A method for identifying large-scale farming practices by combining high-resolution remote sensing images and open-source DEM

Abstract

The present invention provides a method for identifying large-scale tillage measures by combining high-resolution remote sensing images and open source DEM. A method based on manual visual interpretation is used to identify and extract cultivated land units from high-resolution remote sensing image data at the Tiandi Map L17 level; the Canny edge detection operator is used to extract cultivated land texture, and image processing methods such as morphological changes and probabilistic Hough transform are used to identify straight line segments representing the ridge direction of cultivated land; downscaled contour data is generated from PALSAR terrain data; with the help of the mathematical relationship between contour lines and ridge direction, the size of the ridge direction angle is calculated, thereby identifying the category of tillage measures. The above method realizes the identification and extraction of tillage measures within cultivated land while ensuring accuracy. The method of this study can provide a new idea for the identification of large-scale tillage measures, and the research results can provide a scientific basis for the agricultural sector to plan and adjust the reasonable layout of tillage measures according to local conditions.

Description

Technical Field

[0001] The present invention relates to the field of rapid, efficient, and high-precision remote sensing identification of tillage practices within cultivated land, and in particular to a method for rapidly identifying large-scale tillage practices by combining high-resolution satellite remote sensing image data with image processing technology. More specifically, the present invention relates to a method for identifying large-scale tillage practices by combining high-resolution remote sensing images and open-source DEM (digital elevation model). Background Art

[0002] Due to long-term, continuous, high-intensity reclamation and irrational layout of tillage practices, some lands have experienced environmental problems such as decreased soil fertility, severe soil erosion and degradation, severe soil and water loss, significant crop yield reductions, and degradation of land ecological functions. These issues, on the one hand, impact the sustainable development of agriculture, and on the other, pose a serious threat to food security. However, there is currently limited research on the automatic identification and extraction of tillage practices. Monitoring methods still rely primarily on manual visual interpretation combined with field surveys, which are time-consuming and costly, and cannot efficiently extract the spatiotemporal distribution of tillage practices over large areas, resulting in numerous limitations. High-resolution remote sensing imagery, with its advantages of large-scale, synchronous, and periodic Earth observation, is an important means for extracting tillage practices over long time series and large scales. How to fully exploit remote sensing data for the efficient identification of tillage practices while maintaining precision and accuracy while reducing the workload and time required by traditional methods is an urgent issue.

[0003] High-resolution remote sensing imagery and image processing techniques offer a solution to the problem of identifying and extracting tillage practices. Numerous remote sensing-based techniques for identifying cultivated land can be categorized into three main types: manual visual interpretation, pixel-based methods, and object-oriented approaches. However, while manual visual interpretation of tillage practices can yield highly accurate results, it requires the user to possess a certain level of prior knowledge and is unsuitable for large-scale study areas. Pixel-based methods yield fuzzy boundaries for cultivated land ridges, limiting their identification results to spatial resolution. In combination with object-oriented methods, this increases the time required for image segmentation, sample training, and type determination, making it unsuitable for rapidly extracting tillage practices across large areas and, in practice, unable to meet the demands for efficient, real-time management of agricultural production layouts.

[0004] In contrast, the extraction of tillage measures based on high-resolution remote sensing images combined with image processing technology has more advantages. Ridge direction recognition is one of the research focuses, and its calculation depends on the extraction of ridge direction straight line segments. Therefore, the method of straight line segment detection is also the focus of this study. Since straight line segments can use a small amount of data to store rich information and have the characteristics of high data compression rate, they are widely used in lane line detection, three-dimensional reconstruction, remote sensing image analysis and other aspects. Based on the differences in technical foundations and implementation methods, straight line segment detection methods can be divided into two categories: global Hough transform method and local perception combination method.

[0005] Therefore, in order to address the current problems faced by tillage measure identification and extraction, such as reliance on manual interpretation, the need for field sampling, and the lack of research on large-area tillage measure extraction methods, high-precision, low-cost, and large-scale tillage measure identification and extraction can be achieved through a variety of image processing methods based on high-resolution remote sensing images and DEM data, which can provide theoretical support and methodological reference for subsequent tillage measure extraction and agricultural layout-related issues. Summary of the Invention

[0006] The purpose of the present invention is to provide an information extraction method based on high-resolution remote sensing images and image processing technology, which can extract tillage measures efficiently, quickly, over a large range and with high precision. Specifically, it is a method for identifying large-scale tillage measures by combining high-resolution remote sensing images and open source DEM.

[0007] The method provided by the present invention for identifying large-scale farming practices by combining high-resolution remote sensing images and open-source DEMs includes the following steps:

[0008] S1. Obtain the Tiandi Map satellite remote sensing image data, obtain the RGB three-band data from it, and then stitch and crop the data locally to obtain a high-spatial-resolution remote sensing image that includes the study area;

[0009] S2. Obtain PALSAR terrain data. After the data is locally acquired, project, splice, and crop the data to obtain the pre-processed DEM data of the study area.

[0010] S3. Using the high spatial resolution remote sensing data of the study area obtained in step S1, extracting the land use types of the study area from the data by visual interpretation, including: horizontal ridge cultivation, oblique ridge cultivation, along-slope ridge cultivation, and other types;

[0011] S4. For the high spatial resolution remote sensing data of the study area obtained in step S1, the Canny edge detection operator is used to identify the cultivated land texture of the study area, which includes four parts: Gaussian filtering and smoothing, calculating image gradient, performing non-maximum suppression on the gradient amplitude, and double threshold edge segmentation to obtain preliminary cultivated land texture;

[0012] S5. Based on the preliminary farmland texture obtained in step S4, a binary morphological dilation transform is used to process the edges detected by the Canny operator, and the straight line segments at the broken parts are reconnected to obtain a farmland texture after the binary morphological transform.

[0013] S6. Based on the cultivated land texture after binary morphological transformation, the probabilistic Hough test is used to extract the straight line segments with ridge direction characteristics, which can also achieve the effect of converting curves into straight lines, reducing storage space and computational complexity.

[0014] S7, based on the DEM grid data in the DEM data of the study area pre-processed in step S2, setting the contour line spacing to 1m, you can preliminarily extract the contour lines to obtain contour line vector data;

[0015] S8. Smoothing the contour vector data obtained in step S7 to eliminate the effects of jagged edges and local terrain, more accurately restoring the direction and trend of the contour lines on the original basis, and obtaining the final contour vector data;

[0016] S9, extracting the intersection points and the directions of the line elements between the final contour line vector data and the straight line segments with ridge direction characteristics extracted in step S6, wherein the difference between the contour line direction and the ridge direction at the intersection point of each contour line and the ridge direction is the ridge direction angle;

[0017] S10, identifying the tillage measures according to the ridge angle range finally obtained in step S9;

[0018] S11. The result of manual visual interpretation in step S3 and the result of identification in step S10 are used as the classification attributes of each cultivated land unit. The confusion matrix is ​​calculated according to the classification attributes of each cultivated land unit. The accuracy of the tillage measure extraction results is evaluated. If the accuracy evaluation result does not meet the preset expected accuracy, return to step S4 for parameter adjustment. If the accuracy evaluation result meets the preset expected accuracy, the establishment of the tillage measure identification model has been completed.

[0019] In the method of the present invention for identifying large-scale farming practices by combining high-resolution remote sensing images and open source DEM, in step S1, the Tiandi Map satellite remote sensing image data uses L17-level Tiandi Map satellite remote sensing image data with a spatial resolution of 1.19 meters and a projection coordinate system set to WGS1984 UTM Zone 52N, and the grayscale value of each pixel of the high spatial resolution remote sensing image is calculated.

[0020] In the method of the present invention for identifying large-scale farming measures by combining high-resolution remote sensing images and open source DEM, in step S4, the Canny edge detection operator is used to identify the texture of cultivated land in the study area, specifically including: first sampling, quantizing and normalizing the two-dimensional Gaussian function, constructing a 7*7 Gaussian template, removing the sudden pixel points, and then using the Sobel operator to calculate the gradient direction of each point on the edge, and dividing the points into four categories: horizontal, vertical, diagonal and anti-diagonal; secondly, using the linear interpolation method to calculate the gradient value of the edge; finally, comparing the gradient value of the point with the points on both sides along the gradient direction. If the value of the target point is the largest, the point is retained, otherwise it is suppressed; in order to reduce the interference of noise points, a high threshold TH and a low threshold TL are set, and all points above TH are marked as edges, and then the points above TL among the eight connected points of the edge are also marked as edges.

[0021] In the method of combining high-resolution remote sensing images and open source DEM to identify large-scale farming measures, in step S5, a binary morphological dilation transform is used to process the edges detected by the Canny operator. Specifically, each element x in each edge point set X is expanded to S+x using the structural element S, which is recorded as Reconnect the straight line segments at the break, where:

[0022]

[0023] In the method of the present invention for identifying large-scale tillage practices by combining high-resolution remote sensing images and open-source DEMs, in step S6, a probabilistic Hough test is used to randomly select pixels from a set of non-zero pixels in the image for Hough test. Two threshold parameters, minLineLength and maxLineGap, are added to the standard method. A detected straight line is not considered a straight line when its minimum length is less than the threshold minLineLengt, and is considered a straight line when the maximum interval between two points on the detected straight line is less than the threshold maxLineGap.

[0024] In the method of the present invention for identifying large-scale farming measures by combining high-resolution remote sensing images and open source DEM, the accumulator threshold parameter during the probabilistic Hough detection is set to 3, minLineLength is set to 20, and maxLineGap is set to 3.

[0025] In the method of the present invention for identifying large-scale farming measures by combining high-resolution remote sensing images and open source DEM, the contour lines are smoothed using a polynomial approximation method with an exponential kernel and a fitting tolerance of 100 meters to eliminate the influence of jagged edges and local terrain, and to restore the direction and trend of the contour lines on the original basis.

[0026] In the method of identifying large-scale farming measures by combining high-resolution remote sensing images and open-source DEMs, the difference between the contour line direction and the ridge direction at the intersection of each contour line and the ridge direction is the ridge direction angle, which is within the range of [0, 90°] and specifically includes:

[0027] Extract the intersection points of ridge straight line segments and contour lines;

[0028] Construct an intersection buffer with a buffer radius of 0.01 meters. Calculate the intersection points between the buffer and the contour lines and the ridge straight line segments, which are called line feature intersection points, to distinguish the intersection points between the ridge straight line segments and the contour lines.

[0029] Then calculate the x and y coordinates of the intersection point and the intersection point of the line element, and then find the direction α of each line element according to the following formula:

[0030]

[0031] Among them, xi and yi are the x and y coordinates of the intersection point, respectively, and x1 and y1 are the x and y coordinates of the intersection point of the line elements;

[0032] According to the obtained line element direction α, the contour line direction and ridge direction at each contour line and ridge direction intersection are obtained. The absolute value of the difference between the two is taken, and the complementary angle is taken for the angle value whose absolute value is not in [0°, 90°] to obtain the ridge direction angle.

[0033] In the method of the present invention for identifying large-scale tillage measures by combining high-resolution remote sensing images and open source DEM, step S10 for identifying tillage measures is specifically as follows: when the ridge angle range is [0°, 10°], it is defined as cross-slope ridge cultivation; when the ridge angle range is (10°, 80°], it is defined as slope ridge cultivation; when the ridge angle range is (80°, 90°], it is defined as slope ridge cultivation.

[0034] In the method of the present invention for identifying large-scale farming practices by combining high-resolution remote sensing images and open-source DEM, in step S11, the accuracy evaluation is performed using the Kappa coefficient, and the specific calculation formula is as follows:

[0035] Kappa=(pp e ) / (1-p e )

[0036] in,

[0037] p e =(a1*b1+a2*b2+…+a i* b i ) / (n*n)

[0038] Where a k represents the number of real samples of the kth type of ground objects, b kRepresents the predicted value of the sample of the kth category of land features, n represents the total number of samples, p represents the probability of correct classification among all samples, k = 1, 2, 3, ..., i, i is the total number of categories of land features.

[0039] The beneficial effects of the technical solution provided by the present invention are: based on high spatial resolution imaging and image processing technology, the present invention greatly improves the efficiency of extracting tillage measures and the accuracy of the results. The above method realizes the refined identification and extraction of the spatiotemporal distribution of tillage measures while ensuring accuracy. DETAILED DESCRIPTION

[0040] In order to make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be further described below.

[0041] The method provided in this embodiment for identifying large-scale farming practices by combining high-resolution remote sensing images and open-source DEMs includes the following steps:

[0042] S1. Obtain the Tiandi Map satellite remote sensing image data, obtain the RGB three-band data from it and transfer it to the local area, then perform data splicing and cropping to obtain a high-spatial-resolution remote sensing image that includes the study area.

[0043] In this embodiment, the Tiandi Map satellite remote sensing image data uses L17-level Tiandi Map satellite remote sensing image data with a spatial resolution of 1.19 meters and a projection coordinate system set to WGS1984 UTM Zone 52N, and the grayscale value of each pixel of the high spatial resolution remote sensing image is calculated.

[0044] S2. Obtain PALSAR terrain data. After the data is locally acquired, project, splice, and crop the data to obtain the pre-processed DEM data of the study area.

[0045] S3. For the high spatial resolution remote sensing data of the study area obtained in step S1, visual interpretation is used to extract the land use types of the study area, including: transverse slope ridge cultivation, oblique slope ridge cultivation, along slope ridge cultivation and other types, and the codes of transverse slope ridge cultivation, oblique slope ridge cultivation and along slope ridge cultivation are assigned to 1, 2 and 3 respectively, and the codes of the other types of land use are all assigned to 0 to facilitate subsequent type identification and reading.

[0046] S4. For the high spatial resolution remote sensing data of the study area obtained in step S1, the Canny edge detection operator is used to identify the cultivated land texture in the study area, which includes four parts: Gaussian filtering smoothing, calculating image gradient, non-maximum suppression of gradient amplitude and double threshold edge division to obtain preliminary cultivated land texture.

[0047] The Canny edge detection operator was used to identify the texture of cultivated land in the study area. Specifically, the method includes: first sampling, quantizing and normalizing the two-dimensional Gaussian function, constructing a 7*7 Gaussian template, removing the sudden pixel points, and then using the Sobel operator to find the gradient direction of each point on the edge, and dividing the points into four categories: horizontal, vertical, diagonal and anti-diagonal; secondly, using the linear interpolation method to calculate the gradient value of the edge; finally, comparing the gradient value of the point with the points on both sides along the gradient direction. If the value of the target point is the largest, the point is retained, otherwise it is suppressed; in order to reduce the interference of noise points, a high threshold TH and a low threshold TL are set, and all points above TH are marked as edges. Then, the points above TL among the eight connected points of the edge are also marked as edges.

[0048] Among them, the two-dimensional Gaussian function is:

[0049]

[0050] x, y are the horizontal and vertical coordinates of the coordinate point respectively, and σ represents the standard deviation;

[0051] The Gaussian template is:

[0052]

[0053] gradient:

[0054]

[0055] G in the formula x and G y are the horizontal and vertical gradient operator templates of the Sobel operator, and T is the transpose:

[0056]

[0057]

[0058] Gradient Amplitude:

[0059]

[0060] Gradient direction angle:

[0061]

[0062] S5. Based on the preliminary farmland texture obtained in step S4, a binary morphological dilation transformation is used to process the edges detected by the Canny operator, and the straight line segments at the broken parts are reconnected to obtain the farmland texture after the binary morphological transformation.

[0063] The binary morphological dilation transformation is used to process the edges detected by the Canny operator. Specifically, each element x in each edge point set X is expanded to S+x using the structural element S, which is recorded as Expressed in set language as

[0064]

[0065] in:

[0066]

[0067] S6. Based on the farmland texture after binary morphological transformation, the probabilistic Hough test is used to extract straight line segments with ridge direction characteristics, which can also achieve the effect of converting curves into straight lines, reducing storage space and computational complexity.

[0068] Since the standard Hough transform requires calculation and statistics for each point, in order to reduce the amount of calculation, this study uses probabilistic Hough transform to detect straight lines in the edge.

[0069] Using probabilistic Hough detection, pixels are randomly selected from the non-zero pixel set of the image for Hough detection. Two threshold parameters are added to the standard method: minLineLength and maxLineGap. When the minimum length of the detected line is less than the threshold minLineLengt, it is not considered a line. When the maximum interval between two points on the detected line is less than the threshold maxLineGap, it is considered a line. The specific definitions are shown in the following table:

[0070]

[0071] The present invention sets different parameter thresholds. After comparison, it is found that when the accumulator threshold parameter of the probabilistic Hough detection is set to 3, minLineLength is set to 20, and maxLineGap is set to 3, the straight line segment detection effect of the probabilistic Hough transform is relatively better.

[0072] S7: Based on the 12.5m resolution DEM raster data from the study area preprocessed in step S2, set the contour interval to 1m to initially extract contour lines and obtain contour vector data. This step can extract terrain information from the larger-scale DEM data that matches the resolution of the Tiantu Map remote sensing image.

[0073] In step S7, the resolution of the DEM data used differs by an order of magnitude from that of the Tiandi Map data. This means that approximately 10*10 pixels in the Tiandi Map image are matched to one pixel in the DEM data, and the elevation information of many small Tiandi Map pixels is smoothed out. Therefore, directly using the DEM data for tillage practice extraction would significantly affect recognition accuracy. To mitigate this issue, the present invention combines the characteristics of raster and vector data and the conversion relationship between the two, using a downscaling approach to extract terrain information from the DEM that matches the Tiandi Map image.

[0074] S8. Based on the contour vector data obtained in step S7, the data is smoothed to eliminate the influence of jagged edges and local terrain, and the direction and trend of the contour lines are restored more accurately on the original basis to obtain the final contour vector data.

[0075] The data was smoothed using the polynomial approximation with exponential kernel (PAEK) method and a fitting tolerance of 100 meters to eliminate the effects of jagged edges and local terrain, and to restore the direction and trend of the contour lines based on the original data.

[0076] S9. According to the final contour line vector data and the straight line segment with ridge direction characteristics extracted in step S6, the intersection point and the direction of the line element between the two are extracted. The difference between the contour line direction and the ridge direction at each intersection point of the contour line and the ridge direction is the ridge direction angle.

[0077] S10. Identify the tillage measures according to the ridge direction angle range finally obtained in step S9, specifically: when the ridge direction angle range is [0°, 10°], it is defined as cross-slope ridge cultivation; when the ridge direction angle range is (10°, 80°], it is defined as slope ridge cultivation; when the ridge direction angle range is (80°, 90°], it is defined as slope ridge cultivation.

[0078] S11. The results of the manual visual interpretation in step S3 and the results of the identification in step S10 are used as the classification attributes of each piece of cultivated land unit. The confusion matrix is ​​calculated based on the classification attributes of each piece of cultivated land unit, and the accuracy of the tillage measure extraction results is evaluated. If the accuracy evaluation result does not meet the preset expected accuracy, return to step S4 for parameter adjustment. If the accuracy evaluation result meets the preset expected accuracy, the establishment of the tillage measure identification model has been completed. The tillage measure identification model can be used to realize the identification of other large-scale tillage measures. The difference between the contour line direction and the ridge direction at the intersection of each contour line and the ridge direction is the ridge direction angle, and its size is within the range of [0,90°], specifically including:

[0079] Extract the intersection points of ridge straight line segments and contour lines;

[0080] Construct an intersection buffer with a buffer radius of 0.01 meters. Calculate the intersection points between the buffer and the contour lines and the ridge straight line segments, which are called line feature intersection points, to distinguish the intersection points between the ridge straight line segments and the contour lines.

[0081] Then calculate the x and y coordinates of the intersection point and the intersection point of the line element, and then find the direction α of each line element according to the following formula:

[0082]

[0083] Among them, xi and yi are the x and y coordinates of the intersection point, respectively, and x1 and y1 are the x and y coordinates of the intersection point of the line elements;

[0084] Based on the obtained line element direction α, the contour line direction and ridge direction at each intersection are determined. The absolute value of the difference between the two is taken, and the complementary angle is taken for angles whose absolute values ​​are not in the range [0°, 90°] to obtain the ridge angle. This process allows the correct ridge angle to be calculated at each intersection. Finally, the ridge angle at each intersection is classified to determine the tillage policy.

[0085] In the method of the present invention for identifying large-scale farming measures by combining high-resolution remote sensing images and open source DEM, in step S11, the accuracy evaluation is performed using the Kappa coefficient, and the specific calculation formula is as follows:

[0086] Kappa=(pp e ) / (1-p e )

[0087] in,

[0088] p e =(a1*b1+a2*b2+…+a i* b i ) / (n*n)

[0089] Where a k represents the number of real samples of the kth type of ground objects, b k Represents the predicted value of the sample of the kth category of land features, n represents the total number of samples, p represents the probability of correct classification among all samples, k = 1, 2, 3, ..., i, i is the total number of categories of land features.

[0090] In another embodiment of the present invention, on the basis of the Kappa coefficient, three indicators, namely, overall accuracy OA, user accuracy UA, and producer accuracy PA, are selected for auxiliary evaluation. The specific calculation formula is as follows:

[0091] OA=(TP+TN) / (TP+FN+FP+TN)

[0092] UA=TP / (TP+FP)

[0093] PA=TP / (TP+FN)

[0094] Where TP is the number of samples predicted as positive samples when they are actually positive samples, FP is the number of samples predicted as positive samples when they are actually negative samples, TN is the number of samples predicted as negative samples when they are actually negative samples, and FN is the number of samples predicted as negative samples when they are actually positive samples.

[0095] The present invention is based on high spatial resolution imaging and image processing technology, which greatly improves the efficiency of extracting farming measures and the accuracy of the results. The above method can achieve the fine recognition and extraction of the spatiotemporal distribution of farming measures while ensuring the accuracy.

[0096] In the absence of conflict, the above embodiments and features in the embodiments may be combined with each other.

[0097] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for identifying large-scale farming practices by combining high-resolution remote sensing images and open-source DEM, characterized in that: The steps include: S1. Obtain the Tiandi Map satellite remote sensing image data, obtain the RGB three-band data from it, and then stitch and crop the data locally to obtain a high-spatial-resolution remote sensing image that includes the study area; S2. Obtain PALSAR terrain data. After the data is locally acquired, project, splice, and crop the data to obtain the pre-processed DEM data of the study area. S3. Using the high spatial resolution remote sensing data of the study area obtained in step S1, extracting the land use types of the study area from the data by visual interpretation, including: horizontal ridge cultivation, oblique ridge cultivation, along-slope ridge cultivation, and other types; S4. For the high spatial resolution remote sensing data of the study area obtained in step S1, the Canny edge detection operator is used to identify the cultivated land texture of the study area, which includes four parts: Gaussian filtering and smoothing, calculating image gradient, performing non-maximum suppression on the gradient amplitude, and double threshold edge segmentation to obtain preliminary cultivated land texture; S5. Based on the preliminary farmland texture obtained in step S4, a binary morphological dilation transform is used to process the edges detected by the Canny operator, and the straight line segments at the broken parts are reconnected to obtain a farmland texture after the binary morphological transform. S6. Based on the cultivated land texture after binary morphological transformation, the probabilistic Hough test is used to extract the straight line segments with ridge direction characteristics, which can also achieve the effect of converting curves into straight lines, reducing storage space and computational complexity. S7, based on the DEM grid data in the DEM data of the study area pre-processed in step S2, setting the contour line spacing to 1m, you can preliminarily extract the contour lines to obtain contour line vector data; S8. Smoothing the contour vector data obtained in step S7 to eliminate the effects of jagged edges and local terrain, more accurately restoring the direction and trend of the contour lines on the original basis, and obtaining the final contour vector data; S9, extracting the intersection points and the directions of the line elements between the final contour line vector data and the straight line segments with ridge direction characteristics extracted in step S6, wherein the difference between the contour line direction and the ridge direction at the intersection point of each contour line and the ridge direction is the ridge direction angle; S10, identifying the tillage measures according to the ridge angle range finally obtained in step S9; S11. The result of manual visual interpretation in step S3 and the result of identification in step S10 are used as the classification attributes of each cultivated land unit. The confusion matrix is ​​calculated according to the classification attributes of each cultivated land unit. The accuracy of the tillage measure extraction results is evaluated. If the accuracy evaluation result does not meet the preset expected accuracy, return to step S4 for parameter adjustment. If the accuracy evaluation result meets the preset expected accuracy, the establishment of the tillage measure identification model has been completed.

2. The method for identifying large-scale farming practices by combining high-resolution remote sensing images and open-source DEM according to claim 1, characterized in that: In step S1, the Tiandi Map satellite remote sensing image data uses L17 level Tiandi Map satellite remote sensing image data with a spatial resolution of 1.19 meters and a projection coordinate system set to WGS1984 UTM Zone 52N, and the grayscale value of each pixel of the high spatial resolution remote sensing image is calculated.

3. The method for identifying large-scale farming practices by combining high-resolution remote sensing images and open-source DEM according to claim 1, characterized in that: In step S4, the Canny edge detection operator is used to identify the texture of cultivated land in the study area, which specifically includes: first sampling, quantizing and normalizing the two-dimensional Gaussian function, constructing a 7*7 Gaussian template, removing the sudden pixel points, and then using the Sobel operator to find the gradient direction of each point on the edge, and dividing the points into four categories: horizontal, vertical, diagonal and anti-diagonal; secondly, using the linear interpolation method to calculate the gradient value of the edge; finally, comparing the gradient value of the point with the points on both sides along the gradient direction. If the value of the target point is the largest, the point is retained, otherwise it is suppressed; in order to reduce the interference of noise points, a high threshold TH and a low threshold TL are set, and all points above TH are marked as edges, and then the points above TL among the eight connected points of the edge are also marked as edges.

4. The method for identifying large-scale farming practices by combining high-resolution remote sensing images and open-source DEM according to claim 1, characterized in that: In step S5, the binary morphological dilation transform is used to process the edges detected by the Canny operator. Specifically, each element x in each edge point set X is expanded to S+x using the structure element S, denoted as X⊕S, and the straight line segments at the break are reconnected, where:

5. The method for identifying large-scale farming practices by combining high-resolution remote sensing images and open-source DEM according to claim 1, characterized in that: In step S6, a probabilistic Hough detection is used to randomly select pixels from the non-zero pixel set of the image for Hough detection. In addition, two threshold parameters are added to the standard method: minLineLength and maxLineGap. When the minimum length of a detected straight line is less than the threshold minLineLengt, it is not considered a straight line. When the maximum interval between two points on a detected straight line is less than the threshold maxLineGap, it is considered a straight line.

6. The method for identifying large-scale farming practices by combining high-resolution remote sensing images and open-source DEM according to claim 5, characterized in that: The accumulator threshold parameter during the probabilistic Hough detection is set to 3, minLineLength is set to 20, and maxLineGap is set to 3.

7. The method for identifying large-scale farming practices by combining high-resolution remote sensing images and open-source DEM according to claim 1, characterized in that: In step S8, the data is smoothed using a polynomial approximation method with an exponential kernel and a fitting tolerance of 100 meters to eliminate the effects of jagged edges and local terrain, and to restore the direction and trend of the contour lines on the original basis.

8. The method for identifying large-scale farming practices by combining high-resolution remote sensing images and open-source DEM according to claim 1, characterized in that: The difference between the contour line direction and the ridge direction at the intersection of each contour line and the ridge direction is the ridge direction angle, which is within the range of [0,90°] and specifically includes: Extract the intersection points of ridge straight line segments and contour lines; Construct an intersection buffer with a buffer radius of 0.01 meters. Calculate the intersection points between the buffer and the contour lines and the ridge straight line segments, which are called line feature intersection points, to distinguish the intersection points between the ridge straight line segments and the contour lines. Then calculate the x and y coordinates of the intersection point and the intersection point of the line element, and then find the direction α of each line element according to the following formula: Among them, xi and yi are the x and y coordinates of the intersection point, respectively, and x1 and y1 are the x and y coordinates of the intersection point of the line elements; According to the obtained line element direction α, the contour line direction and ridge direction at each contour line and ridge direction intersection are obtained. The absolute value of the difference between the two is taken, and the complementary angle is taken for the angle value whose absolute value is not in [0°, 90°] to obtain the ridge direction angle.

9. The method for identifying large-scale farming practices by combining high-resolution remote sensing images and open-source DEM according to claim 1, characterized in that: Step S10 identifies the tillage measures as follows: when the ridge angle range is [0°, 10°], it is defined as cross-slope ridge cultivation; when the ridge angle range is (10°, 80°], it is defined as slope ridge cultivation; when the ridge angle range is (80°, 90°], it is defined as slope ridge cultivation.

10. The method for identifying large-scale farming practices by combining high-resolution remote sensing images and open-source DEM according to claim 1, characterized in that: In step S11, the accuracy evaluation is performed using the Kappa coefficient, and the specific calculation formula is as follows: Kappa=(pp e ) / (1-p e ) in, <h2 style=";text-align:left;direction:ltr">p<h2 style=";text-align:left;direction:ltr"> e <h2 style=";text-align:left;direction:ltr"> =(a1*b1+a2*b2+…+a<h2 style=";text-align:left;direction:ltr"> i* <h2 style=";text-align:left;direction:ltr"> b<h2 style=";text-align:left;direction:ltr"> i <h2 style=";text-align:left;direction:ltr"> ) / (n*n) Where a k represents the number of real samples of the kth type of ground objects, b k Represents the predicted value of the sample of the kth category of land features, n represents the total number of samples, p represents the probability of correct classification among all samples, k = 1, 2, 3, ..., i, i is the total number of categories of land features.

Images