Remote sensing extraction method of cultivated land boundary

Abstract

The invention relates to the technical field of remote sensing, in particular to a remote sensing extraction method for a cultivated land boundary, which comprises the following steps of: selecting a high-resolution remote sensing image of a research area in a natural year, and making time sequence NDVI (normalized difference vegetation index) image data in the year; performing edge detection, spatial superposition, cultivated land range extraction and other land type mask processing on the intra-year time sequence NDVI data image to obtain a potential cultivated land edge accumulation frequency graph; classifying cultivated land edges in the potential cultivated land edge accumulation frequency graph into permanent edges, temporary edges and non-edges through an automatic threshold segmentation algorithm, and retaining the permanent edges to obtain a cultivated land boundary graph; by calculating an ECR value, a confusion matrix and Spearman correlation analysis are adopted, comparison with a Geo-Wiki Field Size data set is carried out, and the farmland boundary extraction precision is verified. The method is simple in operation, high in extraction efficiency and accuracy, and beneficial to farmland boundary detection and control.

Description

Technical Field

[0001] The invention relates to the fields of land use change, ground object extraction and remote sensing technology, and in particular to a remote sensing extraction method for cultivated land boundaries. Background Art

[0002] Cultivated land is the basic material for human survival. The quantity and quality of cultivated land determine a country's population carrying capacity and sustainable development capabilities. It shoulders the heavy responsibility of providing food security for mankind and is an important guarantee for national ecological security. Using remote sensing technology to accurately obtain cultivated land boundary information, master cultivated land dynamic monitoring data, and form a comprehensive land evaluation is of great significance for guiding agricultural decision-making and carrying out comprehensive land management. Accurate and timely cultivated land boundary information is crucial for agricultural development, resource management and other related fields.

[0003] As important agricultural information data, the traditional method of obtaining cultivated land boundaries is field survey. The statistical results of this method have the disadvantage of very coarse temporal and spatial resolution. Due to the high cost, systematic censuses are usually conducted only once every few years. The data released in the smallest administrative unit cannot accurately describe spatial differences. For high-resolution images, some scholars have adopted the method of visual interpretation to interpret remote sensing images and mark cultivated land boundaries in sample areas. However, due to the inevitable subjective interference and high labor costs of manual interpretation, it cannot be used as a routine monitoring tool. With the continuous development of global remote sensing satellite observation technology and computers, remote sensing methods are effective and reliable methods to obtain spatially clear and objective data, which can overcome the shortcomings of manual interpretation. The research on automatic mapping of cultivated land boundaries through remote sensing images can be roughly divided into edge-based algorithms, region-based algorithms and hybrid algorithms according to the principle. It has been well applied in large farm areas such as the United States and Europe, but it is difficult to achieve satisfactory results in small-scale agricultural areas. In small-scale agricultural areas, cultivated land boundaries are smaller and more fragmented, and the boundaries between fields are thinner, usually less than 10m, which is less than the spatial resolution of the image. Therefore, it is very difficult to extract cultivated land boundaries in small-scale agricultural areas. Summary of the invention

[0004] In order to solve the problems of farmland boundary extraction and land utilization in small-scale agricultural areas, the present invention proposes a remote sensing extraction method for farmland boundaries with small boundary area and high degree of fragmentation.

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

[0006] A remote sensing extraction method for cultivated land boundaries comprises the following steps:

[0007] S1. Select high-resolution remote sensing images of the study area within a natural year and produce annual time series NDVI image data;

[0008] S2, subjecting the annual time series NDVI data images to edge detection, spatial superposition, cultivated land range extraction, and other land type masking to obtain a potential cultivated land edge cumulative frequency map;

[0009] S3, classifying the cultivated land edges in the potential cultivated land edge cumulative frequency map into permanent edges, temporary edges and non-edges by an automatic threshold segmentation algorithm, retaining the permanent edges, and obtaining a cultivated land boundary map;

[0010] S4. The accuracy of cultivated land boundary extraction was verified by calculating the ECR value, using confusion matrix and Spearman correlation analysis, and comparing with the Geo-Wiki FieldSize dataset.

[0011] Furthermore, in the step S1, a Sentinel-2 remote sensing image within a natural year is selected, and the NDVI value of each remote sensing image is calculated using red light band and near infrared band data, which specifically includes the following steps:

[0012] S11. Select Sentinel-2 remote sensing images of the study area within one natural year;

[0013] S12. Calculate the percentage of clean pixels in the image based on the QA band of the Sentinel-2 remote sensing image, remove images with a clean pixel percentage of less than 99%, and retain images with a clean pixel percentage of more than 99%;

[0014] S13, using the Red band and NIR band of the 10m resolution in the Sentinel-2 remote sensing image to calculate the NDVI value of each remote sensing image obtained in step S12, the formula is as follows:

[0015]

[0016] Wherein, NDVI is the normalized difference vegetation index, which is an important parameter reflecting the growth and nutritional information of crops; NIR is the reflectance of the near-infrared band of the remote sensing image; Red is the reflectance of the red light band of the remote sensing image.

[0017] Furthermore, in the step S2, based on the 10m resolution annual time-series NDVI data image obtained in step S1, an edge detection algorithm is used to identify the edges in each image; the spatial overlay method is used to obtain the cumulative number of each edge; the land use data is used to extract the scope of cultivated land in the study area, and other land types such as built-up areas, water bodies, forests, etc. are masked to obtain a potential cultivated land edge cumulative number map.

[0018] Furthermore, the step S2 specifically includes the following steps:

[0019] S21, using the Canny edge detection algorithm, performing convolution calculation on the NDVI value of each remote sensing image obtained in step S1, extracting the area (edge) where the grayscale changes suddenly, and forming a binary image;

[0020] S22, performing equal-weighted summing and superposition processing on each binary image obtained in step S21 to produce a potential edge frequency map;

[0021] S23. Using the cultivated land in the WorldCover product (2021) as the target land type, perform mask processing on the potential edge frequency map in step S22 to obtain a potential cultivated land edge cumulative frequency map.

[0022] Furthermore, in step S21, the threshold of the Canny edge detection algorithm is set to 100, and Sigma is set to 1.5.

[0023] Furthermore, the step S3 specifically includes the following steps:

[0024] S31, using the OTSU automatic threshold segmentation algorithm once, dividing the edges in the potential cultivated land edge cumulative frequency map obtained in step S2 into true edges and non-edges;

[0025] S32, using the OTSU automatic threshold segmentation algorithm to perform convolution calculation on the true edge obtained in step 31, and dividing the true edge into permanent edges and temporary edges;

[0026] S33. By eliminating non-edges and temporary edges and retaining permanent edges, a stable cultivated land boundary map is obtained.

[0027] Furthermore, the step S4 comprises the following steps:

[0028] S41, using spatial analysis tools to calculate the permanent edge ratio per unit area of ​​cultivated land based on the stable cultivated land boundary map obtained in step S3;

[0029]

[0030] In the formula, ECR is the cultivated land edge ratio. Ideally, the lower the ECR value, the larger the farmland area, and the smaller the permanent edge ratio;

[0031] S42. Based on the Geo-Wiki Field Size dataset, within a 1600m radius grid around the sampling point, the classification threshold is determined by correlation coefficient method and expert experience, and farmland is divided into four size levels: L, M, S and XS according to the ECR value of cultivated land;

[0032] S43. The consistency between the cultivated land size predicted based on the ECR value and the cultivated land size in the Geo-Wiki Field Size dataset was analyzed through the confusion matrix, and the Spearman correlation analysis was performed on the two variables.

[0033]

[0034] In the formula, ρ is the Spearman correlation coefficient, and its value range is [-1, 1]; ρ>0 indicates that there is a positive correlation between the two variables, ρ<0 indicates a negative correlation, and ρ=0 indicates that there is no linear correlation between the two variables; the absolute value of the coefficient indicates the strength of the correlation, and the closer the value is to 1, the stronger the correlation.

[0035] Furthermore, in step S42, the classification thresholds of L, M, S and XS are 0.074, 0.085 and 0.099 respectively.

[0036] The present invention has the following characteristics and beneficial effects:

[0037] The present invention utilizes long-term remote sensing image data, combines land use data and cultivated land data, adopts edge detection, automatic threshold segmentation and other algorithms, constructs cultivated land boundary edge extraction rules, realizes cultivated land boundary extraction, and is beneficial to cultivated land fragmentation monitoring.

[0038] The invention is simple to operate, has high extraction efficiency and accuracy, greatly reduces the workload of manual vectorization, and realizes remote sensing extraction of cultivated land boundary information in the absence of high temporal and spatial resolution image information and limited human subjective conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:

[0040] Figure 1 The present invention provides a flow chart of the cumulative frequency of potential cultivated land edges in a certain area;

[0041] Figure 2 All the edges and permanent edges of cultivated land in a certain area in the embodiment of the present invention;

[0042] Figure 3 The probability density districts and confusion matrix results of the ECR values ​​in a certain area in the embodiment of the present invention. DETAILED DESCRIPTION

[0043] The present invention is described in detail below in conjunction with specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but are not intended to limit the present invention in any form. It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other.

[0044] Aiming at the problem of extracting cultivated land boundaries with small boundary area and large degree of fragmentation formed in small-scale agricultural areas, the present invention proposes a remote sensing extraction method for cultivated land boundaries. The extraction method is based on long-term remote sensing image data, combined with land use data and cultivated land data, and uses edge detection, automatic threshold segmentation and other algorithms to construct cultivated land boundary extraction rules, which is conducive to the extraction and monitoring of cultivated land boundaries. Specifically, the method includes the following steps:

[0045] S1. Produce time series NDVI image data within the year;

[0046] Select high-resolution remote sensing images of the study area within a natural year, and use red light band and near infrared band data to calculate the NDVI value of each remote sensing image; specifically:

[0047] S11. Select Sentinel-2 remote sensing images within a natural year in the study area;

[0048] S12. Calculate the percentage of clean pixels in the image based on the QA band of the Sentinel-2 remote sensing image, remove images with a clean pixel percentage of less than 99%, and retain images with a clean pixel percentage of more than 99%;

[0049] S13, using the Red band and NIR band of 10m resolution in the Sentinel-2 remote sensing image to calculate the NDVI value of each remote sensing image processed in step S12, the formula is as follows:

[0050]

[0051] Wherein, NDVI is the normalized difference vegetation index, which is an important parameter reflecting the growth and nutritional information of crops; NIR is the reflectance of the near-infrared band of the remote sensing image; Red is the reflectance of the red light band of the remote sensing image.

[0052] S2. Prepare a cumulative frequency map of potential cultivated land edges;

[0053] Based on the 10m resolution annual time series NDVI data image obtained in step S1, the edge detection algorithm is used to identify the edges in each image; the spatial overlay method is used to obtain the cumulative number of each edge; the land use data is used to extract the cultivated land range in the study area, and other land types such as built-up areas, water bodies, and forests are masked to obtain the potential cultivated land edge cumulative number map: Specifically:

[0054] S21, using the Canny edge detection algorithm, performing convolution calculation on each NDVI data obtained in step S1, extracting the area (edge) where the grayscale changes suddenly, and forming a binary image;

[0055] S22, performing equal-weighted summing and superposition processing on each binary image obtained in step S21 to produce a potential edge frequency map;

[0056] S23, using the cultivated land in the WorldCover product (2021) as the target land type, performing mask processing on the potential edge frequency map in step S22 to obtain a potential cultivated land edge cumulative frequency map;

[0057] S3. Prepare a stable farmland boundary map;

[0058] Based on the potential cultivated land edge cumulative frequency map obtained in step S2, the cultivated land edge is classified by automatic threshold segmentation algorithm to obtain permanent edge, temporary edge and non-edge, and the permanent edge is retained to make a stable cultivated land boundary map; specifically:

[0059] S31, using the OTSU automatic threshold segmentation algorithm once, dividing the edges in the potential cultivated land edge cumulative frequency map obtained in step S2 into true edges and non-edges;

[0060] S32, for the true edge obtained in step S31, the OTSU automatic threshold segmentation algorithm is used again to perform convolution calculation to divide the true edge into permanent edges and temporary edges;

[0061] S33. By eliminating non-edges and temporary edges and retaining permanent edges, a stable cultivated land boundary map is obtained.

[0062] S4, accuracy verification;

[0063] For the stable cultivated land boundary map obtained in step S3, the ratio of the edge area per unit cultivated land area is calculated and compared with the Geo-Wiki Field Size dataset through confusion matrix, Spearman correlation analysis and other methods to verify the accuracy of the method:

[0064] S41, using spatial analysis tools to calculate the permanent edge ratio per unit area of ​​cultivated land for the stable cultivated land boundary map in step S3;

[0065]

[0066] In the formula, ECR is the cultivated land edge ratio. Ideally, the lower the ECR value, the larger the farmland area, and the smaller the permanent edge ratio;

[0067] S42. Based on the 1600m radius grid around the sampling point of the Geo-Wiki Field Size dataset, farmland is divided into four size levels according to the ECR value of cultivated land through correlation coefficient method and expert experience. The division criteria are shown in the following table:

[0068]

[0069] S43. The consistency between the cultivated land size predicted based on the ECR value and the cultivated land size in the Geo-Wiki Field Size dataset was analyzed through the confusion matrix, and the Spearman correlation analysis was performed on the two variables.

[0070]

[0071] Where ρ is the Spearman correlation coefficient, and its value range is [-1, 1]. ρ>0 indicates that there is a positive correlation between the two variables, ρ<0 indicates a negative correlation, and ρ=0 indicates that there is no linear correlation between the two variables. The absolute value of the coefficient indicates the strength of the correlation, and the closer the value is to 1, the stronger the correlation.

[0072] Example:

[0073] The monitoring object of this embodiment is the Hangjiahu Plain. The terrain of this area is a shallow saucer-shaped depression with Taihu Lake as the center, which rises in the east and south and decreases in the west and north. The cultivated land area is small and the cultivated land is seriously fragmented.

[0074] A remote sensing extraction method for cultivated land boundaries comprises the following steps:

[0075] 1) Select all high-resolution remote sensing images of a certain area in the Hangzhou-Jiaxing-Huzhou Plain in 2021, and use the red light band and near-infrared band data to calculate the NDVI value of each remote sensing image. The specific implementation process is as follows:

[0076] 11) Select all Sentinel-2 remote sensing images of a certain area in the Hangzhou-Jiaxing-Huzhou Plain in 2021;

[0077] 12) According to the QA band of Sentinel-2 remote sensing images, calculate the percentage of clean pixels in the image, eliminate images with a clean pixel percentage of less than 99%, and retain images with a clean pixel percentage of more than 99%;

[0078] 13) The NDVI value of each remote sensing image is calculated using the Red band and NIR band with a resolution of 10m in Sentinel-2 remote sensing images.

[0079] 2) Create a cumulative frequency map of potential cultivated land edges, such as Figure 1 As shown, the specific implementation process is as follows:

[0080] 21) Using the Canny edge detection algorithm, perform convolution calculation on each NDVI data obtained in step 1), extract the area (edge) with grayscale mutation, and form a binary image;

[0081] 22) performing equal-weighted summing and superposition processing on each binary image in step 21) to produce a potential edge frequency map;

[0082] 23) Using the cultivated land in the WorldCover product (2021) as the target land type, perform mask processing on the potential edge frequency map in step 22) to obtain a potential cultivated land edge cumulative frequency map.

[0083] 3) Create a stable farmland boundary map, such as Figure 2 As shown, the specific implementation process is as follows:

[0084] 31) Using the OTSU automatic threshold segmentation algorithm, the edges in the potential cultivated land edge cumulative frequency map in step 2) are divided into true edges and non-edges;

[0085] 32) For the true edge obtained in step 31), the OTSU automatic threshold segmentation algorithm is used again to perform convolution calculation to divide the true edge into permanent edges and temporary edges.

[0086] 33) By eliminating non-edges and temporary edges and retaining permanent edges, a stable farmland boundary map is obtained.

[0087] 4) Accuracy verification, such as Figure 3 As shown, the specific implementation process is as follows:

[0088] 41) Use spatial analysis tools to calculate the percentage of permanent margins per unit area of ​​cultivated land;

[0089] 42) Based on the 1600m radius grid around the sampling point of the Geo-Wiki Field Size dataset, farmland is divided into four size levels according to the ECR value of cultivated land by correlation coefficient method and expert experience, and the classification thresholds are 0.074, 0.085 and 0.099 respectively;

[0090] 43) The consistency between the cultivated land size predicted based on the ECR value and the cultivated land size in the Geo-Wiki Field Size dataset was analyzed through the confusion matrix, and the Spearman correlation analysis was performed on the two variables. The confusion matrix results show that the consistent areas are mainly distributed on small-scale farmland, which is in line with the characteristics of small-scale agriculture in my country. The Spearman correlation coefficient between the farmland size prediction results based on the ECR value and the Geo-Wiki Field Size data reached 0.315, and the ρ value was much less than 0.001, indicating that the two variables have a high positive correlation. This proves the feasibility and scientific validity of the present invention.

[0091] In summary, the remote sensing extraction method of cultivated land boundaries of the present invention utilizes public remote sensing data, has low computational cost, adaptive algorithm parameters, and good spatial generalization ability.

[0092] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and descriptions are only preferred examples of the present invention and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A remote sensing extraction method for cultivated land boundaries, characterized in that: The following steps are involved: S1. Select high-resolution remote sensing images of the study area within a natural year and produce annual time series NDVI image data; S2, subjecting the annual time series NDVI data images to edge detection, spatial superposition, cultivated land range extraction, and other land type masking to obtain a potential cultivated land edge cumulative frequency map; S3, classifying the cultivated land edges in the potential cultivated land edge cumulative frequency map into permanent edges, temporary edges and non-edges by an automatic threshold segmentation algorithm, retaining the permanent edges, and obtaining a cultivated land boundary map; S4. By calculating the ECR value, using confusion matrix and Spearman correlation analysis, and comparing with the Geo-Wiki Field Size dataset, the accuracy of cultivated land boundary extraction was verified.

2. The remote sensing extraction method of cultivated land boundary according to claim 1, characterized in that: In the step S1, a Sentinel-2 remote sensing image within a natural year is selected, and the NDVI value of each remote sensing image is calculated using red light band and near infrared band data, which specifically includes the following steps: S11. Select Sentinel-2 remote sensing images of the study area within one natural year; S12. Calculate the percentage of clean pixels in the image based on the QA band of the Sentinel-2 remote sensing image, remove images with a clean pixel percentage of less than 99%, and retain images with a clean pixel percentage of more than 99%; S13, using the Red band and NIR band of the 10m resolution in the Sentinel-2 remote sensing image to calculate the NDVI value of each remote sensing image obtained in step S12, the formula is as follows: Wherein, NDVI is the normalized difference vegetation index, which is an important parameter reflecting the growth and nutritional information of crops; NIR is the reflectance of the near-infrared band of the remote sensing image; Red is the reflectance of the red light band of the remote sensing image.

3. The remote sensing extraction method of cultivated land boundary according to claim 1, characterized in that: In the step S2, based on the 10m resolution annual time-series NDVI data image obtained in step S1, an edge detection algorithm is used to identify the edges in each image; the spatial overlay method is used to obtain the cumulative number of each edge; the land use data is used to extract the cultivated land range in the study area, and other land types are masked to obtain the potential cultivated land edge cumulative number map.

4. The remote sensing extraction method of cultivated land boundary according to claim 3, characterized in that: The step S2 specifically includes the following steps: S21, using the Canny edge detection algorithm, performing convolution calculation on the NDVI value of each remote sensing image obtained in step S1, extracting the area with grayscale mutation, and forming a binary image; S22, performing equal-weighted summing and superposition processing on each binary image obtained in step S21 to produce a potential edge frequency map; S23. Using the cultivated land in the WorldCover product (2021) as the target land type, perform mask processing on the potential edge frequency map in step S22 to obtain a potential cultivated land edge cumulative frequency map.

5. The remote sensing extraction method of cultivated land boundary according to claim 4, characterized in that: In step S21, the threshold of the Canny edge detection algorithm is set to 100, and Sigma is set to 1.

5.

6. The remote sensing extraction method of cultivated land boundary according to claim 1, characterized in that: The step S3 specifically includes the following steps: S31, using the OTSU automatic threshold segmentation algorithm once, dividing the edges in the potential cultivated land edge cumulative frequency map obtained in step S2 into true edges and non-edges; S32, using the OTSU automatic threshold segmentation algorithm to perform convolution calculation on the true edge obtained in step 31, and dividing the true edge into permanent edges and temporary edges; S33. By eliminating non-edges and temporary edges and retaining permanent edges, a stable cultivated land boundary map is obtained.

7. The remote sensing extraction method of cultivated land boundary according to claim 1, characterized in that: The step S4 comprises the following steps: S41, using spatial analysis tools to calculate the permanent edge ratio per unit area of ​​cultivated land based on the stable cultivated land boundary map obtained in step S3; In the formula, ECR is the cultivated land edge ratio. Ideally, the lower the ECR value, the larger the farmland area, and the smaller the permanent edge ratio; S42. Based on the Geo-Wiki Field Size dataset, within a 1600m radius grid around the sampling point, the classification threshold is determined by correlation coefficient method and expert experience, and farmland is divided into four size levels: L, M, S and XS according to the ECR value of cultivated land; S43. The consistency between the cultivated land size predicted based on the ECR value and the cultivated land size in the Geo-Wiki Field Size dataset was analyzed through the confusion matrix, and the Spearman correlation analysis was performed on the two variables. In the formula, ρ is the Spearman correlation coefficient, and its value range is [-1, 1]; ρ>0 indicates that there is a positive correlation between the two variables, ρ<0 indicates a negative correlation, and ρ=0 indicates that there is no linear correlation between the two variables; the absolute value of the coefficient indicates the strength of the correlation, and the closer the value is to 1, the stronger the correlation.

8. The remote sensing extraction method of cultivated land boundary according to claim 7, characterized in that: In step S42, the classification thresholds of L, M, S and XS are 0.074, 0.085 and 0.099 respectively.

Images