System and method for carrying out land intelligent collection work according to cloud platform map data

By using the combination method of intensity calculation, accuracy evaluation and land collection modules in land remote sensing image processing, the model is trained using mean square error and cross-entropy loss function, the problem of noise introduced by data enhancement is solved and the accuracy of land collection is improved Sex and reliability.

CN118966545AInactive Publication Date: 2024-11-15CHONGQING PLANNING & NATURAL RESOURCES INFORMATION CENT
View PDF 8 Cites 0 Cited by

Patent Information

Application Number
CN202411042289.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-31
Publication Date
2024-11-15
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

In the prior art In the processing of land remote sensing image data, data augmentation may introduce noise or distortion, affecting the generalization ability of the model, and it is difficult to simplify the training process and improve the quality of data classification.

Method used

The edge strength calculation module is used to calculate the edge strength, the accuracy evaluation module performs global accuracy evaluation, and the land collection module collects land attributes through probability calculation, and trains the model through mean square error loss function, cross entropy loss function and global accuracy evaluation to improve the Real-time and reliability of land collection.

Benefits of technology

Through pre-trained edge difference cutting images, parallel cutting image data are divided, and the edge difference analysis of land features is carried out. The probability that the edge difference of the output cut image belongs to the planned outline area is significantly improved, which significantly improves the accuracy and reliability of land collection.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118966545A_ABST
    Figure CN118966545A_ABST
Patent Text Reader

Abstract

The invention provides a system for performing land intelligent collection work according to cloud platform map data, and the system comprises an intensity calculation module which is used for obtaining space map data, and carrying out the edge intensity calculation of a planning contour region according to a satellite remote sensing image; the precision evaluation module is used for carrying out global precision evaluation on the cut image subjected to edge strength calculation; and the land collection module is used for carrying out land attribute collection on the evaluation result through probability calculation and uploading the result to the cloud platform.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of data analysis, and in particular to a system and method for intelligently collecting land based on cloud platform map data. Background Art

[0002] Real-time measurement and analysis of land through satellite remote sensing technology can meet the data needs for land construction progress and expected planning and development. The development and changes of land attachments affect planning and construction needs. If such changes are allowed to develop, it will increase the management costs in the later stage, which is not conducive to the requirements of healthy urban development. This requires the optimization of land remote sensing image extraction technology. In most image extraction and analysis fields, deep learning is used for image data mining. For example, in CN114842333A, buildings are extracted through remote sensing information. The SKASPP module obtained by combining the SKNet attention module and the ASPP module replaces the original ASPP module in the DeepLab V3+ model. Data enhancement processing is performed through each original remote sensing image and its label map in the sample data set. Although data enhancement can increase the diversity of training samples and reduce the risk of overfitting, excessive data enhancement may also introduce noise or distortion, affecting the generalization ability of the model. How to simplify the training process and improve the quality of data classification, this urgently requires technicians in this field to solve the corresponding technical problems. Summary of the invention

[0003] The present invention aims to at least solve the technical problems existing in the prior art, and particularly innovatively proposes a system and method for intelligent land collection based on cloud platform map data.

[0004] In order to achieve the above-mentioned object of the present invention, the present invention provides a system for intelligently collecting land according to cloud platform map data, comprising:

[0005] Strength calculation module, used to obtain spatial map data and calculate the edge strength of the planned contour area based on satellite remote sensing images;

[0006] The accuracy evaluation module is used to perform global accuracy evaluation on the cut image after edge strength calculation;

[0007] The land collection module is used to collect land attributes based on the evaluation results through probability calculation and upload them to the cloud platform.

[0008] In a preferred embodiment of the present invention, the intensity calculation module includes:

[0009] Use the imread function to read the cut image to be processed, convert the cut image into a grayscale image, and store it in the candidate variable I. Then, use the RGB2gray function to convert the color image into a grayscale image for edge detection. The converted grayscale image is stored in the grayscale variable I. gray In the horizontal direction, the Sobel operator G x : Used to detect horizontal edges, vertical Sobel operator G y , used to detect vertical edges, using G x Convolve the cut image to get the horizontal edge intensity map, using G y Convolve the image to obtain the vertical edge intensity map. For each pixel in the cut image, calculate the grayscale gradient amplitude I gray , which is the square root of the sum of the squares of the edge strengths in the horizontal and vertical directions,

[0010]

[0011] Use the imfilter function to apply the Sobel operator to a grayscale image. The imfilter function performs a convolution operation where G x and G y They are used to detect edges in the horizontal and vertical directions respectively. x and I y are the edge strengths in the horizontal and vertical directions,

[0012] Use the gradient Sobel operator to perform convolution operations on the cut image to obtain the gradient map of the cut image in the horizontal and vertical directions. The magnitude of the gradient reflects the rate of change of the image grayscale, and the area with a larger gradient often corresponds to edge details. Perform a two-dimensional wavelet transform: Perform a two-dimensional wavelet transform on the image and decompose it into sub-bands of different frequencies. This usually involves performing a one-dimensional wavelet transform on the rows and columns of the image respectively.

[0013] In the result of two-dimensional wavelet transform, the horizontal high-frequency coefficient corresponds to the high-frequency component in the horizontal direction of the transformed image, and the vertical high-frequency coefficient corresponds to the high-frequency component in the vertical direction. The horizontal high-frequency coefficient and the vertical high-frequency coefficient are directly extracted from the result of wavelet transform.

[0014] In a preferred embodiment of the present invention, the intensity calculation module includes:

[0015] Use the mean square error loss function K

[0016]

[0017] Among them, β is the adjustment weight, which is used to balance the importance of the two directions. For Ix The edge strength of M pixels in the predicted horizontal direction and true lateral edge strength The mean of the sum of squared differences, For I y The edge strength of M pixels in the predicted longitudinal direction and true longitudinal edge strength The mean of the sum of squared differences, i is the cut image pixel.

[0018] In a preferred embodiment of the present invention, the intensity calculation module includes:

[0019] By extracting the edge features of the cut image, an edge label difference dataset is formed. By traversing and extracting the edge difference points, the cross entropy loss function E(·) is calculated as the objective function for cutting image edge difference extraction

[0020]

[0021] Among them, N is the total number of edge cutting images extracted, p is the marked edge sample vector, k is the edge difference cutting image output vector calculated by the mean square error loss function K, j is the edge cutting image sequence number, ε is the weight of adjusting the edge difference sample, and λ is the weight of adjusting the edge cutting image sample.

[0022] In a preferred embodiment of the present invention, the accuracy assessment module includes:

[0023] As the calculation condition of the global accuracy F,

[0024]

[0025] Among them, w d is the output weight of the pixel d extracted from the edge difference of the cut image, and f(N) is the number of edge difference cut images in all edge cut images N that have been calculated by the extracted objective function.

[0026] The output weight formula is w d =median(d) / d, where median(d) is the median value of the pixel point d extracted from the edge difference of the cut image in the land planning contour area.

[0027] In a preferred embodiment of the present invention, the land collection module includes:

[0028] The land planning contour area is used to cut image samples. After accuracy evaluation, the pixel information loss caused by image stitching is performed. After multiple samplings, the continuity of semantic information in the restoration process is increased. The trained convolutional coding network model is used as a pre-training model to train the entire encoding-decoding network.

[0029] The satellite remote sensing images collected in real time are input into the trained DeepLab v3+ model, and the trained encoding-decoding network is converted into a network for testing satellite remote sensing images collected in real time. The loss layer and the precision layer are removed, and the Softmax layer is added to calculate the edge difference probability.

[0030] Where A is the probability of forming a z-type edge difference cut image, e is the base of the natural logarithm, and z is the original output value of the edge difference cut image belonging to the edge difference category after global accuracy evaluation;

[0031] The edge difference of the output cut image belongs to the probability of the planning contour area; by setting the judgment threshold, the probability map is converted into a binary image, that is, the pre-labeled planning contour area is obtained, thereby completing the land collection statistical operation.

[0032] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0033] Through the pre-trained edge difference cutting image, the satellite remote sensing image in the input prediction set is divided into parallel cutting image data, and the land feature edge difference analysis of the monitoring area is carried out based on the trained cutting image data to obtain the land feature edge difference prediction result corresponding to the monitoring area. By constructing and training the objective function of cutting image edge difference extraction and performing a global accuracy evaluation of the output image vector according to the time series, the probability that the edge difference of the output cutting image belongs to the planned contour area is obtained.

[0034] Intelligent land collection includes the series-connected mean square error loss function, cross entropy loss function, global accuracy evaluation and difference probability calculation. The purpose is to train the difference between land difference changes and surrounding adjacent spatial information through the corresponding loss function calculation, to help the real-time and reliability of land collection, and to make the collection results more accurate.

[0035] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0037] Figure 1 It is a schematic diagram of the system of the present invention.

[0038] Figure 2 It is a schematic diagram of the method of the present invention.

[0039] Figure 3It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0040] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0041] like Figures 1 to 3 As shown, the present invention discloses a system for intelligently collecting land according to cloud platform map data, comprising:

[0042] The corresponding satellite remote sensing image data is obtained through the data management layer, the image data is interactively accessed through the data access layer, and written into the database layer.

[0043] Strength calculation module, used to obtain spatial map data and calculate the edge strength of the planned contour area based on satellite remote sensing images;

[0044] The accuracy evaluation module is used to perform global accuracy evaluation on the cut image after edge strength calculation;

[0045] The land collection module is used to collect land attributes based on the evaluation results through probability calculation and upload them to the cloud platform.

[0046] After judging the land attachments through spatial map data, the controlled and planned plots are confirmed, referred to as the controlled planning plots. The actual status of the controlled planning plots is defined as existing land, and the newly added planned construction land boundary line or interval is preset to determine whether the controlled planning plot is located in the newly added planned construction land boundary line or interval. If so, the actual status is marked as already used. If not, the actual status definition remains unchanged.

[0047] When the actual status is marked as already in use, a land occupation (transfer, allocation status) range line is preset to determine whether the planning control plot is located on the land occupation range line. If so, the actual status is marked as supplied. If not, the actual status definition remains unchanged.

[0048] By analogy, the construction project land red line, the expected construction project outline, the construction project completion outline and the real estate registration house outline are preset and judged respectively.

[0049] Preferably, in the above technical solution, the strength calculation module comprises:

[0050] The land data to be monitored is marked through satellite remote sensing images, and a planning contour area is established for the land data. The land planning contour area is marked. Different pixels in the remote sensing image need to be marked as corresponding categories (such as buildings, green spaces, roads), and the corresponding categories are assigned a unique integer ID;

[0051] Satellite remote sensing images are usually Landsat series, Sentinel series, etc. These satellites provide high-resolution images around the world.

[0052] The massive satellite remote sensing images collected are used to mark the boundaries of the pre-set planning contour areas. The marked images are in the form of pixel-level annotations, for example, mask images in PNG format are used to represent the planning contour areas.

[0053] After preliminary labeling, the dataset is divided into training set and validation set. The DeepLab v3+ model is trained using the training set. The satellite remote sensing images to be acquired are image data collected by month. The labeled images in the training set need to be segmented so that the segmented labeled images can be processed in parallel. In the process of adjusting the accuracy parameters, the accuracy parameters are adjusted by cross-validation method. In the training set, the edge scale of the labeled building images is calculated by the building edge detection method.

[0054] Wavelet transform is an effective multi-scale analysis tool that decomposes satellite remote sensing images into sub-bands of different frequencies, including horizontal high frequency, vertical high frequency, diagonal high frequency and low frequency sub-bands.

[0055] Use the imread function to read the cut image to be processed, convert the cut image into a grayscale image, and store it in the candidate variable I. Then, use the RGB2gray function to convert the color image into a grayscale image for edge detection. The converted grayscale image is stored in the grayscale variable I. gray In the horizontal direction, the Sobel operator G x : Used to detect horizontal edges, vertical Sobel operator G y , used to detect vertical edges, using G x Convolve the cut image to get the horizontal edge intensity map, using G y Convolve the image to obtain the vertical edge intensity map. For each pixel in the cut image, calculate the grayscale gradient amplitude I gray , which is the square root of the sum of the squares of the edge strengths in the horizontal and vertical directions,

[0056]

[0057] Use the imfilter function to apply the Sobel operator to a grayscale image. The imfilter function performs a convolution operation where G x and G y They are used to detect edges in the horizontal and vertical directions respectively. x and I y are the edge strengths in the horizontal and vertical directions,

[0058] Use the gradient Sobel operator to perform convolution operations on the cut image to obtain the gradient map of the cut image in the horizontal and vertical directions. The magnitude of the gradient reflects the rate of change of the image grayscale, and the area with a larger gradient often corresponds to edge details. Perform a two-dimensional wavelet transform: Perform a two-dimensional wavelet transform on the image and decompose it into sub-bands of different frequencies. This usually involves performing a one-dimensional wavelet transform on the rows and columns of the image respectively.

[0059] In the result of two-dimensional wavelet transform, the horizontal high-frequency coefficient corresponds to the high-frequency component in the horizontal direction of the transformed image, and the vertical high-frequency coefficient corresponds to the high-frequency component in the vertical direction. The horizontal high-frequency coefficient and the vertical high-frequency coefficient are directly extracted from the result of wavelet transform.

[0060] Calculate the loss function, which calculates the predicted edge strength I x and I y The difference between the real edge and the real edge is calculated, and different weights are assigned to different edge areas. During the training process, this loss function is used to adjust the parameters of the Sobel operator to better extract the horizontal high-frequency coefficients and the vertical high-frequency coefficients.

[0061] Use the mean square error loss function K

[0062]

[0063] Among them, β is the adjustment weight, which is used to balance the importance of the two directions. For I x The edge strength of M pixels in the predicted horizontal direction and true lateral edge strength The mean of the sum of squared differences, For I y The edge strength of M pixels in the predicted longitudinal direction and true longitudinal edge strength The mean of the sum of squared differences, i is the cut image pixel;

[0064] After extracting the horizontal high-frequency coefficients and the vertical high-frequency coefficients, the boundary of the planned contour area is better characterized. By extracting the edge features of the cut image, an edge label difference data set is formed. By traversing and extracting the edge difference points, the cross entropy loss function E(·) is calculated as the objective function for cutting image edge difference extraction.

[0065]

[0066] Where N is the total number of edge cutting images extracted, p is the marked edge sample vector, k is the edge difference cutting image output vector calculated by the mean square error loss function K, j is the edge cutting image sequence number, ε is the weight of adjusting the edge difference sample, λ is the weight of adjusting the edge cutting image sample,

[0067] As the calculation condition of the global accuracy F,

[0068]

[0069] Among them, w d is the output weight of the pixel d extracted from the edge difference of the cut image, and f(N) is the number of edge difference cut images in all edge cut images N that have been calculated by the extracted objective function.

[0070] The output weight formula is w d =median(d) / d, where median(d) is the median value of the pixel point d extracted from the edge difference of the cutting image in the land planning contour area,

[0071] The global accuracy F is the proportion of images that successfully identify edge differences among all processed edge-cut images, while taking into account the role of these features in the final land collection decision.

[0072] By cutting the image of the land planning contour area, a multi-resolution fusion feature encoding is formed, so that the global accuracy F is sampled multiple times, and the edge differences of the analogs (buildings, green spaces, roads) marked in the cut image are judged through balanced calculation.

[0073] The land planning contour area is used to cut image samples. After accuracy evaluation, the pixel information loss caused by image stitching is performed. After multiple samplings, the continuity of semantic information in the restoration process is increased. The trained convolutional coding network model is used as a pre-training model to train the entire encoding-decoding network.

[0074] The satellite remote sensing images collected in real time are input into the trained DeepLab v3+ model, and the trained encoding-decoding network is converted into a network for testing satellite remote sensing images collected in real time. The loss layer and the precision layer are removed, and the Softmax layer is added to calculate the edge difference probability.

[0075] Where A is the probability of forming a z-type edge difference cut image, e is the base of the natural logarithm, and z is the original output value of the edge difference cut image belonging to the edge difference category after global accuracy evaluation;

[0076] The edge difference of the output cut image belongs to the probability of the planning contour area; by setting the judgment threshold, the probability map is converted into a binary image, that is, the pre-labeled planning contour area is obtained, thereby completing the land collection statistical operation.

[0077] Through the pre-trained edge difference cutting image, the satellite remote sensing image in the input prediction set is divided into parallel cutting image data, and the land feature edge difference analysis of the monitoring area is carried out based on the trained cutting image data to obtain the land feature edge difference prediction result corresponding to the monitoring area. By constructing and training the objective function of cutting image edge difference extraction and performing a global accuracy evaluation of the output image vector according to the time series, the probability that the edge difference of the output cutting image belongs to the planned contour area is obtained.

[0078] Intelligent land collection includes the series-connected mean square error loss function, cross entropy loss function, global accuracy evaluation and difference probability calculation. The purpose is to train the difference between land difference changes and surrounding adjacent spatial information through the corresponding loss function calculation, to help the real-time and reliability of land collection, and to make the collection results more accurate.

[0079] Carry out data governance work for data with quality problems according to business logic and related records, and formulate targeted business management processes based on the characteristics of data errors to ensure the credibility and availability of the quality of the newly added data, align the data governance results with the data catalog, and support the development of data database construction and collection. (Note: Data supplemented and improved through governance will not be used as a management basis before being recognized by the management department, and will only be used for data statistical analysis)

[0080] 1. Urban detailed planning management. In the case of missing or irregular filling of indicators such as land use nature, volume ratio, building height limit, green space ratio, and building density in the urban detailed planning land use planning master map, they shall be improved according to the sub-map indicators. Spatial errors such as self-intersection of spatial elements shall be repaired.

[0081] 2. Management of the scope lines of newly added construction land. For the missing attributes of the approval document number, approval time, and area of ​​newly added construction land in the scope lines of newly added construction land, the information in the new construction land ledger shall be compared to complete them. For the overlapping scope lines of newly added construction land, the scope lines shall be corrected in the order of approval time, and the scope lines approved later shall replace the scope lines approved earlier. For errors such as self-intersection of spatial elements, the spatial element repair tool shall be used to repair them.

[0082] 3. Management of land transfer scope lines and land allocation scope lines. For the missing attributes of land transfer scope lines and land allocation scope lines, such as the transfer and allocation time, contract number (allocation decision document number), transferee, and land nature, the information in the land transfer and land allocation ledgers shall be compared to complete them. For the overlap between the land transfer scope lines and the land allocation scope lines, the scope lines shall be corrected according to the order of transfer and allocation time, and the scope lines supplied later shall replace the scope lines supplied earlier. For the missing vectors of land transfer and land allocation scope lines, reference shall be made to the control planning plots, the land ownership scope lines of the real estate registration system, and the red lines of construction project land.

[0083] 4. Construction project land redline. Convert the collected construction project land redline DWG files issued in the past three years into GIS data, and assign the construction application number, project name, construction unit, document number, issuance date, land area and other information in the construction land planning permit or site selection opinion to the land redline. Compare the list of construction land planning permits and site selection opinions, and supplement the missing land redlines with reference to the control planning plots, land ownership scope lines of the real estate registration system, land transfer or land allocation scope lines, etc. Determine the validity of overlapping land redlines based on the order of issuance dates, and remove invalid land redlines.

[0084] 5. Real estate registration plots. In the case of overlapping plot boundary lines, the validity shall be determined by the order of obtaining the ownership date, and the invalid plot boundary lines shall be marked.

[0085] 6. Building outlines of the general plan of the construction project. Convert the building outlines in the DWG files of the general plan of the construction projects collected in the past three years into GIS data, assign the construction application number, project name, construction unit, document number, issuance date and other information in the construction project planning permit to the building outlines, and assign the building number, number of floors and other information in the marking to the building outlines. Compare the list of the construction project planning permit and supplement the missing building outlines with reference to the real estate registration house boundary line. Determine the validity of overlapping building outlines based on the order of issuance dates, and remove invalid building outlines.

[0086] 7. Construction project completion drawings. Convert the building boundary lines in the DWG files of the construction project completion drawings collected in the past three years into GIS data, and assign the construction application number, project name, construction unit, document number, issuance date, corresponding construction regulations certificate number and other information in the construction project completion planning verification confirmation to the building outline.

[0087] 8. Real estate registration houses. In the case of overlapping house building boundary lines, the validity is determined by the order of obtaining the ownership date, and the invalid house building boundary lines are marked. The house building boundary lines whose last registration date for all houses in the house building is before the date of signing the land transfer boundary line contract at the same geographical location, or the house building boundary lines whose last registration date for all houses in the house building is before the date of issuing the red line of the construction project land at the same geographical location, are marked as invalid.

[0088] 9. Dynamic management of construction land. Based on the well-managed urban planning plots, the dynamic GIS data of land management is formed by integrating the spatial data of newly added construction land boundary lines, land transfer boundary lines, land allocation boundary lines, real estate registration plots, construction project land red lines, construction project general plan building outlines, construction project completed building outlines, real estate registration houses, etc. The dynamic GIS data of land management contains all the attributes of the planning plots, and adds the implementation status attribute to record the planning implementation status of each plot.

[0089] Although the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

Claims

1. A system for intelligent land collection based on cloud platform map data, characterized in that: include: Strength calculation module, used to obtain spatial map data and calculate the edge strength of the planned contour area based on satellite remote sensing images; The accuracy evaluation module is used to perform global accuracy evaluation on the cut image after edge strength calculation; The land collection module is used to collect land attributes based on the evaluation results through probability calculation and upload them to the cloud platform.

2. The system for intelligently collecting land based on cloud platform map data according to claim 1 is characterized in that: The intensity calculation module comprises: Use the imread function to read the cut image to be processed, convert the cut image into a grayscale image, and store it in the candidate variable I. Then, use the RGB2gray function to convert the color image into a grayscale image for edge detection. The converted grayscale image is stored in the grayscale variable I. gray In the horizontal direction, the Sobel operator G x : Used to detect horizontal edges, vertical Sobel operator G y , used to detect vertical edges, using G x Convolve the cut image to get the horizontal edge intensity map, using G y Convolve the image to obtain the vertical edge intensity map. For each pixel in the cut image, calculate the grayscale gradient amplitude I gray , which is the square root of the sum of the squares of the edge strengths in the horizontal and vertical directions, The Sobel operator is applied to the grayscale image using the imfilter function, which performs a convolution operation where G x and G y They are used to detect edges in the horizontal and vertical directions respectively; x and I y are the edge strengths in the horizontal and vertical directions, Use the gradient Sobel operator to perform convolution operations on the cut image to obtain the gradient map of the cut image in the horizontal and vertical directions. The magnitude of the gradient reflects the rate of change of the image grayscale. The area with a larger gradient often corresponds to the edge details. Perform a two-dimensional wavelet transform: Perform a two-dimensional wavelet transform on the image to decompose it into sub-bands of different frequencies, which usually involves performing a one-dimensional wavelet transform on the rows and columns of the image respectively; In the result of two-dimensional wavelet transform, the horizontal high-frequency coefficient corresponds to the high-frequency component in the horizontal direction of the transformed image, and the vertical high-frequency coefficient corresponds to the high-frequency component in the vertical direction. The horizontal high-frequency coefficient and the vertical high-frequency coefficient are directly extracted from the result of wavelet transform.

3. The system for intelligently collecting land based on cloud platform map data according to claim 1 is characterized in that: The intensity calculation module comprises: Use the mean square error loss function K Among them, β is the adjustment weight, which is used to balance the importance of the two directions. For I x The edge strength of M pixels in the predicted horizontal direction and true lateral edge strength The mean of the sum of squared differences, For I y The edge strength of M pixels in the predicted longitudinal direction and true longitudinal edge strength The mean of the sum of squared differences, i is the cut image pixel.

4. The system for intelligently collecting land based on cloud platform map data according to claim 1 is characterized in that: The intensity calculation module comprises: By extracting the edge features of the cut image, an edge label difference dataset is formed. By traversing and extracting the edge difference points, the cross entropy loss function E(·) is calculated as the objective function for cutting image edge difference extraction Among them, N is the total number of edge cutting images extracted, p is the marked edge sample vector, k is the edge difference cutting image output vector calculated by the mean square error loss function K, j is the edge cutting image sequence number, ε is the weight of adjusting the edge difference sample, and λ is the weight of adjusting the edge cutting image sample.

5. The system for intelligently collecting land based on cloud platform map data according to claim 1 is characterized in that: The accuracy assessment module includes: As the calculation condition of the global accuracy F, Among them, w d is the output weight of the pixel d extracted from the edge difference of the cut image, and f(N) is the number of edge difference cut images calculated by the extraction objective function in all edge cut images N; The output weight formula is w d =median(d) / d, where median(d) is the median value of the pixel point d extracted from the edge difference of the cut image in the land planning contour area.

6. The system for intelligently collecting land based on cloud platform map data according to claim 1 is characterized in that: The land collection module includes: The image sampling is cut using the land planning contour area. After the accuracy evaluation, the pixel information loss caused by image splicing is analyzed. After multiple samplings, the continuity of semantic information in the restoration process is increased. The trained convolutional coding network model is used as a pre-training model to train the entire encoding-decoding network. The satellite remote sensing images collected in real time are input into the trained DeepLab v3+ model, and the trained encoding-decoding network is converted into a network for testing satellite remote sensing images collected in real time. The loss layer and the precision layer are removed, and the Softmax layer is added to calculate the edge difference probability. Where A is the probability of forming a z-type edge difference cut image, e is the base of the natural logarithm, and z is the original output value of the edge difference cut image belonging to the edge difference category after global accuracy evaluation; The edge difference of the output cut image belongs to the probability of the planning contour area; by setting the judgment threshold, the probability map is converted into a binary image, that is, the pre-labeled planning contour area is obtained, thereby completing the land collection statistical operation.

Citation Information

Patent Citations

  • Ploughing plot extraction method and system

    CN114529818A

  • Multi-scale road extraction model method for enhancing road edge information based on U-net

    CN114581665A

  • Small sample remote sensing image water body information extraction method

    CN116486273A

  • Coding and decoding indexed edge representation-based high-resolution remote sensing image semantic segmentation method

    CN116524189A

  • High-resolution remote sensing image semantic segmentation method based on graph convolutional neural network

    CN116597143A