Method and system for identifying rural idle land potential
By combining high-resolution remote sensing imagery and the YOLO-SEG model with ArcGIS methods, the potential of idle rural land is automatically identified, solving the problems of low efficiency and poor accuracy of traditional manual identification, and achieving efficient and accurate identification of idle land.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- GUANGXI ZHUANG AUTONOMOUS REGION NATURAL RESOURCES ECOLOGICAL RESTORATION CENT
- Filing Date
- 2026-02-02
- Publication Date
- 2026-05-19
AI Technical Summary
Existing traditional potential identification methods rely on manual processing and visual interpretation, resulting in low efficiency and poor accuracy in identifying idle rural land. They are unable to quickly identify different roof types, lack intelligence, and cannot use detailed thresholds as a reference.
High-resolution remote sensing image data was used to identify rooftops using the YOLO-SEG model. Data processing and potential land parcel screening were performed using ArcGIS. The proportion of tiled roof area was used as the judgment threshold to generate vector files containing potential land parcels.
It significantly improves the accuracy and consistency of identifying the potential of idle land, avoids human bias, realizes automated and intelligent identification, and enhances identification efficiency and scientific rigor.
Smart Images
Figure CN122067097A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of urban and rural construction technology, and more specifically to a method and system for identifying the potential of idle rural land. Background Technology
[0002] The linkage between urban and rural construction land increases and decreases is a national land and resources management policy implemented to optimize land use structure. It achieves a dynamic balance of the total amount of arable land by exchanging the demolition and reclamation of rural construction land with the replacement of new urban construction land.
[0003] By clearing out inefficient and idle construction land in villages, the policy of linking urban and rural land use quotas can be used to transfer rural construction land quotas to urban areas, thus making up for the shortage of urban construction land quotas.
[0004] Existing traditional potential identification methods rely on manual processes: first, data processing is performed, then overlaying images for visual interpretation to identify the potential of inefficient and idle construction land in villages. Because village construction land is scattered and involves many plots, potential identification requires a significant amount of manpower and time. Furthermore, differences in experience and expertise among personnel can lead to biases, hindering rapid identification. Additionally, the methods cannot identify different roof types within villages, and detailed thresholds cannot be used as a reference when determining the final plot category. Consequently, the accuracy and intelligence of the identification process are relatively weak. Summary of the Invention
[0005] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present invention provide a method and system for identifying the potential of idle rural land, so as to solve the technical problems mentioned in the background art.
[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for identifying the potential of idle rural land, comprising the following steps:
[0007] Step S1: Obtain high-resolution remote sensing images of the target area and the corresponding rural residential map layer, and filter out map patches with an area greater than 400 square meters;
[0008] Step S2: Annotate the rooftops in the remote sensing image according to the preset rooftop category system to generate an annotation file with category information;
[0009] Step S3: Cut the remote sensing image into tiles of a set size and extract the coordinate information of each tile;
[0010] Step S4: Convert the annotation file into the normalized coordinate format required by the target detection model;
[0011] Step S5: Train the YOLO-SEG model using the converted labeled data, and use the trained model to identify the roof of the cut-out image tiles, outputting the contour and category results;
[0012] Step S6: Convert the recognition results into a geospatial format and merge similar roof outlines across tiles;
[0013] Step S7: Calculate the percentage of tiled roof area in each construction map patch, identify plots with a percentage exceeding the set ratio as potential idle land plots, and generate vector files containing spatial and attribute information of potential plots.
[0014] A system for identifying the potential of idle rural land includes a data unit, a sample annotation unit, an image segmentation unit, a coordinate management unit, a format conversion unit, a model training unit, an image recognition unit, a result processing unit, a potential land parcel screening unit, and a data output unit. The data unit acquires image data; the sample annotation unit generates SHP format annotation files; the image segmentation unit segments images and generates JPG format files; the coordinate management unit acquires coordinate information; the format conversion unit performs normalized coordinate transformation on the SHP annotation files; the model training unit trains and optimizes the YOLO-SEG model; the image recognition unit uses the YOLO-SEG model for image recognition and outputs results; the result processing unit performs post-processing on the recognition results; the potential land parcel screening unit selects potential land parcels; and the data output unit outputs standardized result data.
[0015] The data unit acquires the latest 0.5-meter high-resolution remote sensing images covering the target area from a professional platform, and then exports the rural residential map layer that overlaps with the remote sensing image area and time from the latest land change survey database. Through the attribute query and spatial calculation functions of GIS software, it filters out the map patches with an area field greater than 400 square meters and saves them as SHP files.
[0016] In a preferred embodiment, the sample annotation unit imports the SHP file into ArcGIS, annotates the training set according to a preset roof category system, and outlines the roof along its actual boundary during annotation. The sample annotation unit then generates an SHP format annotation file containing the roof outline and category information.
[0017] In a preferred embodiment, the image cutting unit first cuts the remote sensing image into 2048×2048 pixel tiles, then cuts the tiles into 512×512 pixel tiles, naming them in the format of "original file name_number of rows_number of columns_top left corner X coordinate_top left corner Y coordinate", converts the cut image into a 16-bit unsigned format, and then into JPG format.
[0018] In a preferred embodiment, the coordinate management unit is used to extract the top-left (X1, Y1) coordinates of each tile, and the coordinate unit directly obtains the coordinate information of each tile through the file name. The format conversion unit converts the SHP annotation file into the format required for YOLO-SEG model training through normalization. The normalized coordinate conversion formula is:
[0019]
[0020]
[0021] In the formula, min_x and max_y are the X and Y coordinates of the top left corner of the tile, geo_width and geo_height are the width and height of the actual area where the tile is located, point.X and point.Y are the original X and Y coordinates of the annotation point, min_x is the X coordinate of the top left corner of the tile in the current image, and max_y is the Y coordinate of the top left corner of the tile in the current image. The format conversion unit maps the roof category to integer labels and generates a TXT annotation file with each line containing the category label and the normalized coordinate sequence. Each tile corresponds to one annotation file.
[0022] In a preferred embodiment, the model training unit has preset training parameters including: epochs = 2000, initial learning rate lr = 0.001 with a decay to 0.9 times every 10 epochs, batch size = 16, weight decay of 0.0005, and momentum of 0.9. During the training process, the model training unit monitors the loss function, validation set mAP, and recall. If the mAP no longer increases and the loss continues to decrease, it is determined to be overfitting, triggering a model optimization strategy.
[0023] In a preferred embodiment, the image recognition unit calls the trained YOLO-SEG model to perform batch inference on satellite image tiles segmented into 512×512 pixels, and outputs YOLO-SEG format recognition results containing roof outline coordinates and category labels.
[0024] In a preferred embodiment, the post-processing of the recognition results by the result processing unit includes inverse format conversion and cross-tile contour merging. The inverse format conversion restores the normalized coordinates to the actual coordinates. The restoration formula is as follows:
[0025] ;
[0026] ;
[0027] In the formula, tile_x and tile_y are the actual coordinates of the upper left corner of the tile, img_w and img_h are the actual pixel size of the tile, gsd represents the actual size of each pixel in the recognition image, a Polygon geometric object is created using the arcpy library, a roof recognition result SHP file is generated, and adjacent inference result surfaces of the same type are merged across the tile contour.
[0028] In a preferred embodiment, the potential land parcel screening unit determines potential development land parcels according to rules, and the determination rules of the potential land parcel screening unit are as follows: within a construction land parcel, if the total area of tiled roofs accounts for more than 30% of the parcel area, it is marked as a potential land parcel. The potential land parcel screening unit distinguishes potential land parcels by potential value Q, calculated using the following formula: In the formula, SW is the total area of the tiled roof, ST is the area of the map patch, I is the spatial connectivity index between the map patch and roads and public service facilities, DZ is the distance of the map patch from the village center, DM is the distance from the village center to the village edge, k1, k2 and k3 are weight coefficients, and their values range from 0 to 1. The potential value Q is positively correlated with the development potential of the potential plot.
[0029] In a preferred embodiment, the data output unit outputs standardized result data, and the data output unit creates a GeoDataFrame containing fields such as "plot number, plot area, tiled roof area, tiled roof percentage, and geometry" using Python, sets a coordinate system consistent with the original data, and calls the to_file method to save it as an SHP format potential plot vector surface file.
[0030] The technical effects and advantages of this invention are as follows:
[0031] 1. This invention integrates remote sensing images and GIS data, and uses the YOLO-SEG artificial intelligence model to significantly improve the accuracy and consistency of the identification results, avoiding subjective bias caused by differences in human experience. This method improves the efficiency of identifying the potential of idle rural land, points the way for carrying out the linkage between urban and rural construction land increases and decreases, and frees up land quotas for local urban construction.
[0032] 2. This invention ensures the geographic authenticity of training data by outlining the actual boundaries of the roof. The SHP format file is easy to process and verify in the GIS environment and can also be easily converted to other annotation formats, making data processing more convenient in this application. When judging the roof category, it mainly relies on the characteristics of the roof such as color, texture, and shape in satellite imagery to accurately judge the roof.
[0033] 3. When identifying potential land parcels, this invention uses a 30% threshold, which makes the identification more accurate. Furthermore, through the spatial association and statistical functions of ArcGIS, it achieves automated potential calculation, resulting in a higher degree of automation and intelligence. A high proportion of tiled roofs usually indicates aging buildings and high vacancy rates, which are positively correlated with the potential of idle land. This application demonstrates a higher level of scientific rigor in its identification process. Attached Figure Description
[0034] Figure 1 This is a schematic diagram of the identification method of the present invention.
[0035] Figure 2 This is a schematic diagram of the composition structure of the identification system of the present invention. Detailed Implementation
[0036] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. In addition, the forms of the various structures described in the following embodiments are merely illustrative. The method and system for identifying the potential of idle rural land involved in the present invention are not limited to the structures described in the following embodiments. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0037] Reference Figure 1 This invention provides a method for identifying the potential of idle rural land, comprising the following steps:
[0038] Step S1: Obtain high-resolution remote sensing images of the target area and the corresponding rural residential map layer, and filter out map patches with an area greater than 400 square meters;
[0039] Step S2: Annotate the rooftops in the remote sensing image according to the preset rooftop category system to generate an annotation file with category information;
[0040] Step S3: Cut the remote sensing image into tiles of a set size and extract the coordinate information of each tile;
[0041] Step S4: Convert the annotation file into the normalized coordinate format required by the target detection model;
[0042] Step S5 uses the converted labeled data to train the YOLO-SEG model, and uses the trained model to identify the roof of the cut-out image tiles, outputting the contour and category results;
[0043] Step S6: Convert the recognition results into a geospatial format and merge similar roof outlines across tiles;
[0044] Step S7: Calculate the percentage of tiled roof area in each construction map patch, identify plots with a percentage exceeding the set ratio as potential idle land plots, and generate vector files containing spatial and attribute information of potential plots.
[0045] In this embodiment, by integrating remote sensing images and GIS data and using the YOLO-SEG artificial intelligence model, the accuracy and consistency of the identification results are significantly improved, avoiding subjective bias caused by differences in human experience. This method improves the efficiency of identifying the potential of idle rural land, points the way for carrying out the linkage between urban and rural construction land increases and decreases, and frees up land quotas for local urban construction.
[0046] Reference Figure 2 A system for identifying the potential of idle rural land includes a data unit, a sample annotation unit, an image segmentation unit, a coordinate management unit, a format conversion unit, a model training unit, an image recognition unit, a result processing unit, a potential land parcel screening unit, and a data output unit. The data unit acquires image data; the sample annotation unit generates SHP format annotation files; the image segmentation unit segments images and generates JPG format files; the coordinate management unit acquires coordinate information; the format conversion unit performs normalized coordinate transformation on the SHP annotation files; the model training unit trains and optimizes the YOLO-SEG model; the image recognition unit uses the YOLO-SEG model for image recognition and result output; the result processing unit performs post-processing on the recognition results; the potential land parcel screening unit selects potential land parcels; and the data output unit outputs standardized result data.
[0047] In this embodiment, tiled roofs are used as the key identification category, and their area ratio within the map patch is used as the potential judgment threshold. This makes the identification more accurate. In rural areas, tiled roofs are usually a typical feature of traditional old houses. Their large-scale presence often means low building density, low land use efficiency, or vacant houses. Therefore, this limitation establishes a strong correlation between technical identification and actual land use conditions, making the judgment of idle land potential an objective indicator, thereby improving the scientific nature of the identification results.
[0048] Reference Figure 2 The data unit acquires the latest 0.5-meter high-resolution remote sensing image covering the target area from a professional platform, and then exports the rural residential map layer that overlaps with the remote sensing image area and time from the latest land change survey database. Through the attribute query and spatial calculation functions of GIS software, the map patches with an area field greater than 400 square meters are selected and saved as SHP files.
[0049] In this embodiment of the application, images are obtained through remote sensing, which yields more accurate images. Furthermore, by filtering out image patches with an area field greater than 400 square meters, old houses can be accurately identified, thereby improving subsequent accuracy.
[0050] Reference Figure 2 The sample annotation unit imports the SHP file into ArcGIS and annotates the training set according to the preset roof category system, which includes asbestos roof, cement roof, solar panel, corrugated iron roof and tile roof. When annotating, the outline is drawn along the actual boundary of the roof, and the roof category is determined based on color, texture and shape features. The sample annotation unit finally generates an SHP format annotation file containing roof outline and category information.
[0051] In this embodiment, the application ensures the geographic authenticity of the training data by outlining the actual boundary of the roof. The application pre-defines roof categories (such as tile roofs, cement roofs, etc.) to facilitate model learning and classification. The SHP format file is easy to process and verify in the GIS environment and can be easily converted to other annotation formats, making the application more convenient for data processing. Furthermore, the asbestos tile category is merged into tile roofs. When determining the roof category, the application mainly relies on the characteristics of the roof, such as color, texture, and shape in satellite imagery, to accurately determine the roof.
[0052] Reference Figure 2 The image cropping unit first calls ArcGIS's Split Raster tool to crop the remote sensing image, setting the cropping method to SIZE_OF_TILE, the tile size to 2048×2048 pixels, and the output format to TIF. Then, a Python script calls the ArcPy package to crop the tiles a second time to 512×512 pixels, naming them in the format "original filename_number of rows_number of columns_top-left X coordinate_top-left Y coordinate". Finally, ArcGIS Pro converts the cropped image to a 16-bit unsigned format and then to JPG format.
[0053] In this embodiment, the image is cut into 512×512 pixels after two cuts. 512×512 pixels is suitable for the input size of models such as YOLO-SEG, improving training and inference efficiency. Furthermore, coordinates are embedded in the filename, facilitating subsequent geographic coordinate reconstruction, reducing computational intensity, and decreasing storage pressure after conversion to JPG, thus improving processing speed.
[0054] Reference Figure 2The coordinate management unit is used to extract the top-left (X1, Y1) coordinates of each tile, and the coordinate unit directly obtains the coordinate information of each tile through the file name. The format conversion unit converts the SHP annotation file into the format required for YOLO-SEG model training through normalization. The normalized coordinate conversion formula is:
[0055]
[0056]
[0057] In the formula, min_x and max_y are the X and Y coordinates of the top left corner of the tile, geo_width and geo_height are the width and height of the actual area where the tile is located, point.X and point.Y are the original X and Y coordinates of the annotation point, min_x is the X coordinate of the top left corner of the tile in the current image, and max_y is the Y coordinate of the top left corner of the tile in the current image. The format conversion unit maps the roof category to integer labels and generates a TXT annotation file with each line containing the category label and the normalized coordinate sequence. Each tile corresponds to one annotation file.
[0058] In this embodiment, the actual coordinates are converted into normalized coordinates required for model training, which improves the model's convergence speed and stability, thereby facilitating model training and subsequent calculations. The roof category is mapped to integer labels, which facilitates model recognition and output. After format conversion, each tile corresponds to a label file, making the overall structure clear and facilitating training data management and retrieval.
[0059] Reference Figure 2 The model training unit has preset training parameters including: epochs = 2000, initial learning rate lr = 0.001 with a decay to 0.9 every 10 epochs, batch size = 16, weight decay of 0.0005, and momentum of 0.9. During the training process, the model training unit monitors the loss function, validation set mAP, and recall. If the mAP no longer increases and the loss continues to decrease, it is determined to be overfitting, triggering a model optimization strategy.
[0060] In this embodiment, during training, reasonable settings of learning rate decay, momentum, and weight decay parameters help the model converge quickly. Furthermore, monitoring mAP and loss function during training allows for dynamic adjustment of the training strategy to improve the model's generalization ability. If mAP no longer increases and loss continues to decrease, it indicates a fitted state, and timely optimization is performed to ensure the accuracy of the model in this application.
[0061] Reference Figure 2The image recognition unit calls the trained YOLO-SEG model to perform batch inference on satellite image tiles segmented into 512×512 pixels, and outputs YOLO-SEG format recognition results containing roof outline coordinates and category labels.
[0062] In this embodiment, the trained YOLO-SEG model is called for batch inference: it supports parallel processing of the segmented image tiles, greatly improves recognition efficiency, and outputs structured results containing contour coordinates and category labels, which facilitates subsequent spatial analysis and statistics, thereby making subsequent processing more efficient.
[0063] Reference Figure 2 The result processing unit performs post-processing on the recognition results, including inverse format conversion and cross-tile contour merging. Inverse format conversion restores the normalized coordinates to the actual coordinates. The restoration formula is:
[0064] ;
[0065] ;
[0066] In the formula, tile_x and tile_y are the actual coordinates of the upper left corner of the tile, img_w and img_h are the actual pixel size of the tile, gsd represents the actual size of each pixel in the recognition image, a Polygon geometric object is created using the arcpy library, a roof recognition result SHP file is generated, and adjacent inference result surfaces of the same type are merged across the tile contour.
[0067] In this embodiment, the normalized coordinates output by the model are restored to the actual geographic coordinates to ensure spatial accuracy. Cross-tile merging avoids fragmentation of the identification caused by the same roof being cut into multiple tiles, improves the integrity of the outline, and finally generates an SHP file, which is convenient for direct use and visualization in GIS.
[0068] Reference Figure 2 The potential land parcel screening unit determines the development potential land parcels according to rules, and the determination rules of the potential land parcel screening unit are as follows: within the construction map parcel, if the total area of tiled roofs accounts for more than 30% of the parcel area, it is marked as a potential land parcel. Through the spatial association function of ArcGIS, the identified roof range is associated with the construction map parcel. The area ratio of the roof range to the construction map parcel area is calculated. After calculating the area ratio of each roof, the area ratio of tiled roofs in the parcel is summarized through the BSM field of the construction map parcel using the summary statistics function of ArcGIS. Parcels with a ratio exceeding 30% are selected as potential land parcels.
[0069] In this embodiment of the application, a threshold of 30% is used when identifying potential land parcels, thus making the identification more accurate. Furthermore, through the spatial association and statistical functions of ArcGIS, automated potential calculation is achieved, resulting in a higher degree of automation and intelligence. A high proportion of tiled roofs usually indicates aging buildings and high vacancy rates, which are positively correlated with the potential of idle land. Therefore, the scientific basis of this application in the identification process is higher.
[0070] Reference Figure 2 The potential land parcel screening unit distinguishes potential land parcels based on a potential value Q, calculated using the following formula: In the formula, SW is the total area of the tiled roof, ST is the area of the map patch, I is the spatial connectivity index between the map patch and roads and public service facilities, DZ is the distance of the map patch from the village center, DM is the distance from the village center to the village edge, k1, k2 and k3 are weight coefficients, and their values range from 0 to 1. The potential value Q is positively correlated with the development potential of the potential plot.
[0071] In this embodiment, the potential of a land parcel is ranked by calculating a potential value Q based on the sum of the roof area, the area of the land parcel, the spatial connectivity index, and the difference between 1 and the ratio of the distance from the land parcel to the village center to the distance from the village center to the village edge. The larger the calculated potential value Q, the greater the potential. Therefore, development can be prioritized based on the ranking of potential values Q, thus making better site selections. The difference between 1 and the ratio of the distance from the land parcel to the village center to the distance from the village center to the village edge has a maximum value of 1 when a land parcel is located in the village center and a minimum value of 0 when a land parcel is located at the edge of the village. The closer the land parcel is to the center, the higher its "location convenience" score. Idle land parcels located inside the village have relatively lower costs for relocation, resettlement, and infrastructure support when the indicators generated after reclamation are used for urban construction, and the feasibility of implementation is higher. Therefore, they have greater potential.
[0072] Reference Figure 2 The data output unit outputs standardized result data, and the data output unit creates a GeoDataFrame containing fields such as "plot number, plot area, tiled roof area, tiled roof percentage, and geometry" using Python, sets a coordinate system consistent with the original data, and calls the to_file method to save it as an SHP format potential plot vector surface file.
[0073] In this embodiment, the output includes data such as plot number, area, tiled roof ratio, and geometric information to meet subsequent analysis needs. The output coordinate system is consistent with the input data, making it easy to integrate into existing GIS platforms. The final SHP format is widely supported and can be directly used in mapping, planning, and decision support systems, making the application more convenient.
[0074] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. The units and algorithm steps of the various examples described in the embodiments can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.
[0075] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.
[0076] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
[0077] In conclusion, 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 within the protection scope of the present invention.
Claims
1. A method for identifying the potential of idle rural land, characterized in that: Includes the following steps: Step S1: Obtain high-resolution remote sensing images of the target area and the corresponding rural residential map layer, and filter out map patches with an area greater than 400 square meters; Step S2: Annotate the rooftops in the remote sensing image according to the preset rooftop category system to generate an annotation file with category information; Step S3: Cut the remote sensing image into tiles of a set size and extract the coordinate information of each tile; Step S4: Convert the annotation file into the normalized coordinate format required by the target detection model; Step S5: Train the YOLO-SEG model using the converted labeled data, and use the trained model to identify the roof of the cut-out image tiles, outputting the contour and category results; Step S6: Convert the recognition results into a geospatial format and merge similar roof outlines across tiles; Step S7: Calculate the percentage of tiled roof area in each construction map patch, identify plots with a percentage exceeding the set ratio as potential idle land plots, and generate vector files containing spatial and attribute information of potential plots.
2. A system for identifying the potential of idle rural land, characterized in that: The system includes a data unit, a sample annotation unit, an image segmentation unit, a coordinate management unit, a format conversion unit, a model training unit, an image recognition unit, a result processing unit, a potential land parcel screening unit, and a data output unit. The data unit is used for image data acquisition; the sample annotation unit is used to generate SHP format annotation files; the image segmentation unit segments the image and generates JPG format files; the coordinate management unit is used to acquire coordinate information; the format conversion unit performs normalized coordinate transformation on the SHP annotation files; the model training unit trains and optimizes the YOLO-SEG model; the image recognition unit uses the YOLO-SEG model to perform image recognition and output results; the result processing unit performs post-processing on the recognition results; the potential land parcel screening unit screens out potential land parcels; and the data output unit outputs standardized result data. The data unit acquires the latest 0.5-meter high-resolution remote sensing images covering the target area from a professional platform, and then exports the rural residential map layer that overlaps with the remote sensing image area and time from the latest land change survey database. Through the attribute query and spatial calculation functions of GIS software, it filters out the map patches with an area field greater than 400 square meters and saves them as SHP files.
3. The system for identifying the potential of idle rural land according to claim 2, characterized in that: The sample annotation unit imports the SHP file into ArcGIS and annotates the training set according to the preset roof category system. During annotation, the outline is drawn along the actual boundary of the roof. The sample annotation unit finally generates an SHP format annotation file containing the roof outline and category information.
4. The system for identifying the potential of idle rural land according to claim 2, characterized in that: The image cutting unit first cuts the remote sensing image into 2048×2048 pixel tiles, then cuts the tiles into 512×512 pixels, naming them in the format of "original file name_number of rows_number of columns_top left corner X coordinate_top left corner Y coordinate". The cut image is then converted into a 16-bit unsigned format and then into JPG format.
5. A system for identifying the potential of idle rural land according to claim 2, characterized in that: The coordinate unit directly obtains the coordinate information of each tile through the filename. The format conversion unit converts the SHP annotation file into the format required for YOLO-SEG model training through normalization. The normalized coordinate conversion formula is: In the formula, min_x and max_y are the X and Y coordinates of the top left corner of the tile, geo_width and geo_height are the width and height of the actual area where the tile is located, point.X and point.Y are the original X and Y coordinates of the annotation point, min_x is the X coordinate of the top left corner of the tile in the current image, and max_y is the Y coordinate of the top left corner of the tile in the current image. The format conversion unit maps the roof category to integer labels and generates a TXT annotation file with each line containing the category label and the normalized coordinate sequence. Each tile corresponds to one annotation file.
6. The system for identifying the potential of idle rural land according to claim 2, characterized in that: The model training unit has the following preset training parameters: number of training epochs = 2000, initial learning rate lr = 0.001 which decays to 0.9 times every 10 epochs, batch size = 16, weight decay of 0.0005, and momentum of 0.
9. During the training process, the model training unit monitors the loss function, validation set mAP, and recall. If the mAP no longer increases and the loss continues to decrease, it is determined to be overfitting, and the model optimization strategy is triggered.
7. The system for identifying the potential of idle rural land according to claim 2, characterized in that: The image recognition unit calls the trained YOLO-SEG model to perform batch inference on satellite image tiles segmented into 512×512 pixels, and outputs YOLO-SEG format recognition results containing roof outline coordinates and category labels.
8. A system for identifying the potential of idle rural land according to claim 2, characterized in that: The result processing unit performs post-processing on the recognition results, including inverse format conversion and cross-tile contour merging. Inverse format conversion restores the normalized coordinates to the actual coordinates. The restoration formula is as follows: ； ； In the formula, tile_x and tile_y are the actual coordinates of the upper left corner of the tile, img_w and img_h are the actual pixel size of the tile, gsd represents the actual size of each pixel in the recognition image, a Polygon geometric object is created using the arcpy library, a roof recognition result SHP file is generated, and adjacent inference result surfaces of the same type are merged across the tile contour.
9. A system for identifying the potential of idle rural land according to claim 2, characterized in that: The potential land parcel screening unit determines development potential land parcels according to rules, and the determination rules of the potential land parcel screening unit are as follows: within a construction land parcel, if the total area of tiled roofs accounts for more than 30% of the parcel area, it is marked as a potential land parcel. The potential land parcel screening unit distinguishes potential land parcels by potential value Q, calculated using the following formula: In the formula, SW is the total area of the tiled roof, ST is the area of the map patch, I is the spatial connectivity index between the map patch and roads and public service facilities, DZ is the distance of the map patch from the village center, DM is the distance from the village center to the village edge, k1, k2 and k3 are weight coefficients, and their values range from 0 to 1. The potential value Q is positively correlated with the development potential of the potential plot.
10. A system for identifying the potential of idle rural land according to claim 2, characterized in that: The data output unit outputs standardized result data, and the data output unit creates a GeoDataFrame containing fields such as "plot number, plot area, tiled roof area, tiled roof percentage, and geometry" using Python, sets a coordinate system consistent with the original data, and saves it as a potential plot vector surface file in SHP format.