Large-scale forest fragmentation classification algorithm based on geographic information cloud computing platform

Abstract

The invention relates to a large-scale forest fragmentation classification algorithm based on a geographic information cloud computing platform. The algorithm comprises the following steps: loading forest coverage data of a target area; generating a binarized forest mask image; constructing a neighborhood window; calculating a local forest coverage rate in the neighborhood window; extracting numerical values of all pixels in the neighborhood window, generating corresponding adjacency images and counting adjacency relations; calculating a local forest adjacency probability in the neighborhood window; calculating a local forest coverage rate and a local forest adjacency probability corresponding to each pixel; and determining the forest fragmentation type to which the pixel belongs. The beneficial effects of the invention are that the method achieves the efficient processing of a high-resolution and large-scale region based on the advantages of a geographic information cloud computing platform employing a multi-level cache and parallel operation strategy, breaks through the bottlenecks of a conventional algorithm in speed and scale, and effectively shortens the processing time.

Description

Technical Field

[0001] This invention belongs to the field of remote sensing technology, and in particular relates to a large-scale forest fragmentation classification algorithm based on a geographic information cloud computing platform. Background Technology

[0002] Forest fragmentation refers to the process by which large, continuous forest landscapes are divided into smaller, isolated patches. It is typically caused by human activities (such as deforestation, agricultural expansion, and urbanization) or natural disasters (such as fires), and has a profound impact on biodiversity, carbon storage, and ecosystem function. Traditional methods for monitoring and analyzing forest fragmentation rely on low- to medium-resolution satellite imagery and localized ground surveys. These methods suffer from problems such as the conflict between data timeliness and spatial resolution, insufficient fusion of multi-source data (such as optical / microwave remote sensing), and low efficiency in long-term series analysis. Furthermore, assessment indicators are often limited to area changes, lacking a comprehensive quantification of landscape patterns and ecological processes, making it difficult to achieve efficient and accurate monitoring globally.

[0003] In recent years, with the rapid development of remote sensing big data and cloud computing technologies, geographic information cloud computing platforms, with their petabyte-level remote sensing data storage capabilities (covering Landsat, Sentinel, and other satellite data series) and distributed cloud computing frameworks, have provided massive amounts of remote sensing data and high-performance computing capabilities, offering new technical means for global forest fragmentation monitoring. However, current applications based on geographic information cloud computing platforms mostly focus on detecting forest cover change, lacking a systematic algorithm framework for identifying and classifying forest fragmentation pattern characteristics. Especially at large spatial and multi-temporal scales, how to fully extract structural information from remote sensing images and construct stable and transferable fragmentation classification methods remains a key issue that urgently needs to be addressed.

[0004] Therefore, developing a large-scale forest fragmentation classification algorithm based on geographic information cloud computing is not only of great scientific significance, but can also provide a key decision support tool for global and regional sustainable forest management. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a large-scale forest fragmentation classification algorithm based on a geographic information cloud computing platform.

[0006] Firstly, a large-scale forest fragmentation classification algorithm based on a geographic information cloud computing platform is provided, including:

[0007] Step 1: Load the forest cover data of the target area into the geographic information cloud computing platform;

[0008] Step 2: Based on the forest cover data, generate a binarized forest mask image;

[0009] Step 3: Construct a neighborhood window of size K×K, where K is an odd number;

[0010] Step 4: Take the first pixel in the forest mask image as the center point of the neighborhood window, and calculate the local forest coverage rate within the neighborhood window;

[0011] Step 5: Extract the values ​​of all pixels within the neighborhood window, generate the corresponding adjacency image, and count the adjacency relationships;

[0012] Step 6: Calculate the local forest adjacency probability within the neighborhood window based on the adjacency relationship;

[0013] Step 7: Traverse the forest mask image and calculate the local forest coverage and local forest adjacency probability for each pixel;

[0014] Step 8: Based on the numerical relationship between the local forest cover rate and the local forest adjacency probability corresponding to each pixel, and according to the preset multi-level condition threshold, determine the forest fragmentation type to which the pixel belongs.

[0015] Preferably, in step 4, the formula for calculating the local forest coverage rate is:

[0016]

[0017] in, Let f(x+i, y+j) represent the local forest cover rate, K represent the window size, and r represent the window radius. f(x+i, y+j) represents the binary attribute value of the neighboring cell with an offset of (i, j) within the window. x and y represent the coordinates of the center cell of the moving window, and i and j represent the row and column offsets of the cells inside the window relative to the center cell, respectively.

[0018] Preferably, in step 6, the formula for calculating the local forest adjacency probability is:

[0019]

[0020] in, This represents the local forest adjacency probability. This indicates the frequency of horizontally adjacent pixels being part of a forest. This indicates the frequency of vertically adjacent pixels belonging to the forest. This indicates the frequency of at least one pixel representing a forest in the horizontal direction. This indicates the frequency of at least one pixel representing a forest in the vertical direction.

[0021] Preferably, in step 5, the values ​​of all pixels within the neighborhood window are extracted using a neighborhood-to-multi-band function or a neighborhood-to-array function.

[0022] Preferably, step 8 includes:

[0023] A first threshold, a second threshold, and a third threshold are set, wherein the first threshold is greater than the third threshold, and the third threshold is greater than the second threshold;

[0024] according to Whether it equals the first threshold determines whether it is an inner forest;

[0025] according to Whether it is less than the second threshold determines whether it is a patchy forest;

[0026] according to Whether it is greater than or equal to the second threshold and less than the third threshold determines whether it is a transition forest;

[0027] according to Whether it is greater than or equal to the third threshold, and in combination with and If the difference is greater than zero, determine whether it is a hole forest;

[0028] according to Whether it is greater than or equal to the third threshold, and in combination with and If the difference is less than zero, it is determined whether it is a fringe forest.

[0029] Preferably, in step 8, for pixels that meet the preset uncertain conditions, the forest fragmentation type is further verified by combining auxiliary terrain data; the preset uncertain conditions include: Greater than or equal to the third threshold, and equal .

[0030] Secondly, a large-scale forest fragmentation classification system based on a geographic information cloud computing platform is provided for performing any of the methods described in the first aspect, including:

[0031] The loading module is used to load forest cover data for a target area into the geographic information cloud computing platform.

[0032] The generation module is used to generate a binarized forest mask image based on the forest cover data;

[0033] The building module is used to construct a neighborhood window of size K×K, where K is an odd number;

[0034] The first calculation module is used to take the first pixel in the forest mask image as the center point of the neighborhood window and calculate the local forest coverage rate within the neighborhood window.

[0035] The extraction module is used to extract the values ​​of all pixels within the neighborhood window, generate the corresponding adjacency image, and count the adjacency relationships.

[0036] The second calculation module calculates the local forest adjacency probability within the neighborhood window based on the adjacency relationship.

[0037] The traversal module is used to traverse the forest mask image and calculate the local forest coverage and local forest adjacency probability for each pixel.

[0038] The determination module is used to determine the forest fragmentation type of a pixel based on the numerical relationship between the local forest cover rate and the local forest adjacency probability corresponding to each pixel, and according to preset multi-level condition thresholds.

[0039] Thirdly, a computer storage medium is provided, wherein a computer program is stored therein; when the computer program is run on a computer, the computer causes the computer to perform any of the methods described in the first aspect.

[0040] Fourthly, an electronic device is provided, comprising:

[0041] Memory, used to store computer programs;

[0042] A processor for executing the computer program to implement the method as described in any of the first aspects.

[0043] The beneficial effects of this invention are:

[0044] 1. This invention supports the selection of convolution kernels (neighborhood windows) with larger odd-numbered sizes, which improves the flexibility of spatial feature extraction and classification accuracy.

[0045] 2. Based on the advantages of the geographic information cloud computing platform, this invention adopts a multi-level caching and parallel computing strategy to achieve efficient processing of high-resolution, large-scale areas, breaking through the bottlenecks of traditional algorithms in terms of speed and scale, and effectively shortening the processing time.

[0046] 3. This invention fully utilizes the ability of geographic information cloud computing to process image archives instantly, and can be easily extended to long-term series analysis when the image dataset has enough years.

[0047] 4. This invention can directly call up massive, multi-source remote sensing images (such as Landsat and Sentinel series) archived by the geographic information cloud computing platform, and seamlessly integrate data acquisition, preprocessing (such as cloud masking and atmospheric correction) with core analysis algorithms, which greatly reduces the pre-research preparation costs and technical barriers. Attached Figure Description

[0048] Figure 1 A flowchart of a large-scale forest fragmentation classification algorithm based on a geographic information cloud computing platform provided for this invention;

[0049] Figure 2 This is a schematic diagram illustrating the calculation of forest coverage and forest adjacency probability within a neighborhood window provided by the present invention.

[0050] Figure 3 This invention provides a schematic diagram of forest fragmentation classification using the state of Veracruz, Mexico in 2000 as an example. Detailed Implementation

[0051] The present invention will be further described below with reference to embodiments. The description of the embodiments below is only for the purpose of helping to understand the present invention. It should be noted that those skilled in the art can make several modifications to the present invention without departing from the principle of the present invention, and these improvements and modifications also fall within the protection scope of the claims of the present invention.

[0052] Example 1:

[0053] Existing technologies, such as Guidos Toolbox software, can identify fragmented categories at a smaller regional scale. However, such software typically has a hard limitation on the pixel size of the input image, making it difficult to obtain classification results at a larger regional scale at high resolution.

[0054] To address the problems of low efficiency and excessive time consumption caused by the massive amount of remote sensing data in existing technologies, Embodiment 1 of this application provides a large-scale forest fragmentation classification algorithm based on a geographic information cloud computing platform, such as... Figure 1 As shown, it includes:

[0055] Step 1: Load the forest cover data of the target area into the geographic information cloud computing platform.

[0056] Specifically, this step involves analyzing one or more phases of remote sensing image data covering the target area, and using image classification or information extraction techniques to identify and generate raster data that characterizes the spatial distribution range of the forest.

[0057] For example, this embodiment is based on the HansenGlobal Forest Change v1.12 (2000–2024) forest change dataset built into the geographic information cloud computing platform. Specifically, the "treecover2000" band is extracted from this dataset as remote sensing image data representing global forest cover in 2000. Depending on research or application needs, a forest coverage threshold is set, for example, a threshold of 30%, to determine whether a pixel is classified as a forest area.

[0058] Step 2: Based on the forest cover data, generate a binarized forest mask image.

[0059] In step 2, the forest mask imagery serves as input data for subsequent forest fragmentation analysis. For example, based on the forest cover threshold set in step 1, a threshold segmentation operation is performed on the remote sensing image data to generate a binarized forest mask imagery with the same spatial resolution as the original image. In the binarized mask, a pixel value of "1" indicates a forest area, and a pixel value of "0" indicates a non-forest area. This binarized mask is used as input data for subsequent forest fragmentation analysis.

[0060] Step 3: Construct a neighborhood window of size K×K, where K is an odd number.

[0061] In step 3, the square convolution kernel function (Kernel.square) provided by the geographic information cloud computing platform is called to create a square convolution kernel with a preset window size, which is used to construct a local neighborhood analysis window. The window size is set to an odd number of pixel values, and the kernel radius is half the window size rounded down. For example, when the window size is 5×5 pixels, the corresponding kernel radius is 2 pixels. This neighborhood window is used to define the range of pixels that each target pixel participates in the analysis within the local space.

[0062] Step 4: Take the first pixel in the forest mask image as the center point of the neighborhood window, and calculate the local forest coverage rate within the neighborhood window.

[0063] In step 4, based on the neighborhood window constructed in step 3, the first pixel in the binarized mask image selected in step 2 is used as the center point of the window, and statistical analysis is performed on the neighborhood space. The local forest cover is obtained by calculating the proportion of forest pixels within the neighborhood of this pixel. The formula for calculating the local forest cover is:

[0064]

[0065] in, The local forest coverage rate is represented by the proportion of forest pixels covering the analysis window. K represents the window size and r represents the window radius. f(x+i, y+j) represents the binary attribute value of the neighboring pixel with an offset of (i, j) within the window. x and y represent the coordinates of the center pixel of the moving window, and i and j represent the row and column offsets of the pixels inside the window relative to the center pixel, respectively.

[0066] Step 5: Extract the values ​​of all pixels within the neighborhood window, generate the corresponding adjacency image, and count the adjacency relationships.

[0067] Step 6: Calculate the local forest adjacency probability within the neighborhood window based on the adjacency relationship.

[0068] In step 6, all adjacent pixel pairs along the horizontal and vertical directions within the established neighborhood window are traversed, and the frequencies of dual-forest pixel pairs and single-forest pixel pairs in each direction are counted. The frequencies of dual-forest pixel pairs and single-forest pixel pairs obtained in the two directions are summed, and then the total frequency of dual-forest pixel pairs is divided by the total frequency of single-forest pixel pairs to calculate the local forest adjacency probability. The formula for calculating the local forest adjacency probability is as follows:

[0069]

[0070] in, This represents the local forest adjacency probability, which is the conditional probability that a given forest pixel is adjacent to another forest pixel within a window. This indicates the frequency of horizontally adjacent pixels being forests (forest-forest pixel pairs). This indicates the frequency of vertically adjacent pixels belonging to the forest. This indicates the frequency of at least one pixel in the horizontal direction belonging to a forest (forest-non-forest and forest-forest pixel pairs). This indicates the frequency of at least one pixel representing a forest in the vertical direction.

[0071] Step 7: Traverse the forest mask image and calculate the local forest coverage and local forest adjacency probability for each pixel.

[0072] Step 8: Based on the numerical relationship between the local forest cover rate and the local forest adjacency probability corresponding to each pixel, and according to the preset multi-level condition threshold, determine the forest fragmentation type to which the pixel belongs.

[0073] Example 2:

[0074] Based on Example 1, Example 2 of this application provides a more specific large-scale forest fragmentation classification algorithm based on a geographic information cloud computing platform, including:

[0075] Step 1: Load the forest cover data of the target area into the geographic information cloud computing platform.

[0076] In step 1, a remote sensing image dataset (including but not limited to optical images such as Landsat and Sentinel-2, or radar data such as ALOS and PALSAR) of the target area is loaded into the geographic information cloud computing platform. Forest cover information is then quickly extracted using vegetation index threshold segmentation or machine learning classification methods. Alternatively, high-quality forest products built into the platform or uploaded by users can be used directly.

[0077] Step 2: Based on the forest cover data, generate a binarized forest mask image.

[0078] Step 3: Construct a neighborhood window of size K×K, where K is an odd number.

[0079] In step 3, the size of the neighborhood window can be adaptively adjusted according to the resolution of the remote sensing image dataset to ensure the rationality of spatial context information extraction. For example, a larger window (such as 21×21) can be used for a 10-meter resolution Sentinel-2 image, while a medium window (such as 9×9 or 15×15) can be used for a 30-meter resolution Landsat image.

[0080] Step 4: Take the first pixel in the forest mask image as the center point of the neighborhood window, and calculate the local forest coverage rate within the neighborhood window.

[0081] Step 5: Extract the values ​​of all pixels within the neighborhood window, generate the corresponding adjacency image, and count the adjacency relationships.

[0082] In step 5, within the geographic information cloud computing platform, a built-in function for generating neighborhood pixel data is invoked to calculate the adjacency relationships between pixels. Specifically, by analyzing the types of adjacent pixels within a preset neighborhood, adjacent pixel pairs in dual forests (forest-forest) and single forests (forest-non-forest) are extracted, and their adjacency relationships are statistically analyzed.

[0083] For example, one feasible implementation is to call the neighborhoodToBands() function. This method reconstructs the pixels within the window into multi-band data, thus easily extracting horizontal (row-wise) and vertical (column-wise) adjacent pixel pairs. Subsequently, Boolean algebra operations are performed on the extracted pixel pairs to achieve accurate statistics on various adjacency relationships.

[0084] Neighborhood-to-multiband functions are intuitive and efficient when processing small windows (such as 5x5) because they can directly map neighboring pixels to bands. However, as the neighborhood window size increases, the number of bands generated by this method increases dramatically, leading to huge computational overhead and easily exceeding the processing limits of geographic information cloud computing platforms.

[0085] To address this issue, another feasible approach is to use a pre-defined neighborhood window. By calling the neighborhood-to-array function (neighborhoodToArray()) built into the geographic information cloud computing platform, the values ​​of all pixels within the window are extracted to generate the corresponding neighboring image (multidimensional array image). Subsequently, based on Boolean logic operations between the neighboring image and the original image, the relationships between adjacent pixels are accurately calculated. This method stores the neighboring pixels as an array within a single band. By performing operations on this array, all neighboring pixel values ​​can be accessed. Furthermore, this method offers better scalability and can stably and efficiently support neighborhood analysis for windows of arbitrary sizes, with particularly significant performance advantages when handling medium or large windows.

[0086] Step 6: Calculate the local forest adjacency probability within the neighborhood window based on the adjacency relationship.

[0087] Step 7: Traverse the forest mask image and calculate the local forest coverage and local forest adjacency probability for each pixel.

[0088] In step 7, this embodiment of the application utilizes the parallel computing capabilities of the geographic information cloud computing platform, using a sliding window of the same size as the neighborhood window to perform sliding window processing on all pixels of the image called in step 1, and calculates the forest cover probability in the neighborhood range with each pixel as the center. ) and adjacency probability ( ).

[0089] Furthermore, neighborhood calculation can also be achieved using the following alternative functions:

[0090] Displacement function: Generate a neighborhood image that is displaced by one pixel in both the horizontal and vertical directions using the translate() function, and then calculate the statistics;

[0091] Convolution operation function: Generate convolution kernels using the convolve() function, define kernels in the horizontal and vertical directions, and then calculate statistics.

[0092] Step 8: Based on the numerical relationship between the local forest cover rate and the local forest adjacency probability corresponding to each pixel, and according to the preset multi-level condition threshold, determine the forest fragmentation type to which the pixel belongs.

[0093] In step 8, this application determines the process based on each sliding window. and The numerical differences are used to set multi-level condition thresholds based on the permeation theory proposed by Broadbent and Hammersley to determine the forest fragmentation type of each pixel.

[0094] Specifically, step 8 includes:

[0095] A first threshold, a second threshold, and a third threshold are set, wherein the first threshold is greater than the third threshold, and the third threshold is greater than the second threshold;

[0096] according to Whether it equals the first threshold determines whether it is an inner forest;

[0097] according to Whether it is less than the second threshold determines whether it is a patchy forest;

[0098] according to Whether it is greater than or equal to the second threshold and less than the third threshold determines whether it is a transition forest;

[0099] according to Whether it is greater than or equal to the third threshold, and in combination with and If the difference is greater than zero, determine whether it is a hole forest;

[0100] according to Whether it is greater than or equal to the third threshold, and in combination with and If the difference is less than zero, it is determined whether it is a fringe forest.

[0101] Furthermore, for pixels that meet preset uncertain conditions, the forest fragmentation type is further verified by combining auxiliary terrain data; the preset uncertain conditions include: Greater than or equal to the third threshold, and equal .

[0102] For example, the first threshold is 1, the third threshold is 0.6, and the second threshold is 0.4. In this case, step 8 includes:

[0103] Internal Forest: When P_f = 1, it indicates that the entire window is a forest, and is determined to be a continuous and complete internal forest region;

[0104] Patch forest: When P_f < 0.4, it indicates low forest cover and is judged as isolated small patches with low connectivity;

[0105] Transitional forest: When 0.4 ≤ P_f < 0.6, it indicates that the forest coverage is moderate and is judged as a transitional region with moderate connectivity;

[0106] Hole forest: When P_f≥0.6 and P_f-P_ff>0, it indicates a high forest coverage and relatively high aggregation, but there are non-forest gaps inside, which are identified as hole areas;

[0107] Marginal forest: When P_f≥0.6 and P_f-P_ff<0, it indicates that the forest coverage is high but the aggregation is relatively low. The forest edge and the external non-forest area are interspersed, and it is judged as a boundary transition zone.

[0108] Undetermined Areas: When P_f ≥ 0.6 and P_f = P_ff, it indicates high forest coverage and balanced connectivity and clustering, and is initially identified as an undetermined area. First, the total number of pixels in the undetermined area or its proportion within the study area is counted. If the number of pixels in the undetermined area is extremely small or its proportion is below a preset ignore threshold (e.g., less than 0.1%), it will not be further classified. If the number or proportion of pixels in the undetermined area meets preset conditions, then geographic environmental feature data needs to be overlaid for secondary verification. The geographic environmental feature data includes, but is not limited to: topographic data (e.g., slope, altitude), distance factors (e.g., distance to roads, distance to settlements), or land cover features (e.g., soil type, hydrological distribution). In this embodiment, slope (S) and altitude (E) are selected as verification indicators. The specific judgment logic is as follows: (1) Obtain the slope value S and altitude value E of the undetermined area pixel; (2) Compare the slope value S and altitude value E with the preset slope threshold (S_th) and altitude threshold (E_th) respectively; (3) If the value indicates that the area is significantly restricted by the terrain (in this embodiment, S>S_th or E>E_th), it indicates that the area may be located at natural geographical boundaries such as ridges and cliffs. Although the coverage is high, it belongs to the transition zone. It is corrected and judged as a marginal forest; (4) If the value indicates that the terrain of the area is flat and easily disturbed (in this embodiment, S≤S_th or E≤E_th), it indicates that the area is easily penetrated by human activities. Under high coverage, it may contain potential man-made gaps. It is corrected and judged as a hole forest.

[0109] It should be noted that the parts in this embodiment that are the same as or similar to those in Embodiment 1 can be referred to each other, and will not be repeated in this application.

[0110] Example 3:

[0111] Based on Example 2, Example 3 of this application provides a large-scale forest fragmentation classification system based on a geographic information cloud computing platform, including:

[0112] The loading module is used to load forest cover data for a target area into the geographic information cloud computing platform.

[0113] The generation module is used to generate a binarized forest mask image based on the forest cover data;

[0114] The building module is used to construct a neighborhood window of size K×K, where K is an odd number;

[0115] The first calculation module is used to take the first pixel in the forest mask image as the center point of the neighborhood window and calculate the local forest coverage rate within the neighborhood window.

[0116] The extraction module is used to extract the values ​​of all pixels within the neighborhood window, generate the corresponding adjacency image, and count the adjacency relationships.

[0117] The second calculation module calculates the local forest adjacency probability within the neighborhood window based on the adjacency relationship.

[0118] The traversal module is used to traverse the forest mask image and calculate the local forest coverage and local forest adjacency probability for each pixel.

[0119] The determination module is used to determine the forest fragmentation type of a pixel based on the numerical relationship between the local forest cover rate and the local forest adjacency probability corresponding to each pixel, and according to preset multi-level condition thresholds.

[0120] It should be noted that the system provided in this embodiment is the corresponding system of the method provided in embodiment 2. Therefore, the parts that are the same as or similar to those in embodiment 2 in this embodiment can be referred to each other, and will not be described again in this application.

[0121] In summary, this invention, by introducing a convolution kernel size adjustment mechanism, can adapt to the spatial resolution differences of multi-source remote sensing data (e.g., Sentinel-2 data with 10-meter resolution and Landsat data with 30-meter resolution), thus improving the applicability of the method while ensuring processing accuracy. Simultaneously, relying on the distributed cloud computing capabilities of the geographic information cloud computing platform, it achieves pixel-level parallel processing of remote sensing imagery covering the entire region, effectively overcoming the computational bottleneck problem of traditional methods when processing large-scale data. This significantly improves the efficiency and accuracy of classification processing, providing an efficient and scalable technical solution for large-scale forest fragmentation analysis and dynamic change monitoring.

Claims

1. A large-scale forest fragmentation classification algorithm based on a geographic information cloud computing platform, characterized in that, include: Step 1: Load the forest cover data of the target area into the geographic information cloud computing platform; Step 2: Based on the forest cover data, generate a binarized forest mask image; Step 3: Construct a neighborhood window of size K×K, where K is an odd number; Step 4: Take the first pixel in the forest mask image as the center point of the neighborhood window, and calculate the local forest coverage rate within the neighborhood window; Step 5: Extract the values ​​of all pixels within the neighborhood window, generate the corresponding adjacency image, and count the adjacency relationships; Step 6: Calculate the local forest adjacency probability within the neighborhood window based on the adjacency relationship; Step 7: Traverse the forest mask image and calculate the local forest coverage and local forest adjacency probability for each pixel; Step 8: Based on the numerical relationship between the local forest cover rate and the local forest adjacency probability corresponding to each pixel, and according to the preset multi-level condition threshold, determine the forest fragmentation type to which the pixel belongs.

2. The large-scale forest fragmentation classification algorithm based on a geographic information cloud computing platform according to claim 1, characterized in that, In step 4, the formula for calculating the local forest coverage rate is: ； in, Let f(x+i, y+j) represent the local forest cover rate, K represent the window size, r represent the window radius, f(x+i, y+j) represent the binary attribute value of the neighboring cell with offset (i, j) within the window, x and y represent the coordinates of the center cell of the moving window, and i and j represent the row and column offsets of the cells within the window relative to the center cell, respectively.

3. The large-scale forest fragmentation classification algorithm based on a geographic information cloud computing platform according to claim 2, characterized in that, In step 6, the formula for calculating the local forest adjacency probability is: ； in, This represents the local forest adjacency probability. This indicates the frequency of horizontally adjacent pixels being part of a forest. This indicates the frequency of vertically adjacent pixels belonging to the forest. This indicates the frequency of at least one pixel representing a forest in the horizontal direction. This indicates the frequency of at least one pixel representing a forest in the vertical direction.

4. The large-scale forest fragmentation classification algorithm based on a geographic information cloud computing platform according to claim 3, characterized in that, In step 5, the values ​​of all pixels within the neighborhood window are extracted using a neighborhood-to-multi-band function or a neighborhood-to-array function.

5. The large-scale forest fragmentation classification algorithm based on a geographic information cloud computing platform according to claim 4, characterized in that, Step 8 includes: A first threshold, a second threshold, and a third threshold are set, wherein the first threshold is greater than the third threshold, and the third threshold is greater than the second threshold; according to Whether it equals the first threshold determines whether it is an inner forest; according to Whether it is less than the second threshold determines whether it is a patchy forest; according to Whether it is greater than or equal to the second threshold and less than the third threshold determines whether it is a transition forest; according to Whether it is greater than or equal to the third threshold, and in combination with and If the difference is greater than zero, determine whether it is a hole forest; according to Whether it is greater than or equal to the third threshold, and in combination with and If the difference is less than zero, it is determined whether it is a fringe forest.

6. The large-scale forest fragmentation classification algorithm based on a geographic information cloud computing platform according to claim 4, characterized in that, In step 8, for pixels that meet the preset undetermined conditions, the forest fragmentation type is verified a second time by combining auxiliary terrain data. The preset undetermined conditions include: Greater than or equal to the third threshold, and equal .

7. A large-scale forest fragmentation classification system based on a geographic information cloud computing platform, characterized in that: For performing the method as described in any one of claims 1 to 6, comprising: The loading module is used to load forest cover data for a target area into the geographic information cloud computing platform. The generation module is used to generate a binarized forest mask image based on the forest cover data; The building module is used to construct a neighborhood window of size K×K, where K is an odd number; The first calculation module is used to take the first pixel in the forest mask image as the center point of the neighborhood window and calculate the local forest coverage rate within the neighborhood window. The extraction module is used to extract the values ​​of all pixels within the neighborhood window, generate the corresponding adjacency image, and count the adjacency relationships. The second calculation module calculates the local forest adjacency probability within the neighborhood window based on the adjacency relationship. The traversal module is used to traverse the forest mask image and calculate the local forest coverage and local forest adjacency probability for each pixel. The determination module is used to determine the forest fragmentation type of a pixel based on the numerical relationship between the local forest cover rate and the local forest adjacency probability corresponding to each pixel, and according to preset multi-level condition thresholds.

8. A computer storage medium, characterized in that, The computer storage medium stores a computer program; when the computer program is run on the computer, it causes the computer to perform the method described in any one of claims 1 to 7.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the method as described in any one of claims 1 to 7.

Images