A method and system for extracting farmland plots in mountain and hilly areas

Abstract

The present application provides a kind of mountain hilly region cultivated land plot precision extraction method, comprising: obtaining the remote sensing image of the region to be extracted, and pre-processing remote sensing image;The rough cultivated land plot boundary graph is obtained by inputting the remote sensing image after pre-processing into cultivated land plot extraction model;Whether the rough cultivated land plot boundary graph is completely segmented based on the coupling strategy of decision rule is judged;Based on the optimization of the rough cultivated land plot boundary graph that is completely segmented by smooth contour line reconstruction rule, the optimized cultivated land plot boundary graph is used as the final cultivated land plot extraction result.The present application realizes the accurate demarcation of the fine cultivated land plot in complex mountain hilly region by constructing the cultivated land plot extraction model suitable for the characteristics of agricultural planting area, provides reference for large-scale land plot fine classification, meets the social needs such as land fine management and agricultural industrial structure adjustment dynamic monitoring.

Description

Technical Field

[0001] This invention belongs to the field of remote sensing image processing technology, and in particular relates to a method and system for extracting cultivated land plots in mountainous and hilly areas. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] The precision planting, visualization of management, and intelligent decision-making in modern agricultural production require not only large-scale real-time agricultural monitoring but also precise surveys of crop planting information at a fine scale. The rapid development of remote sensing technology, especially the widespread application of high-resolution imagery, has promoted the modernization of agriculture and supported the trend of agricultural monitoring towards regionalization, refinement, and quantification.

[0004] Currently, scholars both domestically and internationally are conducting research on precise mapping of cultivated land using high-resolution imagery, focusing on key technologies such as edge detection algorithms and region classification and recognition at the object, pixel, and sub-pixel scales. Among these efforts, accurately and efficiently extracting existing cultivated land plots in mountainous and hilly areas remains a long-standing challenge in the field of agricultural remote sensing.

[0005] Currently, most methods for extracting farmland plots rely on manual vectorization and high-resolution image segmentation. While manual vectorization offers high accuracy in boundary extraction, it is time-consuming and inefficient for large-scale farmland monitoring applications. High-resolution image segmentation methods, particularly edge-based methods, rely on various spatial filters such as Scharr, Sobel, and Canny operators. Conversely, region-based methods group adjacent pixels and classify them into objects based on predefined criteria. However, while these methods are time- and labor-saving, the optimal segmentation scale is influenced by land cover type, surrounding environmental contrast, and internal heterogeneity. They perform poorly in extracting farmland plots in mountainous and hilly areas, exhibiting low accuracy and a mismatch with geographic entities. Summary of the Invention

[0006] To overcome the shortcomings of the prior art, the present invention provides a method and system for the precise extraction of cultivated land plots in mountainous and hilly areas, thereby achieving the precise extraction of cultivated land plots in mountainous and hilly areas.

[0007] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions:

[0008] The first aspect of this invention provides a method for accurately extracting cultivated land plots in mountainous and hilly areas, comprising:

[0009] Acquire remote sensing images of the area to be extracted and preprocess the remote sensing images;

[0010] The preprocessed remote sensing images are input into the farmland plot extraction model to obtain a rough farmland plot boundary map;

[0011] A decision rule-based coupling strategy is used to determine whether a rough farmland plot boundary map is completely segmented.

[0012] The decision rule-based coupling strategy for determining whether a rough farmland parcel boundary map is completely segmented includes:

[0013] The random forest algorithm is used to classify cultivated land plots in preprocessed remote sensing images, resulting in a classification map of the random forest algorithm. Category attribute values ​​are then assigned to crop categories in the classification map.

[0014] The number of crop categories in each plot in the rough cultivated land plot boundary map is counted based on the category attribute values; if the number of crop categories in each plot in the rough cultivated land plot boundary map is only one, then the rough cultivated land plot boundary map is completely segmented.

[0015] The rough farmland plot boundary map that is completely segmented is optimized based on the smooth contour line reconstruction rule, and the optimized farmland plot boundary map is used as the final farmland plot extraction result.

[0016] A second aspect of the present invention provides a system for precise extraction of cultivated land plots in mountainous and hilly areas, comprising:

[0017] The remote sensing image acquisition module is configured to: acquire remote sensing images of the area to be extracted and preprocess the remote sensing images;

[0018] The land parcel segmentation module is configured to input the preprocessed remote sensing image into the cultivated land parcel extraction model to obtain a rough cultivated land parcel boundary map.

[0019] The segmentation result judgment module is configured to: use a coupling strategy based on decision rules to determine whether the rough farmland plot extraction map is completely segmented;

[0020] The decision rule-based coupling strategy for determining whether a rough farmland parcel boundary map is completely segmented includes:

[0021] The random forest algorithm is used to classify cultivated land plots in preprocessed remote sensing images, resulting in a classification map of the random forest algorithm. Category attribute values ​​are then assigned to crop categories in the classification map.

[0022] The number of crop categories in each plot in the rough cultivated land plot boundary map is counted based on the category attribute values; if the number of crop categories in each plot in the rough cultivated land plot boundary map is only one, then the rough cultivated land plot boundary map is completely segmented.

[0023] The land parcel boundary optimization module is configured to optimize the coarsely divided farmland parcel boundary map based on the smooth contour line reconstruction rule, and use the optimized farmland parcel boundary map as the final farmland parcel extraction result.

[0024] A third aspect of the present invention provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the steps of the method for accurately extracting cultivated land plots in mountainous and hilly areas as described in the first aspect of the present invention.

[0025] The fourth aspect of the present invention provides an electronic device, including a memory, a processor, and a program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the method for accurately extracting cultivated land plots in mountainous and hilly areas as described in the first aspect of the present invention.

[0026] The above one or more technical solutions have the following beneficial effects:

[0027] (1) This invention uses high-resolution satellite imagery as the data source and improves the R2U-Net model with recurrent (recursive) convolutional layers and residual units to obtain a model for delineating cultivated land plots in mountainous and hilly areas. This achieves high-precision delineation of farmland, greatly saving manpower and material resources. It can build an accurate agricultural detection system to implement precision agriculture and crop yield estimation, and maintain food security.

[0028] (2) This invention uses a residual convolution module to replace the traditional forward convolution layer, continuously extracts image features based on different time steps, and achieves feature accumulation, which helps to develop more effective deep models and ensures better and stronger feature representation. The introduction of residual units also effectively avoids the network degradation problem caused by increasing network depth.

[0029] (3) This invention can determine the boundaries of cultivated land plots in mountainous and hilly areas, provide real-time information on crop conditions for developing an accurate agricultural monitoring system, and achieve precise delineation of fragmented cultivated land plots in complex mountainous and hilly areas by constructing an algorithm model adapted to the characteristics of agricultural planting areas, providing a reference for the fine classification of large-scale plots, and meeting social needs such as refined land management and dynamic monitoring of agricultural industrial structure adjustment.

[0030] (4) The present invention develops the classification results of the random forest classification algorithm based on the decision rule coupling strategy to correct the farmland boundary map extracted by the R2U-Net model. Ideally, the target category value of each farmland plot should be 1. If it is greater than 1, the boundary line of the farmland plot is corrected based on the generalized adaptive buffer algorithm to obtain a more realistic plot boundary.

[0031] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0032] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0033] Figure 1 The flowchart shows a method for accurately extracting cultivated land plots in mountainous and hilly areas, as shown in the first embodiment.

[0034] Figure 2 The image data of the study area used in the training set in the first embodiment.

[0035] Figure 3 This is a structural diagram of the improved R2U-Net model in the first embodiment.

[0036] Figure 4 This is a flowchart illustrating the optimization of rough farmland plot boundary contours using a decision rule-based coupling strategy in the first embodiment.

[0037] Figure 5 The flowchart of the random forest classification algorithm in the first embodiment is shown.

[0038] Figure 6 (a) shows the study areas at different locations selected in the first embodiment.

[0039] Figure 6 (b) is Figure 6 (a) The farmland boundary map obtained by segmenting remote sensing images of regions S1, S2, S3 and S4 in the middle using an improved R2U-Net model.

[0040] Figure 7 The study areas are selected at different locations in the first embodiment.

[0041] Figure 8 for Figure 7 The classification boundary results of cultivated land plots obtained by applying the random forest classification algorithm to the remote sensing images of regions S1, S2 and S3 in the middle region.

[0042] Figure 9 for Figure 7 A comparison of the final farmland plot boundary map obtained after segmentation and optimization of remote sensing images of central regions S1, S2, and S3 using an improved R2U-Net model with the RF classification boundary results. Detailed Implementation

[0043] Example 1

[0044] This embodiment discloses a method for accurately extracting cultivated land plots in mountainous and hilly areas, including:

[0045] Step 1: Acquire remote sensing images of the area to be extracted and preprocess the remote sensing images;

[0046] Step 2: Input the preprocessed remote sensing image into the farmland plot extraction model to obtain a rough farmland plot boundary map;

[0047] Step 3: Determine whether the rough farmland plot extraction map is completely segmented based on the coupling strategy of decision rules. If the rough farmland plot boundary map is not completely segmented, correct the plot boundary lines in the rough farmland plot boundary map based on the adaptive generalized buffer algorithm.

[0048] Step 4: Optimize the rough farmland plot boundary map after complete segmentation or correction based on the smooth contour line reconstruction rule, optimize the boundary contour line, and use the optimized farmland plot boundary map as the final farmland plot boundary extraction result.

[0049] Step 5: The accuracy of the final farmland plot boundary extraction results is evaluated using region-based and edge-based methods, thereby analyzing the applicability of the farmland plot extraction model in mountainous and hilly areas.

[0050] Step 1 involves preprocessing the remote sensing images, including fusing the temporal and spectral features of the remote sensing images, creating an optimal feature dataset, and using the feature dataset as input to the model.

[0051] Step 2 includes: Step 201: Construction of the training dataset; Step 202: Construction of the recurrent residual convolutional neural network R2U-Net; Step 203: Training of the recurrent residual convolutional neural network R2U-Net;

[0052] Specifically, step 201: Collect remote sensing images of the study area, and obtain sample data of farmland plot boundaries in the study area through visual interpretation and manual vectorization, thereby obtaining the training dataset for the target area, including the following specific steps:

[0053] Step 2011: The research area in this invention is Yangming District, Mudanjiang City, Heilongjiang Province, such as... Figure 2 As shown, Sentinel-2 remote sensing images of this area were obtained.

[0054] Step 2012: Perform preprocessing operations such as radiometric correction, atmospheric correction, and geometric registration on the acquired remote sensing images.

[0055] Step 2013: Use ArcGIS to vectorize the cultivated land plots in the preprocessed remote sensing image to obtain artificially digitized farmland plot boundary vector data.

[0056] Step 2014: Perform mask extraction on the digitized farmland plot vector boundary data to obtain corresponding farmland plot sample data. Combine the remote sensing imagery of the study area with the farmland plot boundary map to form a training sample set.

[0057] Step 202: Construct a farmland plot extraction model based on a recurrent residual convolutional neural network (improved R2U-Net model).

[0058] Starting with the plot-scale approach to crop remote sensing identification, this invention employs a recurrent residual U-Net structure (recurrent residual convolutional neural network) originating from R2U-Net. In this invention, by studying U-Net models with basic forward convolutional layers, U-Net models with forward convolutional layers and residual units (i.e., residual U-Net (ResU-Net), and U-Net models with recurrent (recursive) convolutional layers and residual units (i.e., R2U-Net), a farmland plot extraction model adapted to the characteristics of agricultural planting areas is constructed. This enables accurate extraction of fragmented plots within complex mountainous and hilly terrain, providing a reference for the refined classification of large-scale plots.

[0059] like Figure 3 As shown, the recurrent residual convolutional neural network (RNN) comprises an encoder, a decoder, and a predictor. The encoder downsamples the input remote sensing image for each band, extracting features and generating feature maps at different scales. The encoder consists of multiple sequentially connected encoding layers, each containing a sequentially connected residual convolutional module, a ReLU nonlinear activation function (linear rectified layer), and a 3×3 max-pooling layer. The remote sensing image is input into the model, and features are extracted from the target image through recurrent convolutional layers (RCLs) with residual units, thereby repeatedly uncovering deeper features. To accelerate network training, the output is standardized. The decoder fuses the numerous low-dimensional feature maps at different scales generated by the encoder to reconstruct the segmented image (coarse parcel boundary map). The decoder consists of multiple sequentially connected decoding layers, each containing a sequentially connected upsampling layer, a residual recurrent deconvolutional block, a ReLU nonlinear activation function (linear rectified layer), and a full-scale skip layer. The predictor supervises the segmentation results. The predictor contains convolutional layers with a kernel size of 3*3. This predictor maps the final feature map to the same dimension as the number of categories in the remote sensing image and uses a flexible maximum transfer loss function to supervise the segmentation results.

[0060] The construction process of a recurrent residual convolutional neural network includes the following specific steps:

[0061] Step 2021: The R2U-Net model used in this invention is an improvement on the U-Net model with recurrent convolutional layers and residual units, while retaining the convolutional encoding and decoding units of the original U-Net model. Compared with the basic forward convolutional layers in U-Net, R2U-Net uses recurrent convolutional layers, which helps to develop more efficient and deeper models.

[0062] Step 2022: Use only concatenation operations in the R2U-Net model and remove the pruning and copying units used in the convolutional network (U-Net) model.

[0063] Step 2023: Compared to U-Net, this invention performs normalization processing in the R2U-Net model. Normalization processing is a technique that can accelerate the training of deep networks by normalizing the mean and variance, thus alleviating the problem of internal covariate shifting.

[0064] The specific operational procedures for standardization are as follows:

[0065] The formula for calculating the mean of each training batch of data is:

[0066]

[0067] The variance of each training batch of data is calculated using the following formula:

[0068]

[0069] Normalize the batch of training data using the obtained mean and variance to obtain a 0-1 distribution, as shown in the formula:

[0070]

[0071] The formulas for scaling and offset are as follows:

[0072]

[0073] Where B represents a batch of input data, μ B and Let B be the mean and variance, ε be a small positive number used to avoid division by zero, γ and β be the scaling and shifting factors, respectively, and β be trainable parameters.

[0074] Standardization can significantly improve model training speed. Without standardization, the learning rate needs to be adjusted gradually. With standardization, a large initial learning rate can be used, followed by a rapid decay of the learning rate, resulting in faster convergence and significantly reducing the number of iterations required to reach convergence, thus improving final performance. Standardization effectively uses only the linear part of the sigmoid activation function, mitigating the gradient vanishing problem during backpropagation.

[0075] Step 2024: Compared to the U-Net model with forward convolutional layers and residual units, a recurrent convolutional layer is added to the model. Within this added recurrent convolutional layer, an efficient feature accumulation method with different time steps (t) is employed to continuously extract image features. The time steps t are 1, 2, and 3. The input image first passes through two recurrent convolutional layers with 3×3 kernels. Figure 3 This describes the unfolding process of the circular convolution operation at a time step of t=3. At t=0, the image is input into the circular convolution layer. At t=1, the image is subjected to the first 3×3 forward convolution operation. At t=2, the input image and the output of the first forward convolution operation are used as the input for the second forward convolution operation. This process continues until the output at t=3 is the output of the circular convolution layer.

[0076] The low-level feature information obtained after the operation of the recurrent convolution module is fused with the input image to achieve residual connectivity. The feature accumulation of the recurrent convolution layer based on different time steps ensures better and stronger feature representation, which helps to extract very low-level features. These features provide more detailed features for remote sensing image segmentation tasks with complex features, and improve the accuracy of segmentation.

[0077] Step 2025: Since convolutional layers with increased stride outperform max pooling in multiple image recognition benchmarks, the max pooling layers are replaced with 3×3 convolutional layers with a stride of 2 in the R2U-Net model.

[0078] Step 203: Input training samples, learn representative and discriminative features based on the model, and train the farmland plot extraction model.

[0079] Based on the training dataset obtained after artificial vectorization, the training samples are input into the constructed recurrent residual convolutional neural network model to learn representative and discriminative features, optimize various parameters, and train the farmland parcel boundary delineation model, including the following specific steps:

[0080] Step 2031: Oversample the samples in the training sample set that contain farmland plots, and undersample the samples containing only background with the same number of samples as the oversampled samples.

[0081] Step 2032: Input the oversampled and undersampled training sample sets into the recurrent residual convolutional neural network model, and continuously adjust and optimize the model parameters using a step-by-step training method. Utilize the stochastic gradient descent algorithm to obtain the trained recurrent residual convolutional neural network model.

[0082] Step 2033: Apply the trained recurrent residual convolutional neural network model to process the Sentinel-2 data of the study area to obtain preliminary rough farmland plot extraction results.

[0083] like Figure 4 As shown, in step 3, for the under-segmentation problem of extracting fragmented land parcels in mountainous and hilly areas, this invention develops a coupling strategy based on decision rules to handle it. The specific operation steps are as follows:

[0084] Step 301: Assign a unique attribute value "m" to each individual plot of land based on the output of the improved R2U-Net model. i ,i=1,2,3…”;

[0085] Step 302: Classify farmland plots in remote sensing images based on the Random Forest (RF) algorithm to obtain the classification result map of the Random Forest algorithm.

[0086] like Figure 5 As shown, the RF classification model is built using optimal features, and the required number of classification trees (n) tree The number of predictors (m) is set to 1000. try The value is equal to the square root of the number of input variables. In this invention, 450 training samples are used to classify five crop categories (corn, soybean, rice, wheat, and other crops), and 250 validation samples based on a confusion matrix are used to evaluate the accuracy, resulting in a trained RF classification model.

[0087] To improve the differentiation between different crop types, "n" (n = 1, 2, 3...I, where I is the attribute of the crop category) was used to assign each crop plot. Based on the actual situation that wheat, soybeans, and corn are mainly grown in the study area, soybeans were assigned a value of n = 1, corn a value of n = 2, wheat a value of n = 3, and rice a value of n = 4.

[0088] Step 303: Calculate m for a single plot of land i Number of crop categories C i

[0089] Based on the classification results map of the RF classification model and the crop category n within the plot, the number of crop category n in each cultivated land plot in the cultivated land plot boundary drawing results based on the improved R2U-Net model is counted, and set as C. i ;

[0090] For example, Figure 4 In the given plot, the crop categories n in plot m1 are: n = 1, 2, 1, 2; the crop categories n in plot m2 are: n = 1; the crop categories n in plot m3 are: n = 2; therefore, the number of crop categories in plots m1, m2, and m3 is C1 = 4, C2 = 1, and C3 = 1.

[0091] Step 304: Develop a coupling strategy based on decision rules. Ideally, the target category value C for each farmland plot... i If the value is 1, then the final result of farmland demarcation is based on the improved R2U-Net model; if C i >1, which means that the improved R2U-Net model incorrectly depicts the plots, which contain more than one crop, and the RF classification result is used as the final boundary.

[0092]

[0093] Where Bplot is the final farmland boundary, B R2U-Net and B RF These are the boundary delineation results based on the improved R2U-Net model and the RF classification model, respectively. i It represents the number of crop types within the i-th plot of cultivated land.

[0094] For the problem of incomplete segmentation in the extraction of fragmented land parcels in mountainous and hilly areas, an adaptive generalized buffer algorithm is required. The specific operation steps are as follows:

[0095] Step 305: For the boundary of each plot obtained based on the improved R2U-Net model, calculate the average distance between it and the boundary of the plot obtained from the RF classification results to obtain a threshold. Then, automatically generate a buffer based on the actual situation of each plot. Its mathematical model can be expressed as:

[0096]

[0097] Where r CL It is the generalized buffer radius, which is subject to similar rules; Let r be the set of instances, where the distance between all instances is less than or equal to r. CL ;(p,A j ) represent the farmland plot boundary polygon objects obtained from the classified and corrected R2U-Net, respectively; d(p, A j ) is p and A j The distance between them. Generalized buffer radius r CL It is a function of instance properties and can be evaluated using the Class Relationship Rule (CLR), that is,

[0098] rCL =d[(x, y), CLR]

[0099] Where d[(x, y), CLR] is the distance function of the coordinates (x, y) of the boundary point obtained using the homogeneity rule.

[0100] Step 306: Within a plot, if the buffer has exceeded 65% of the boundary depicted by the RF classification results, it means there is a small gap between the segmentation and classification. Therefore, the segmentation result obtained by the improved R2U-Net model is used as the final boundary output; otherwise, continue to extend the buffer to the RF classification boundary until the buffer contains more than 65% of the RF classification results.

[0101] like Figure 6 (a) shows the research areas at different locations selected in this embodiment; Figure 6 (b) is Figure 6 (a) The farmland boundary maps obtained by segmenting remote sensing images of regions S1, S2, S3, and S4 using an improved R2U-Net model, and the judgment of the segmentation results of farmland boundary maps of different regions based on a coupling strategy of decision rules, show that some plots are not completely segmented; such as Figure 7 The image shows the region to be extracted in this embodiment, based on the random forest algorithm. Figure 7 The classification results of remote sensing images of different regions in China are as follows: Figure 8 As shown.

[0102] Image segmentation refers to dividing an image into multiple independent parts, each with a specific meaning, such as separating objects, background, and masks. Image segmentation typically involves preprocessing the image to allow for further processing of its information. The result of image segmentation is usually a binary image, where object parts are labeled as 1 and background parts are labeled as 0.

[0103] Image classification refers to dividing images into multiple categories, each corresponding to a specific object or scene, such as animals, plants, or buildings. Image classification is usually the final processing step in image analysis to extract information from the image and make related decisions or actions. The result of image classification is typically a label indicating which category the image belongs to.

[0104] This invention develops a decision rule-based coupling strategy to correct image segmentation results based on an improved R2U-Net based on RF classification results, thereby obtaining more realistic land parcel boundaries and improving the accuracy of land parcel boundary extraction.

[0105] In step 4, considering that the contour detection result based on R2U-Net is a fragmented contour, the smoothness and continuity of the edge contour lines are not guaranteed. To solve this problem, morphological filtering and the Douglas-Peucker algorithm (DPA) are needed to generate efficient and stable smooth contour line reconstruction rules, including the following specific steps:

[0106] Step 401: Connect the start and end points of the curve with a straight line, and then draw the distances from the remaining points to this straight line.

[0107] Step 402: If the maximum value is greater than the threshold, the point with the largest distance from the line is retained; otherwise, all points between the two endpoints of the line are eliminated.

[0108] The threshold is used to control the accuracy of the simplified curve. By adjusting the size of the threshold, the degree to which the straight line result approximates the original curve can be controlled: the larger the threshold, the greater the degree of simplification and the more points are reduced; conversely, the lower the degree of simplification, the more points are retained, and the shape is closer to the original curve.

[0109] Step 403: Based on the retained points, the known curve is divided into two parts and iteratively processed. That is, the remaining maximum distance is selected and compared with the threshold, and discarded in turn until there are no points that need to be excluded. Finally, the coordinates of the curve points that satisfy the given precision limit difference are obtained, and the smooth boundary is obtained.

[0110] This embodiment processes remote sensing images of different regions separately, and the optimization results of the edge contours are as follows: Figure 9 As shown.

[0111] Step 5 involves evaluating the accuracy of the generated farmland plot results. This invention employs region-based and edge-based methods to evaluate the accuracy of the farmland plot extraction results, thereby analyzing the applicability of the farmland plot extraction model for mountainous and hilly areas. Specifically, it includes the following steps:

[0112] Step 501: The accuracy of the final farmland plot extraction results is evaluated using a region-based method. The region-based metric is calculated using a confusion matrix, which consists of 10,000 samples. 5,000 "boundary" samples are randomly selected along the reference boundary, and the other 5,000 "non-boundary" samples are randomly selected at a distance of more than 10 meters from the reference boundary.

[0113] Accuracy metrics include Overall Accuracy (OA), Kappa coefficient (K), Commission Error (CE), Omission Error (OE), and Precision Recall (PR). CE represents false boundaries of the field, often caused by over-segmentation. OE provides an accuracy metric for extracting boundaries along a reference field boundary, where high error indicates that segmentation does not accurately describe the field boundary. PR is a framework for calculating precision (P), recall (R), and F1 score (or F metric). For farmland boundary maps, the PR metric is calculated based on four terms: true positives, true negatives, false positives, and false negatives. P is the number of true positive predicted pixels divided by the total number of positive pixels returned by the network, measuring how close the detected boundary is to the reference boundary; R is the number of true positive predicted pixels divided by the total number of pixels that should be identified as positive, representing the proportion of correctly detected reference boundaries; the F1 score (F) is the harmonic mean of P and R, ranging from 0 to 1, tending towards 1 for optimal performance, and is calculated as follows:

[0114]

[0115] Step 502: The accuracy of the final farmland plot boundary extraction results is evaluated using an edge-based method. The accuracy of the edge-based method is assessed by four indicators (oversegmentation rate, undersegmentation rate, location offset, and eccentricity rate), which reflect the shape, size, and offset of the extracted area relative to the target or reference area, respectively.

[0116] First, the overlapping area between the reference field and the delineated field is extracted, and then four indices are calculated: (1) the relative area of ​​the overlapping area and the reference field (oversegmentation rate, RAor); (2) the relative area of ​​the overlapping area and the extracted field (undersegmentation rate, RAos); (3) the positional difference between the extracted field (s) and the reference field (positional offset, Dsr), which is calculated as the average distance between the center point of the extracted field and the center point of the reference field; (4) the absolute difference in shape (eccentricity coefficient, ε).

[0117] The calculation formula is as follows:

[0118]

[0119]

[0120]

[0121] ε=||Eccentricity or -Eccentricity os ||;

[0122] Where n represents the number of fields generated by the R2U-Net model, A o(i) is the area of ​​the i-th overlapping region related to the reference field, A r D is the area of ​​the reference field, As(i) is the area of ​​the i-th extracted field, and D sr It is the average distance between the center point of the extracted field and the center point of the reference field, representing the positional accuracy of the delineated field; X s (i) and Y s (i) are the coordinates of the center point of the i-th extraction field, X r and Y r These are the coordinates of the center point of the reference field; RA or and RA os Used to evaluate the topological (partial) similarity between the extracted region and the reference region, providing a ratio value from 0 to 100: the closer the value is to 100, the less likely oversegmentation and undersegmentation will occur. RA or RA os and D sr The average value represents the overall segmentation quality; the eccentricity indicates the degree to which the shape of a region deviates from a circle (eccentricity = 0). Table 1 shows the measurement results based on regions and edges in the study area.

[0123] Table 1

[0124]

[0125] The extraction results demonstrate high accuracy in both region-based and edge-based metrics. The OA value reaches 78.21%, and the Kappa value reaches 0.74, indicating a high level of accuracy in the final classification result. The CE value is 4.23, indicating that there is almost no oversegmentation during the classification process and few erroneous boundaries. The RAor and RAos values ​​are 76.74% and 78.98%, respectively, indicating that regions with incomplete segmentation problems are well corrected. The final extracted field shows little difference in shape, size, and positional bias relative to the reference field, achieving significant results.

[0126] Example 2

[0127] This embodiment discloses a system for accurately extracting cultivated land plots in mountainous and hilly areas, including:

[0128] The remote sensing image acquisition module is configured to: acquire remote sensing images of the area to be extracted and preprocess the remote sensing images;

[0129] The land parcel segmentation module is configured to input the preprocessed remote sensing image into the cultivated land parcel extraction model to obtain a rough cultivated land parcel boundary map.

[0130] The segmentation result judgment module is configured to: use a coupling strategy based on decision rules to determine whether the rough farmland plot extraction map is completely segmented;

[0131] The decision rule-based coupling strategy for determining whether a rough farmland parcel boundary map is completely segmented includes:

[0132] The random forest algorithm is used to classify cultivated land plots in preprocessed remote sensing images, resulting in a classification map of the random forest algorithm. Category attribute values ​​are then assigned to crop categories in the classification map.

[0133] The number of crop categories in each plot in the rough cultivated land plot boundary map is counted based on the category attribute values; if the number of crop categories in each plot in the rough cultivated land plot boundary map is only one, then the rough cultivated land plot boundary map is completely segmented.

[0134] The land parcel boundary optimization module is configured to optimize the coarsely divided farmland parcel boundary map based on the smooth contour line reconstruction rule, and use the optimized farmland parcel boundary map as the final farmland parcel extraction result.

[0135] Example 3

[0136] The purpose of this embodiment is to provide a computer-readable storage medium.

[0137] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a method for accurately extracting cultivated land plots in mountainous and hilly areas as described in Embodiment 1 of this disclosure.

[0138] Example 4

[0139] The purpose of this embodiment is to provide an electronic device.

[0140] An electronic device includes a memory, a processor, and a program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps in a method for accurately extracting cultivated land plots in mountainous and hilly areas as described in Embodiment 1 of this disclosure.

[0141] The steps and methods involved in the apparatuses of Embodiments 2, 3, and 4 above correspond to those in Embodiment 1. For specific implementation details, please refer to the relevant description section of Embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood as including any medium capable of storing, encoding, or carrying an instruction set for execution by a processor and enabling the processor to perform any of the methods in this invention.

[0142] Those skilled in the art will understand that the modules or steps of the present invention described above can be implemented using general-purpose computer devices. Optionally, they can be implemented using computer-executable program code, thereby allowing them to be stored in a storage device for execution by a computer device, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. The present invention is not limited to any particular combination of hardware and software.

[0143] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A method for precisely extracting a farmland plot in a mountain hilly area, characterized in that, The method comprises the following steps: acquiring remote sensing images of an area to be extracted, and preprocessing the remote sensing images; inputting the preprocessed remote sensing images into a cultivated land plot extraction model to obtain a rough cultivated land plot boundary map; the cultivated land plot extraction model is obtained by training a residual recurrent convolutional neural network (R2U-Net) using the processed remote sensing images; the residual recurrent convolutional neural network (R2U-Net) comprises an encoder and a decoder; the encoder is configured to perform down-sampling operation on the input remote sensing images, extract features of the remote sensing images, and generate feature maps of different scales; the decoder is configured to perform feature fusion on the feature maps of different scales generated by the encoder to obtain a rough cultivated land plot extraction map; determining whether the rough cultivated land plot boundary map is completely segmented based on a coupling strategy of a decision rule; the step of determining whether the rough cultivated land plot boundary map is completely segmented based on the coupling strategy of the decision rule comprises: classifying cultivated land plots in the preprocessed remote sensing images based on a random forest algorithm to obtain a classification result map of the random forest algorithm, and assigning a category attribute value to a crop category in the classification result map; counting the number of crop categories in each plot in the rough cultivated land plot boundary map according to the category attribute value; if the number of crop categories in each plot in the rough cultivated land plot boundary map is one, the rough cultivated land plot boundary map is completely segmented; if the rough cultivated land plot boundary map is not completely segmented, correcting plot boundary lines in the rough cultivated land plot boundary map based on an adaptive generalized buffer algorithm; establishing a smooth contour line reconstruction rule according to a morphological filter and a Douglas-Peucker algorithm; optimizing the completely segmented rough cultivated land plot boundary map based on the smooth contour line reconstruction rule, and taking the optimized cultivated land plot boundary map as a final cultivated land plot extraction result; the step of optimizing the completely segmented rough cultivated land plot boundary map based on the smooth contour line reconstruction rule comprises: connecting a straight line at the starting point and the ending point of a curve, then drawing the distance from the remaining points on the curve to the straight line, if the maximum value is greater than a threshold value, retaining the point with the maximum distance from the straight line, otherwise all points between the two endpoints of the straight line will be eliminated; dividing the known curve into two parts according to the retained points, and performing iterative processing: selecting the maximum distance remaining and comparing it with the threshold value, and discarding in turn until there is no point to be excluded, finally obtaining the curve point coordinates that meet the given accuracy limit difference, and obtaining the optimized smooth plot boundary.

2. The method for accurately extracting cultivated land plots in mountainous and hilly areas as described in claim 1, characterized in that, the step of correcting the plot boundary lines in the rough cultivated land plot boundary map based on the adaptive generalized buffer algorithm comprises: calculating the average distance between the plot boundaries in the rough cultivated land plot boundary map and the classification result boundary obtained based on the random forest algorithm to obtain a threshold value, and automatically generating a buffer zone; determining whether the buffer zone exceeds 65% of the classification result boundary; if yes, the plot boundaries in the rough cultivated land plot boundary map are taken as the final plot boundaries; otherwise, continue to expand the buffer zone to the classification result boundary until the buffer zone contains more than 65% of the classification result boundary, link the plot boundaries to the classification result boundary, and obtain the corrected plot boundaries.

3. The method for accurately extracting cultivated land plots in mountainous and hilly areas as described in claim 1, characterized in that, The encoder comprises a plurality of encoding layers connected in sequence, each encoding layer comprising a residual cyclic convolution module, a Relu nonlinear activation function and a 3*3 maximum pooling layer connected in sequence; The decoder comprises a plurality of decoding layers connected in sequence, each decoding layer comprising an up-sampling layer, a residual cyclic de-convolution block, a Relu nonlinear activation function and a full-scale skip layer connected in sequence.

4. The method for accurately extracting cultivated land plots in mountainous and hilly areas as described in claim 3, characterized in that, The residual cyclic convolution module continuously extracts image features based on an efficient feature accumulation method at different time steps, comprising: Let the time step t be 1, 2 and 3 respectively; The input image first passes through a cyclic convolution layer with two convolution kernels of 3*3, at t=0, the image is input to the cyclic convolution layer, at t=1, the first 3*3 forward convolution operation is performed on the image, at t=2, the input image and the output of the first forward convolution operation are taken as a whole as the input of the second forward convolution operation, and so on, the output at t=3 is the output of the residual cyclic convolution module.

5. A system for precise extraction of cultivated land plots in mountain hilly areas, characterized in that it comprises: Comprise: A remote sensing image acquisition module configured to acquire a remote sensing image of a region to be extracted, and pre-process the remote sensing image; A land parcel segmentation module configured to input the pre-processed remote sensing image into a cultivated land parcel extraction model to obtain a rough cultivated land parcel boundary map; the cultivated land parcel extraction model is obtained by training a recurrent residual convolutional neural network R2U-Net using the processed remote sensing image; the recurrent residual convolutional neural network R2U-Net comprises an encoder and a decoder, the encoder is used to perform down-sampling operation on the input remote sensing image, extract features of the remote sensing image, and generate feature maps of different scales; the decoder is used to fuse the feature maps of different scales generated by the encoder to obtain a rough cultivated land parcel extraction map; A segmentation result judgment module configured to judge whether the rough cultivated land parcel extraction map is completely segmented based on a coupling strategy of decision rules; The judgment whether the rough cultivated land parcel boundary map is completely segmented based on the coupling strategy of decision rules comprises: Classifying the cultivated land parcels in the pre-processed remote sensing image based on a random forest algorithm to obtain a classification result map of the random forest algorithm, and assigning a class attribute value to the crop class in the classification result map; According to the class attribute value, the number of crop classes in each land parcel in the rough cultivated land parcel boundary map is counted; if the number of crop classes in each land parcel in the rough cultivated land parcel boundary map is one, the rough cultivated land parcel boundary map is completely segmented; if the rough cultivated land parcel boundary map is not completely segmented, the land parcel boundary lines in the rough cultivated land parcel boundary map are corrected based on an adaptive generalized buffer algorithm; A land parcel boundary optimization module configured to establish a smooth contour line reconstruction rule according to a morphological filter and a Douglas-Peucker algorithm; based on the smooth contour line reconstruction rule, the rough cultivated land parcel boundary map that is completely segmented is optimized, and the optimized cultivated land parcel boundary map is taken as the final cultivated land parcel extraction result; the optimization of the rough cultivated land parcel boundary map that is completely segmented based on the smooth contour line reconstruction rule comprises: Connecting a straight line at the starting point and the ending point of the curve, then drawing the distance from the remaining points on the curve to the straight line, if the maximum value is greater than the threshold value, the point with the maximum distance to the straight line is retained, otherwise all points between the two endpoints of the straight line will be eliminated; According to the retained points, the known curve is divided into two parts, and iterative processing is performed: the maximum distance remaining is compared with the threshold value, and is discarded in turn until there is no point to be excluded, finally the curve point coordinates satisfying the given precision limit difference are obtained, and the optimized smooth land parcel boundary is obtained.

6. A computer-readable storage medium having stored thereon a program, characterized in that, The program is executed by the processor to implement the steps in the mountain and hilly area cultivated land parcel precision extraction method according to any one of claims 1-4.

7. An electronic device comprising a memory, a processor, and a program stored in the memory and capable of running on the processor, characterized by The processor executes the program to implement the steps in the mountain and hilly area cultivated land parcel precision extraction method according to any one of claims 1-4.

Images