use product; According to the result of the matching and overlaying operation, the number of matches between the cultivated land contained in the verification sample and the cultivated land and non-cultivated land in the data set of the land use product is determined, and the numbers are used as the first matching parameter and the second matching parameter respectively; and the number of matches between the non-cultivated land contained in the verification sample and the cultivated land and non-cultivated land in the data set of the land use product is determined, and the numbers are used as the third matching parameter and the fourth matching parameter respectively; The first matching parameter, the second matching parameter, the third matching parameter and the fourth matching parameter are used as elements of the confusion matrix.

4. The method for mapping cultivated land according to claim 1, characterized in that: The step of acquiring a remote sensing image corresponding to the target area and performing interpolation processing on the remote sensing image to determine a vegetation index, a water index and terrain data corresponding to the target area includes: Using preset image data and using a computing platform corresponding to the preset image data, screening and cropping the preset image data, and performing mask processing on image pixels with clouds, snow and cloud shadows; Using linear interpolation processing to fill in the covered or missing data in the preset image data; Based on the processed image data, the vegetation index, water index and terrain data corresponding to the target area are determined.

5. The method for mapping cultivated land according to any one of claims 1, 3 or 4, characterized in that: Determining the fusion score layer of the land use product according to the confusion matrix and the vegetation index includes: When determining that the land use product is cultivated land, determining the overall accuracy corresponding to each cultivated land according to the elements in the confusion matrix; The overall accuracy corresponding to each cultivated land is accumulated to determine the synergy index; Determining phenological parameters according to the normalized difference vegetation index in the vegetation index; According to the phenological parameters and the synergy index, a fusion score layer of the land use product is determined.

6. The method for mapping cultivated land according to claim 5, characterized in that: The synergy index is determined as follows: Among them, Score crop represents the synergy index, k represents the type of land use product. When the land use product is cultivated land, k is 1; when the land use product is non-cultivated land, k is 0; n represents the total number of land use products, OA i Represents the overall accuracy of the ith land use product.

7. The method for mapping cultivated land according to claim 5, characterized in that: Determining phenological parameters according to the normalized difference vegetation index includes: The 95% percentile value of the normalized difference vegetation index in a year is taken as the maximum NDVI value of NDVI in a year. max , and the quantile value at 20% of the first half of the normalized difference vegetation index peak is taken as the starting value of the crop growing season NDVI start , and the quantile value at 30% of the second half of the peak of the Normalized Difference Vegetation Index is taken as the end value of the crop growing season NDVI end ; According to the maximum NDVI max , NDVI at the beginning of crop growing season start and NDVI at the end of the crop growing season end , determine the phenological parameters, wherein the phenological parameters are determined in the following manner: Score ndvi =(NDVI max -NDVI start )+(NDVI max -NDVI end ) Among them, Score ndvi Represents phenological parameters.

8. The method for mapping cultivated land according to claim 1, characterized in that: Determining the land product distribution map of the target area according to the area of ​​cultivated land of each administrative unit in the target area and the fusion score map of the land use product includes: According to the size of the fused score layer of the land use product, the pixels of the administrative units corresponding to the fused score layer of the land use product are sorted; The area of ​​the cultivated land pixels of each administrative unit is selected in sequence and added up. When it is determined that the accumulated value of the cultivated land pixels is the same as or tends to be the same as the total area of ​​the target area, the cultivated land of each administrative unit selected is integrated to determine the land product distribution map of the target area.

9. The method for mapping cultivated land according to claim 1, characterized in that: The selecting of land meeting set requirements as training samples from the land product distribution map of the target area includes: The fused score layer in the land product distribution map of the target area is divided into grades according to percentage; Selecting a land product distribution map located in a preset range, and sampling from a set proportion of the area in the land product distribution map to obtain candidate training samples; Perform spatial filtering processing on the candidate training samples, and use the processed candidate training samples as the training samples.

10. A farmland mapping device, characterized in that: include: A memory and a processor, wherein the memory is suitable for storing one or more computer instructions, and when the processor runs the computer instructions, it executes the farmland mapping method described in any one of claims 1 to 9.

Images