A method and product for dynamically updating a field plot

Abstract

The application discloses a farmland plot dynamic updating method and product, relates to the farmland right field, and comprises the following steps: acquiring a remote sensing image data pair of a rural farmland history and corresponding farmland boundary data, and constructing a data set; according to the data set, constructing a super-resolution-plot joint model based on a super-resolution network and a plot segmentation network; determining current super-resolution data and current plot data of the rural farmland according to current medium-resolution remote sensing image of the rural farmland and based on the trained super-resolution-plot joint model; further determining current plot vector data of the rural farmland; and associating and matching the current plot vector data with a right ownership database, and righting the rural farmland. The application can improve land righting accuracy and efficiency, and dynamically updates the farmland plot regularly.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of cultivated land right confirmation, and in particular to a cultivated land plot dynamic updating method and product. BACKGROUND

[0002] In recent years, with the rapid development of remote sensing technology, deep learning and geographic information system (GIS), automatic technology has been gradually introduced into the field of agricultural informatization and land management to improve efficiency. In the field of cultivated land right confirmation, existing technologies mainly include medium and low resolution satellite remote sensing image technology, super-resolution reconstruction technology and instance segmentation model technology. Medium and low resolution satellite remote sensing image technology (such as Landsat, Sentinel-2) has a wide coverage and low cost, and is widely used in large-scale cultivated land monitoring. However, its resolution (10-30 meters) is difficult to accurately extract the field boundary, especially in the complex scene of cross field ridge and crop shading, the boundary blur problem is particularly prominent. Although unmanned aerial vehicles can obtain high-resolution images (0.1-2 meters), they are limited by flight costs, data processing period and dependence on manual interpretation, making it difficult to achieve large-scale application. Traditional image segmentation methods (such as threshold segmentation and edge detection algorithm) have high false detection rate in complex farmland scenes, and are difficult to meet the demand of high-precision right confirmation. Super-resolution reconstruction technology (such as ESRGAN, SRCNN) focuses on general scenes in existing research, lacks optimization for farmland texture features, resulting in reconstructed images still having artifacts or missing details. Instance segmentation model technology (such as Mask R-CNN, U-Net) has been preliminarily applied in farmland segmentation, but its performance significantly decreases under low resolution images, and it is not deeply combined with land ownership data, so it cannot directly serve the right confirmation process.

[0003] Currently, the right confirmation work is still dominated by manual mode, relying on field measurement, paper archive management and manual checking, which is low in efficiency, high in cost and easy to be affected by subjective factors. Some semi-automatic systems try to combine GIS and remote sensing technology, but the data updating cycle is long (usually 1-2 years), which cannot realize dynamic management, and lacks a reliable evidence mechanism, making it difficult to meet the needs of land transfer and dispute arbitration. Although some progress has been made, rural cultivated land right confirmation still faces some core bottlenecks: first, the contradiction between image resolution and boundary precision. Low-cost satellite image resolution is insufficient, resulting in more than 10% error rate in extracting field boundary. High-resolution unmanned aerial image is high in cost and difficult to update frequently, which cannot meet the economic needs of large-scale right confirmation. Second, the algorithm adaptability is insufficient. The general super-resolution model is not optimized for farmland edges (such as ridges and ditches), and the reconstructed image still has artifacts or blurred details. The segmentation model is easily disturbed by crop coverage and shadow in complex farmland scenes, resulting in high omission rate and false detection rate, especially in the crop growth period. Third, the right confirmation data updating is lagging. The existing system relies on periodic manual survey, and the data updating cycle is as long as 1-2 years, which cannot reflect the cultivated land changes (such as encroachment, abandonment and land consolidation) in time, leading to the disconnection between right confirmation information and actual situation, and increasing the risk of disputes. Fourth, the data credibility and traceability problem. The right confirmation data are stored in centralized database, which has the risk of tampering, and lacks tamper-proof change records, making it difficult to meet the requirements of data authenticity and traceability in land transfer and dispute arbitration.

[0004] Therefore, based on the above problems, it is necessary to provide a cultivated land plot dynamic updating method, which can improve the accuracy and efficiency of land right confirmation, and dynamically update the plots of rural cultivated land regularly. SUMMARY

[0005] The purpose of the present application is to provide a cultivated land plot dynamic updating method and product, which can improve the accuracy and efficiency of land right confirmation, and dynamically update the plots of rural cultivated land regularly.

[0006] To achieve the above purpose, the present application provides the following solutions:

[0007] In a first aspect, the present application provides a cultivated land plot dynamic updating method, which comprises:

[0008] acquiring a pair of historical remote sensing image data and corresponding farmland boundary data of rural cultivated land, and constructing a data set; the pair of remote sensing image data comprises: medium-resolution remote sensing image and high-resolution remote sensing image;

[0009] According to the data set, an ultra-resolution field block joint model is constructed based on an ultra-resolution network and a field block segmentation network; the ultra-resolution field block joint model is used to output corresponding ultra-resolution data and field block data of rural cultivated land; the ultra-resolution field block joint model comprises an encoder, a decoder, an output head and a loss function; the encoder comprises a plurality of convolutional blocks in a positive pyramid type in cascade and is used to extract image features of a remote sensing image data pair; the decoder comprises a plurality of convolutional blocks in an inverse pyramid type in cascade and is used to decode the image features; the convolutional blocks of the encoder and the convolutional blocks of the decoder are symmetrically arranged and have the same number; the loss function comprises a mean absolute error loss, an adversarial loss, an edge enhancement loss, a cross-entropy loss and a loss.

[0010] According to the current medium-resolution remote sensing image of rural cultivated land, the current ultra-resolution data and the current field block data of rural cultivated land are determined based on the trained ultra-resolution field block joint model.

[0011] According to the current ultra-resolution data and the current field block data, current field block vector data of rural cultivated land is determined.

[0012] The current field block vector data is associated and matched with the ownership database to determine the ownership of rural cultivated land.

[0013] Optionally, the historical remote sensing image data pair and the corresponding farmland boundary data of rural cultivated land are obtained, and a data set is constructed, specifically comprising:

[0014] The historical remote sensing image data pair is preprocessed to determine a farmland region; the preprocessing process comprises radiation correction, geometric registration and farmland region extraction; the farmland region extraction process comprises extracting a region of interest of the farmland region by using the cultivated land classification in the land classification data set, and cropping the non-farmland region;

[0015] The farmland region is spatially aligned with the ownership database to obtain an aligned remote sensing image data pair;

[0016] The aligned remote sensing image data pair and the corresponding farmland boundary data are used to construct a data set.

[0017] Optionally, the determination process of the loss function comprises:

[0018] The ultra-resolution loss is determined by using the formula

[0019] The segmentation loss is determined by using the formula

[0020] The loss function is determined by using the formula ​​​​​ ;

[0021] wherein, is a mean absolute error loss coefficient, is a mean absolute error loss, is an adversarial loss coefficient, is an adversarial loss, is an edge enhancement loss coefficient, is an edge enhancement loss, is a cross-entropy loss coefficient, is a cross-entropy loss, is a loss coefficient, is a loss.

[0022] Optionally, the determining of the current field plot vector data of the rural cultivated land according to the current super segmentation data and the current field plot data specifically comprises:

[0023] aligning and correcting the current super segmentation data and the current field plot data to obtain aligned field plot data;

[0024] vectorizing the aligned field plot data to obtain the current field plot vector data of the rural cultivated land.

[0025] Optionally, the associating and matching of the current field plot vector data and the right ownership database to right the rural cultivated land specifically comprises:

[0026] overlapping the longitude and latitude coordinates of the current field plot vector data and the right ownership database; and determining an area error of the current field plot vector data and the right ownership database;

[0027] taking a matching rule with a longitude and latitude coordinate overlap degree greater than 90% and the area error less than 5% as an optimal matching rule, and obtaining a matching result;

[0028] righting the rural cultivated land by using the matching result.

[0029] Optionally, the associating and matching of the current field plot vector data and the right ownership database to right the rural cultivated land further comprises:

[0030] judging whether an abnormal situation exists according to the righting result of the rural cultivated land; if yes, issuing an alarm to prompt manual intervention; if no, storing the righting result of the rural cultivated land; the abnormal situation includes a field plot area change degree higher than an area change threshold, a field plot right ownership matching conflict, and newly added unregistered cultivated land.

[0031] In a second aspect, the present application provides a computer program product comprising a computer program which, when executed by a processor, implements the steps of the farmland plot dynamic updating method of any one of the above.

[0032] According to the specific embodiments provided in the present application, the present application has the following technical effects:

[0033] The present application provides a farmland plot dynamic updating method and product. By fusing a super-resolution network and a field plot segmentation network, a super-field joint model is constructed for unified optimization training, and the loss function is adaptively improved, so that it is more suitable for farmland scenes, the problem of boundary blur caused by low-resolution images is solved, the farmland edge information is enhanced, and the extraction accuracy of field plot data is improved. The field plot vector data and the right ownership database are associated and matched to generate field plot vector data with spatial position, area and right ownership information, further improving the right ownership accuracy and efficiency. When the right ownership of rural farmland is determined, the current medium-resolution remote sensing image of rural farmland is input each time, i.e. the right ownership of the current rural farmland is determined periodically, achieving the effect of dynamic updating of rural farmland. BRIEF DESCRIPTION OF DRAWINGS

[0034] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0035] Figure 1 FIG. 1 is a flowchart of a farmland plot dynamic updating method according to an embodiment of the present application;

[0036] Figure 2 FIG. 2 is a specific flowchart of the farmland plot dynamic updating method according to an embodiment of the present application;

[0037] Figure 3 FIG. 3 is a schematic diagram of historical remote sensing image data pairs and corresponding farmland boundary data according to an embodiment of the present application; Figure 3 (a) part of FIG. 3 is a schematic diagram of a medium-resolution remote sensing image, Figure 3 (b) part of FIG. 3 is a schematic diagram of a high-resolution remote sensing image, Figure 3 (c) part of FIG. 3 is a schematic diagram of farmland boundary data;

[0038] Figure 4 FIG. 4 is a schematic diagram of a super-field joint model according to an embodiment of the present application;

[0039] Figure 5 FIG. 5 is a schematic diagram of a current medium-resolution remote sensing image, super-resolution data and field plot data according to an embodiment of the present application;Figure 5 Part (a) is a schematic diagram of the current medium-resolution remote sensing image. Figure 5 Part (b) is a schematic diagram of the current super-resolution data. Figure 5 (Part (c) is a schematic diagram of the current field data).

[0040] Figure 6 This is a schematic diagram of the results of rural farmland ownership confirmation in one embodiment of this application. Detailed Implementation

[0041] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0042] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0043] In one exemplary embodiment, such as Figure 1 and Figure 2 As shown, a method for dynamic updating of arable land plots is provided, which includes the following steps S1 to S5. Wherein:

[0044] S1: Obtain remote sensing image data pairs of rural cultivated land history and corresponding farmland boundary data, and construct a dataset.

[0045] S1 specifically includes:

[0046] S11: Obtain remote sensing image data pairs of rural cultivated land history and corresponding farmland boundary data.

[0047] like Figure 3 As shown, the remote sensing image data pair includes medium-resolution and high-resolution remote sensing images. The medium-resolution images are from Sentinel-2 (S2) satellites, provided for download by the European Space Agency (ESA), while the high-resolution images are from Gaofen-2 (GF-2) satellites, provided for download by the Land Observation Satellite Data Service Platform. The temporal error of the remote sensing image data pair is less than two days, and the spatial error is less than 10 meters.

[0048] The farmland boundary data is from the open-source Field Boundaries for Agriculture (Fiboa).

[0049] S12: Preprocess historical remote sensing image data to identify farmland areas.

[0050] Historical remote sensing image data pairs were sequentially radiometrically corrected and geometrically registered. Simultaneously, the remote sensing image data pairs were reprojected onto the geographic coordinate system to ensure consistency between the historical remote sensing image data pairs and the Geographic Information System (GIS) coordinate system. A land classification dataset (European Space Agency World Cover, ESA World Cover) was downloaded using Google Earth Engine (GEE). Regions of Interest (ROIs) were extracted from farmland areas based on the farmland classification in the land classification dataset. Non-farmland areas, including areas such as houses and roads, were then cropped using the remote sensing image processing environment software ENvironment for Visualizing Images (ENVI).

[0051] S13: Align the farmland area and the ownership database with spatial coordinates to obtain aligned remote sensing image data pairs.

[0052] The coordinate system of the farmland area is projected onto the coordinate system of the ownership database. Then, control points are added using ENVI software to spatially register the two, so that their coordinates are completely aligned, resulting in an aligned pair of remote sensing image data.

[0053] S14: Construct a dataset based on the aligned remote sensing image data pairs and the corresponding farmland boundary data.

[0054] S2: Based on the dataset, construct a super-resolution-field joint model using a super-resolution network and a field segmentation network.

[0055] This application integrates traditional Super-Resolution Convolutional Neural Networks (SRCNN) and Field Parcel Segmentation Networks (FPSN) into a novel super-resolution-field parcel joint model. The overall super-resolution-field parcel joint model adopts a U-net network structure, consisting of an encoder, decoder, output head, and loss function. The encoder comprises multiple cascaded convolutional blocks in a pyramidal shape, which compress image features into a high-level semantic feature space through several convolutional downsampling operations; that is, the encoder is used to extract image features from remote sensing image data pairs. The decoder comprises multiple cascaded convolutional blocks in an inverted pyramidal shape, with the encoder's convolutional blocks and decoder's convolutional blocks symmetrically arranged and having the same number. Through several convolutional upsampling operations, the extracted image features are re-decoded into target features. The output head comprises two parallel multi-layer convolutions; one multi-layer convolution is responsible for translating the target features into higher-resolution super-resolution data (super-resolution task), and the other multi-layer convolution is responsible for translating the target features into parcel data (semantic segmentation task). Loss functions include: mean absolute error loss, adversarial loss, edge enhancement loss, cross-entropy loss, and... The formula for determining the loss function is as follows:

[0056] ;

[0057] in, For the excess score loss, , To divide the loss, , For loss function, The mean absolute error loss coefficient, For the average absolute error loss, To counteract the loss coefficient, To combat the losses, The edge enhancement loss coefficient, To enhance the loss at the edge, The cross-entropy loss coefficient is... For cross-entropy loss, for Loss coefficient, for loss.

[0058] In summary, this application jointly trains the super-resolution network and the field segmentation network, establishing a joint framework of "feature extractor-spatial attention-feature decoder-multi-task head," and constructs a super-resolution-field joint model with a U-net network structure, such as... Figure 4 As shown, where, Figure 4The numbers in the diagram represent the number of channels in the convolutional layer. This application integrates the spatial attention module (Convolutional Block Attention Module) as a single convolutional block into the decoder and encoder, enabling the super-resolution-field joint model to accurately reconstruct the edge information and internal texture information of the fields, thereby improving the recognition accuracy of the fields, also known as cultivated land plots. The output head configuration allows the super-resolution-field joint model to simultaneously perform super-resolution image recognition and field recognition functions, further enhancing the extraction accuracy of the fields. Experiments show that compared to traditional single-task learning, the recognition accuracy of the model in this application is improved by 5% to 8%.

[0059] Furthermore, this application combines the classic loss functions from super-resolution and field extraction tasks, and improves the loss function according to the task objectives of this application, utilizing mean absolute error loss, adversarial loss, edge enhancement loss, cross-entropy loss, and... The super-resolution loss and segmentation loss are determined separately, and the super-resolution loss and segmentation loss are weighted and fused to obtain the loss function (weight ratio 6:4). The improvement of the loss function makes it more suitable for farmland scenes, solves the problem of blurred farmland boundaries caused by low-resolution images, enhances the information of farmland edges, improves the prediction accuracy of the super-resolution-field joint model, and further improves the accuracy of field extraction.

[0060] When training the super-resolution-field joint model, the input consists of historical aligned remote sensing image data pairs and corresponding farmland boundary data. The aim is to enable the super-resolution-field joint model to learn how to extract and decode target features, as well as how to translate the output head. At the same time, the super-resolution-field joint model is trained using publicly available farmland boundary data and existing farmland boundary data. Simulation training is also conducted using remote sensing image data pairs under different lighting conditions and different crop growth cycles to enhance the data of the farmland scene.

[0061] S3: Based on the current medium-resolution remote sensing image of rural farmland, and using the trained super-resolution-field joint model, determine the current super-resolution data and current field data of rural farmland.

[0062] Once the super-resolution-field joint model provided in this application is trained, it only requires input of the current medium-resolution remote sensing imagery, i.e., Sentinel-2 data, to obtain the current super-resolution data and current field data of rural farmland, such as... Figure 5 As shown, the resolution of the super-resolution data is 2m.

[0063] S4: Determine the current field vector data of rural cultivated land based on the current super-resolution data and the current field data.

[0064] S4 specifically includes:

[0065] S41: Align and correct the current super-resolution data and the current field data to obtain aligned field data.

[0066] S42: Vectorize the aligned field data to obtain the current field vector data of rural farmland.

[0067] Morphological optimization of field vector data is performed, such as closing operations to fill small holes and removing noise areas that are too small.

[0068] S5: Connect and match the current field vector data with the ownership database to confirm the ownership of rural farmland.

[0069] S5 specifically includes:

[0070] S51: Overlay the latitude and longitude coordinates of the current field vector data and the ownership database; and determine the area error between the current field vector data and the ownership database.

[0071] ArcGIS software is used to associate and match the current field vector data with the ownership database. The matching rules include spatial location matching and area consistency verification. Spatial location matching involves overlaying the latitude and longitude coordinates of the field vector data and the ownership database. Area consistency verification involves using ArcGIS software to calculate the area deviation between the contracted area in the ownership database and the area of ​​the field vector data.

[0072] S52: The matching rule with a latitude and longitude coordinate overlap greater than 90% and an area error less than 5% is taken as the optimal matching rule, and the matching result is obtained.

[0073] Specifically, the overlap threshold is set to 90%, the area threshold is set to 5%, and the matching rule when the latitude and longitude coordinate overlap is greater than 90% and the area error is less than 5% is taken as the optimal matching rule. If the optimal matching rule is met, the field vector data and the ownership database are considered to match. After the matching conditions are met, the contracted area in the ownership database is assigned to the field vector data to obtain the matching result. The assignment function is implemented in ArcGIS software.

[0074] For fields that cannot be automatically matched in the field vector data, such as newly added farmland and farmland with disputed ownership, they are marked as "farmland to be manually verified".

[0075] S53: Use the matching results to confirm the ownership of rural cultivated land.

[0076] The matching results are used to confirm the ownership of rural farmland, resulting in ownership confirmation data, which includes vector data such as the spatial location, area, and ownership information of the plots. After the ownership confirmation results are linked to the GIS platform, a standard electronic certificate template can be automatically generated, and PDF and blockchain evidence storage are supported.

[0077] Specifically, acquire the latest medium-resolution remote sensing imagery each quarter, re-execute the rights confirmation process, and update the rights confirmation results, such as... Figure 6 As shown, comparing historical and current land ownership confirmation results can automatically identify changes in farmland, such as land parcel merging and splitting, and trigger updates to ownership information to achieve automated and incremental updates of cultivated land parcel data. In addition, a manual screening and alerting rule is used to handle anomalies. This involves comparing historical and current land ownership confirmation results to determine if any anomalies exist; if so, an alert is issued prompting manual intervention; otherwise, the rural cultivated land ownership confirmation results are stored. Anomalies include changes in parcel area exceeding the area change threshold, conflicts in parcel ownership matching, and newly added unregistered cultivated land. The area change threshold is 10%.

[0078] In summary, this application enables an integrated process design for super-division, segmentation, rights confirmation, and blockchain-based evidence storage.

[0079] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0080] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for dynamic updating of arable land plots, characterized in that, The method for dynamically updating cultivated land plots includes: Acquire pairs of remote sensing image data of rural cultivated land history and corresponding farmland boundary data, and construct a dataset; the remote sensing image data pairs include: medium-resolution remote sensing images and high-resolution remote sensing images; Based on the dataset, a super-resolution-field joint model is constructed using a super-resolution network and a field segmentation network. This model outputs super-resolution data and field data corresponding to rural farmland. The super-resolution-field joint model includes an encoder, a decoder, an output head, and a loss function. The encoder comprises multiple cascaded convolutional blocks in a pyramidal shape and is used to extract image features from remote sensing image data pairs. The decoder comprises multiple cascaded convolutional blocks in an inverted pyramid shape and is used to decode the image features. The convolutional blocks of the encoder and the decoder are symmetrically arranged and have the same number. The loss function includes: mean absolute error loss, adversarial loss, edge enhancement loss, cross-entropy loss, and... loss; Based on the current medium-resolution remote sensing images of rural farmland, and using a trained super-resolution-field joint model, the current super-resolution data and current field data of rural farmland are determined. Based on the current super-resolution data and the current field data, determine the current field vector data of rural cultivated land; The current field vector data and ownership database are correlated and matched to confirm the ownership of rural arable land.

2. The method for dynamic updating of cultivated land plots according to claim 1, characterized in that, The acquisition of remote sensing image data pairs of rural cultivated land history and corresponding farmland boundary data, and the construction of a dataset, specifically includes: Historical remote sensing image data pairs are preprocessed to identify farmland areas; the preprocessing process includes: radiometric correction, geometric registration, and farmland area extraction; the farmland area extraction process includes: extracting regions of interest for farmland areas using cultivated land classification in the land classification dataset, and cropping non-farmland areas; The farmland area and the ownership database are spatially aligned to obtain an aligned pair of remote sensing image data. A dataset is constructed based on the aligned remote sensing image data pairs and the corresponding farmland boundary data.

3. The method for dynamic updating of cultivated land plots according to claim 1, characterized in that, The process of determining the loss function includes: Using formula Determine the over-division loss ; Using formula Determine the segmentation loss ; Using formula Determine the loss function ; in, The mean absolute error loss coefficient, For the average absolute error loss, To counteract the loss coefficient, To combat the losses, The edge enhancement loss coefficient, To enhance the loss at the edge, The cross-entropy loss coefficient is... For cross-entropy loss, for Loss coefficient, for loss.

4. The method for dynamic updating of cultivated land plots according to claim 1, characterized in that, The process of determining the current field vector data of rural cultivated land based on the current super-resolution data and the current field data specifically includes: Align and correct the current over-resolution data and the current field data to obtain aligned field data; The aligned field data is vectorized to obtain the current field vector data of rural arable land.

5. The method for dynamic updating of cultivated land plots according to claim 1, characterized in that, The step of associating and matching the current field vector data with the ownership database to confirm the ownership of rural cultivated land specifically includes: Overlay the latitude and longitude coordinates of the current field vector data and the ownership database; and determine the area error between the current field vector data and the ownership database; The matching rule with a latitude and longitude coordinate overlap greater than 90% and an area error less than 5% is taken as the optimal matching rule, and the matching result is obtained. The matching results are used to confirm the ownership of rural arable land.

6. The method for dynamic updating of cultivated land plots according to claim 1, characterized in that, The step of associating and matching the current field vector data with the ownership database to confirm the ownership of rural cultivated land also includes: Based on the results of rural farmland ownership confirmation, determine whether there are any abnormalities; if so, issue an alarm to prompt manual intervention; if not, store the results of rural farmland ownership confirmation. The abnormalities include: changes in field area exceeding the area change threshold, conflicts in field ownership matching, and newly added unregistered farmland.

7. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the method for dynamic updating of cultivated land plots as described in any one of claims 1-6.

Images