An Adaptive Threshold Optimization Method and System for the Division of the Wild-City Junction Region

Abstract

The present invention belongs to the cross field of remote sensing image processing and disaster risk management, and provides an adaptive threshold optimization method and system for wildland-urban interface (WUI) zoning. The technical solution is as follows: calculate the building density according to the building segmentation result; calculate the vegetation coverage, the wildland patch area and the distance measure respectively; determine the WUI determination rule, and determine the type of WUI in combination with the determination rule; set the threshold ranges of the building density, the vegetation coverage, the wildland patch area and the buffer distance measure, and calculate the WUI areas of different types of WUI under different thresholds; combine the historical burned area data and the WUI area to define the fire exposure index, use the spatial distribution characteristics of the WUI area as the independent variable of the random forest (RF) model, and the fire exposure index under different thresholds as the dependent variable of the RF model, and fit to obtain the optimal threshold, effectively solving the technical bottlenecks such as low accuracy of element extraction, single quantification standard and strong subjectivity in threshold setting in traditional WUI identification.

Description

Technical Field

[0001] The present invention belongs to the cross - field of remote sensing image processing and disaster risk management, and particularly relates to an adaptive threshold optimization method and system for wildland - urban interface zoning. Background Technique

[0002] The statements in this part only provide background technical information related to the present invention and do not necessarily constitute prior art.

[0003] The Wildland - Urban Interface (WUI) is a transitional zone between natural ecosystems and human urban development, with important ecological sensitivity. Due to the close contact or mixing of natural vegetation and human - used land in this area, it has become a high - incidence area of forest fires, posing a serious threat to the safety of residents. At the same time, rich natural resources, cultural resources, and high - value ecological environment assets are distributed in the WUI and its surrounding areas. With climate change and frequent forest fires, these resources and assets are facing increasingly severe survival threats.

[0004] The wildland - urban interface in the urban - rural fringe shows a significant expansion trend. On the one hand, the rapid expansion of urban space has led to the close contact or mixing of high - frequency land uses such as communities, enterprises, and facilities with vegetation such as forests, grasslands, and shrubs, forming a mosaic landscape structure; on the other hand, the continuous implementation of greening projects and ecological protection projects has effectively improved the vegetation coverage level in urban and surrounding areas, providing the necessary ecological support for the expansion of the wildland - urban interface. This development trend has led to significant distribution characteristics of the wildland - urban interface: high building density, large area of combustible vegetation coverage, small and scattered natural vegetation patches. These characteristics make the frequently occurring forest fires in the region show obvious regional aggregation characteristics and dynamic change trends.

[0005] Currently, the prior art has established a relatively complete set of wildland - urban interface delineation standards and fire risk assessment methods, and has respectively established a delineation system based on regional division (WUI - Z) and a delineation system based on point - by - point division (WUI - P); however, it does not consider the particularity of natural - human landscapes in different regions, resulting in significant differences in the WUI distribution pattern and dynamic changes. Summary of the Invention

[0006] In order to solve at least one of the technical problems in the above - mentioned background technique, the present invention provides an adaptive threshold optimization method and system for wildland - urban interface zoning, which takes into account the actual characteristics of different regions and innovatively constructs a technical framework integrating "high - precision element extraction - multi - dimensional index quantification - dynamic threshold optimization", effectively solving technical bottlenecks such as low accuracy of element extraction, single quantification standard, and strong subjectivity in threshold setting in traditional WUI recognition.

[0007] To achieve the above object, the present invention adopts the following technical solutions:

[0008] The first aspect of the present invention provides an adaptive threshold optimization method for wildland-urban interface (WUI) zoning, including the following steps:

[0009] Segment the image of the target area to obtain the building segmentation result;

[0010] Calculate the building density according to the building segmentation result;

[0011] Based on the obtained vegetation dataset, screen the wildland vegetation types, and calculate the vegetation coverage, wildland patch area, and distance measure respectively;

[0012] Determine the WUI determination rule, and combine the building density, vegetation coverage, wildland patch area, buffer distance measure, and determination rule to determine the type of WUI;

[0013] Set the threshold ranges of the building density, vegetation coverage, wildland patch area, and buffer distance measure, and calculate the WUI areas of different types of WUI under different thresholds;

[0014] Combine the historical burned area data and the WUI area to define the fire exposure index. Take the spatial distribution characteristics of the WUI area as the independent variable of the RF model, and the fire exposure index under different thresholds as the dependent variable of the RF model, and fit to obtain the optimal threshold.

[0015] Further, the segmenting the image of the target area to obtain the building segmentation result includes:

[0016] Pre-train the ResNet-50 backbone network based on the source domain dataset to obtain general image features;

[0017] Transfer the learned general image features to the target domain task;

[0018] Perform cross-stage feature connection between the corresponding layers of the ResNet-50 backbone network and the U-Net decoder, and realize the dynamic weighted fusion of multi-scale features through the spatial attention mechanism. Obtain the building segmentation result based on the fused features.

[0019] Further, the calculating the building density according to the building segmentation result includes:

[0020] Convert the building raster data corresponding to the building segmentation result into vector surface data;

[0021] Generate spatial point features representing buildings based on the vector surface data;

[0022] Based on spatial point features, perform spatial density calculation. Construct buffers with each building point as the center, and count the building distribution density around each analysis point.

[0023] Furthermore, screen wildland vegetation types based on the obtained vegetation dataset, and calculate the vegetation coverage, wildland patch area, and distance measure respectively, including:

[0024] Use the moving window statistical method to count the number of wildland vegetation pixels in the moving window pixel by pixel, and obtain the regional wildland vegetation coverage by dividing the number of wildland vegetation pixels by the total number of pixels in the buffer;

[0025] Convert the screened wildland raster data into vector polygon data through GDAL, calculate the area of each vector patch, and screen wildland patches with an area larger than the set area to obtain the wildland patch area;

[0026] Generate n -level concentric ring buffers along the wildland patch boundary, and obtain the distance measure according to the set buffer distance and level threshold.

[0027] Furthermore, the threshold ranges of building density, vegetation coverage, wildland patch area, and buffer distance measure include:

[0028] For building density, the threshold range is set from 1 per square kilometer to the maximum building density, with a step size of 5 per square kilometer;

[0029] For vegetation coverage, the threshold range is set from 0% to 100%, with a gradient step size of 1%;

[0030] For wildland patch area, the threshold range is set from 0.1 square kilometer to the maximum patch area, with a step size of 0.1 square kilometer;

[0031] For buffer distance measure, the threshold range is from 0.1 kilometer to 5.0 kilometers, with a step size of 0.1 kilometer.

[0032] Furthermore, calculate the WUI areas of different types of WUI under different thresholds, including:

[0033] The calculation process of the WUI area of the mixed WUI under different thresholds includes:

[0034] Generate candidate regions based on the building density threshold, extract all connected regions with a building density ≥ x per square kilometer, and then, overlay the vegetation coverage threshold, and extract sub-regions with a vegetation coverage ≥ y% within the candidate regions;

[0035] The calculation process of the WUI area of the mixed WUI under different thresholds includes:

[0036] First, candidate areas sharing the building density threshold with the hybrid-WUI are identified. Subsequently, subject to the reverse vegetation coverage constraint, sub-areas within the candidate areas with vegetation coverage < y% are extracted. In the third step, spatial correlation analysis of wildland patches is carried out to screen wildland patches with an area ≥ m square kilometers, and a buffer zone with a width of n kilometers is generated based on the boundaries of the wildland patches, and the candidate areas of the interface type WUI that intersect with the buffer zone are retained.

[0037] Further, the fire exposure index WEI is expressed as:

[0038] 。

[0039] The second aspect of the present invention provides an adaptive threshold optimization system for wildland-urban interface zoning, including:

[0040] A segmentation module for segmenting the image of the target area to obtain the building segmentation result;

[0041] A parameter calculation module for calculating the building density based on the building segmentation result; screening wildland vegetation types based on the obtained vegetation dataset, and respectively calculating the vegetation coverage, the area of wildland patches, and the distance measure;

[0042] A WUI type determination module for determining the WUI determination rule and determining the type of WUI in combination with the building density, vegetation coverage, wildland patch area, buffer zone distance measure, and the determination rule;

[0043] A WUI area determination module for setting the threshold ranges of the building density, vegetation coverage, wildland patch area, and buffer zone distance measure, and calculating the WUI areas of different types of WUI at different thresholds;

[0044] A threshold adaptive optimization module for combining historical burned area data and the WUI area to define the fire exposure index, using the spatial distribution characteristics of the WUI area as the independent variable of the RF model, and the fire exposure index at different thresholds as the dependent variable of the RF model, and fitting to obtain the optimal threshold.

[0045] The third aspect of the present invention provides a computer-readable storage medium.

[0046] A computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the steps in an adaptive threshold optimization method for wildland-urban interface zoning as described above are implemented.

[0047] The fourth aspect of the present invention provides a computer device.

[0048] A computer device includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements the steps in an adaptive threshold optimization method for wildland-urban interface zoning as described above.

[0049] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0050] The present invention utilizes the high-precision extraction (qualitative) of WUI elements and the adaptive quantization (quantitative) of WUI determination elements by means of deep learning and machine learning technologies. Thus, a technical framework integrating "high-precision element extraction - multi-dimensional index quantization - dynamic threshold optimization" is innovatively constructed, effectively solving the technical bottlenecks in traditional WUI recognition, such as low element extraction accuracy, single quantization standard, and strong subjectivity in threshold setting. The present invention is a high-resolution remote sensing image building semantic segmentation method based on transfer learning and improved U-Net. By constructing a cross-stage feature fusion architecture and a spatial attention mechanism, the building recognition accuracy in complex urban scenes is significantly improved. The pre-training strategy of the ResNet-50 backbone network on the ImageNet dataset, combined with the parameter freezing mechanism in transfer learning technology, enables the model to effectively extract general image features while avoiding overfitting problems. The improved dynamic weighted fusion formula reduces the positioning error of building edges through the spatial attention guidance of multi-scale features. At the same time, in the small-sample training scenario, a composite data augmentation strategy integrating random rotation, elastic deformation, and adversarial generated samples improves the generalization ability of the model and significantly reduces the annotation cost of training, providing feasible support for the extraction of WUI elements in large-scale regions. 3. The present invention is a hybrid and interface-type WUI impact factor quantization method based on the WUI-P model. By constructing a dynamic calculation system with adaptive spatial heterogeneity, it solves the applicability defects of the traditional static threshold method in complex terrain areas. The introduction of an initial buffer and a dynamic radius contraction mechanism for building density calculation effectively overcomes the sensitivity of the fixed window method to uneven spatial distribution. The calculation of vegetation coverage combined with existing datasets significantly reduces the amount of calculation data and reduces the dependence on remote sensing data. The composite constraint conditions (area threshold, shape index, spatial distribution) integrated in the wildland patch screening link lay a precise data foundation for subsequent fire risk modeling. The intelligent construction strategy of multi-level buffers through the annular gradient division mechanism makes the spatio-temporal coincidence degree of fire risk spread simulation better match the local fire risk situation compared with the traditional buffer.

[0051] 4. The technical solution of the present invention is a method for dynamically optimizing the WUI threshold based on random forest. By constructing a multi-dimensional parameter space and an intelligent decision-making system, it breaks through the subjective limitations of relying on manual experience in traditional threshold setting. By defining the spatio-temporal weighted fire exposure index (WEI) and integrating the spatial alignment and resolution matching mechanism of long-term historical burned area data, it innovatively dynamically correlates the spatial distribution characteristics of fire risk with threshold parameters. In the threshold decision-making link, by analyzing the partial dependence curve, the geographical spatial rationality and practical operability of the threshold scheme are significantly improved. It overcomes the timeliness lag problem of traditional static threshold schemes in different geographical scenarios.

[0052] Advantages of additional aspects of the present invention will be partly given in the following description, partly will become obvious from the following description, or will be understood through the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] The accompanying drawings forming a part of this specification are used to provide a further understanding of the present invention. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention.

[0054] Figure 1 is a flowchart of an adaptive threshold optimization method for wildland-urban interface zoning provided by an embodiment of the present invention;

[0055] Figure 2 is a schematic structural diagram of a building semantic segmentation recognition model based on transfer learning and U-Net network provided by an embodiment of the present invention;

[0056] Figure 3 is a schematic structural diagram of an improved U-Net network provided by an embodiment of the present invention;

[0057] Figure 4 is a schematic diagram of the existing standard WUI determination process based on the WUI-P model provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0058] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0059] It should be noted that the following detailed description is illustrative and is intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention belongs.

[0060] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular forms are also intended to include the plural forms. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they specify the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0061] Currently, the prior art has established a relatively complete set of standards for delimiting the wildland-urban interface (WUI) and methods for fire risk assessment. For example, the WUI is divided into two types of classification methods: mixed-WUI (building density ≥ 6.17 / km², combustible vegetation ≥ 50%) and interface-WUI (building density ≥ 6.17 / km², combustible vegetation < 50%, wildland patch area ≥ 5 km², distance measure ≤ 2.4 km), and a delimitation system based on regional division (WUI-Z) and a delimitation system based on point division (WUI-P) have been established respectively. However, it does not take into account the significant differences in the distribution pattern and dynamic changes of the WUI due to the particularity of the natural and human landscapes in different regions.

[0062] The present invention combines the actual characteristics of the area to be evaluated and establishes a scientific and practical method for calculating the dynamic threshold of the wildland-urban interface, providing technical support for the scientific delimitation of the wildland-urban interface, fire risk assessment, and regional planning and management.

[0063] Example 1

[0064] As Figure 1 shown, this embodiment provides an adaptive threshold optimization method for wildland-urban interface zoning, including the following steps:

[0065] Step 1: Obtain the source domain dataset, target domain dataset, and auxiliary data;

[0066] In this embodiment, the ImageNet dataset is used as the source domain dataset;

[0067] Among them, the process of obtaining the target domain dataset includes: determining the target area to be analyzed, for example, selecting the area with a high incidence of historical wildfires; obtaining the high-resolution remote sensing image data of the target area, and annotating the buildings in the high-resolution remote sensing image data to construct the target domain dataset;

[0068] In this embodiment, the high-resolution remote sensing image data with a resolution of 512×512 pixels is annotated, and the sample size is expanded to 5 times the original data through data augmentation means such as random rotation (±30°), mirror flipping, and Gaussian noise injection (σ = 0.01) to construct the target domain dataset in a set style;

[0069] The auxiliary data includes ASTER DEM, for example, with a resolution of 30 meters, which is used for building screening;

[0070] The ESA WorldCover vegetation dataset, such as at a 10-meter resolution, is used for building screening and wildland feature extraction;

[0071] A historical fire exposure dataset is constructed based on a fire exposure calculation mechanism, integrating the burned area data in the target area over the past 20 years, overlaying the burned area data, and unifying the spatial resolution to 10 meters; it is used for calculating the WUI threshold.

[0072] Step 2: Segment the images of the target area to obtain the building segmentation results;

[0073] As Figure 2 and Figure 3 shown, it specifically includes the following steps:

[0074] Specifically, it includes the following steps:

[0075] Step 201: Pre-train the ResNet-50 backbone network based on the source domain dataset to obtain general image features;

[0076] In this embodiment, the initial learning rate is set to 0.001 during the pre-training stage, the cosine annealing learning rate scheduling strategy is adopted, the batch size is 32, and the number of iterations is 100 epochs;

[0077] Step 202: Transfer the learned general image features to the target domain task, specifically including:

[0078] Freeze the parameters of the first three convolutional stages (stage1-stage3) of ResNet-50, only fine-tune stage4, and use the pre-trained ResNet-50 as the encoder of the improved U-Net. The multi-scale feature maps output by it are transferred to the decoder through cross-stage connections;

[0079] In this embodiment, the output of stage 1 (256 channels) is connected to the 4th layer of the decoder; the output of stage 2 (512 channels) is connected to the 3rd layer of the decoder; the output of stage 3 (1024 channels) is connected to the 2nd layer of the decoder; the output of stage 4 (2048 channels) is used as the input benchmark for the decoder;

[0080] Step 203: Perform cross-stage feature connections between the corresponding layers of the ResNet-50 backbone network and the U-Net decoder, and achieve dynamic weighted fusion of multi-scale features through a spatial attention mechanism. Its output features are directly used as the input basis for subsequent decoding upsampling; the formula is expressed as:

[0081] (1),

[0082] Among them, and represent the feature maps of the encoder and decoder respectively, σ represents the Sigmoid function, Conv is a 1×1 convolution operation, and ⊙ represents element-wise multiplication.

[0083] The Focal Loss dynamic loss function is used to optimize the model training process. Set the adjustment factor γ = 2 and the class balance parameter α = 0.25. Its mathematical expression is:

[0084] (2),

[0085] where, represents the model prediction probability value.

[0086] The attention weight matrix calculated by formula (1) and the Focal Loss function of formula (2) form a joint optimization objective. Among them, the spatial attention mechanism preferentially retains the significant feature regions related to buildings in the encoder, and the Focal Loss strengthens the gradient backpropagation of difficult samples (such as building edges) through the adjustment factor γ = 2.

[0087] Finally, the Intersection over Union (IoU), F1-score, and Overall Accuracy (OA) are used as three indicators to evaluate the model performance on the test set. Set the threshold judgment condition as IoU≥0.75 and F1-score≥0.85 as the model compliance standard.

[0088] The building semantic segmentation method for high-resolution remote sensing images based on transfer learning and improved U-Net significantly improves the building recognition accuracy in complex urban scenes by constructing a cross-stage feature fusion architecture and a spatial attention mechanism. The pre-training strategy of the ResNet-50 backbone network on the ImageNet dataset, combined with the parameter freezing mechanism in transfer learning technology, enables the model to effectively extract general image features while avoiding overfitting problems. The improved dynamic weighted fusion formula reduces the positioning error of building edges through the spatial attention guidance of multi-scale features. At the same time, in the small-sample training scenario, the composite data augmentation strategy that combines random rotation, elastic deformation, and adversarial generated samples improves the generalization ability of the model and significantly reduces the annotation cost of training, providing feasible support for the extraction of WUI elements in large-scale regions.

[0089] Step 3: Calculate the building density based on the building segmentation results;

[0090] Specifically, it includes the following steps:

[0091] Step 301: Vectorize the obtained building segmentation results;

[0092] In this embodiment, the building raster data obtained in step 2 is converted into vector surface data through the Polygonize function of the GDAL library;

[0093] Step 302: Generate spatial point features representing buildings based on the vector surface data;

[0094] In this embodiment, the geometric centroid coordinates are calculated according to the polygon vertex coordinates formed by the vector surface data, and the centroid points are projected onto the UTM coordinate system to generate spatial point features representing buildings.

[0095] Step 303: Based on the spatial point features, perform spatial density calculation, construct buffers with each building point as the center, and count the building distribution density around each analysis point.

[0096] In this embodiment, a buffer is constructed with an initial radius of 500 meters, and the building distribution density around each analysis point is counted.

[0097] Step 4: Screen wildland vegetation types based on the vegetation dataset, and calculate the vegetation coverage and wildland patch area based on the wildland vegetation types;

[0098] In this embodiment, for the vegetation coverage, wildland vegetation types are screened based on the ESA WorldCover 10-meter resolution data, including: forest land, shrubs, grasslands, herbaceous wetlands, mangroves, mosses, and lichens, etc. The moving window statistical method is used to count the number of wildland vegetation pixels in a 500 m × 500 m moving window pixel by pixel, and the ratio of the number of wildland vegetation pixels to the total number of pixels in the buffer area is used to obtain the regional wildland vegetation coverage;

[0099] For the wildland patch area, the selected wildland raster data is converted into vector surface data through GDAL, the areas of each vector patch are calculated, and wildland patches with an area greater than 0.1 square kilometers are screened;

[0100] For the distance measure, generate n concentric ring buffers of level, and the distance of this buffer depends on the threshold setting with the level.

[0101] Step 5: Determine the determination rules for the wildland-urban boundary region, and combine the building density, vegetation coverage, wildland patch area, buffer distance measure, and determination rules to determine the type of the wildland-urban boundary region;

[0102] In this embodiment, the determination rules for the wildland-urban boundary region adopt traditional determination rules, and the specific judgment process includes:

[0103] Judge whether the building density is greater than 6.17 / km². If so, further judge the vegetation coverage, otherwise it is a non-boundary region;

[0104] Judge whether the vegetation coverage is greater than 50%. If so, the buffer zone is a mixed - WUI. Otherwise, judge whether the wildland patch area is greater than 5 km² and the distance measure is less than 2.4 km. If it is satisfied, the buffer zone is an interface - WUI. If not, it is a non - junction area.

[0105] By constructing a dynamic calculation system adaptable to spatial heterogeneity, the applicability defect of the traditional static threshold method in complex terrain areas is solved. For the calculation of building density, a 500 - meter initial buffer zone and a dynamic radius contraction mechanism are introduced, effectively overcoming the sensitivity of the fixed - window method to uneven spatial distribution. The calculation of vegetation coverage combined with existing data sets significantly reduces the amount of calculation data and the dependence on remote sensing data. The composite constraint conditions (area threshold, shape index, spatial distribution) integrated in the wildland patch screening link lay a precise data foundation for subsequent fire risk modeling. The intelligent construction strategy of multi - level buffer zones through the annular gradient division mechanism makes the spatio - temporal coincidence degree of fire risk spread simulation better match the local fire risk situation compared with the traditional buffer zone.

[0106] Step 6: Set gradient thresholds for the mixed - WUI and the interface - WUI, and calculate the WUI areas of the mixed WUI and the interface - WUI under different thresholds respectively.

[0107] Step 601: Set threshold gradients for the mixed - WUI (building density, vegetation coverage) and the interface - WUI (building density, wildland patch area, distance measure) based on expert experience, specifically including:

[0108] For building density, the lower threshold is set to 1 per square kilometer (fixed value), and the upper limit is dynamically calculated as the maximum value of building density statistics in the study area (extracted from the density raster generated in step 3), with a step size of 5 per square kilometer.

[0109] For vegetation coverage, the threshold range remains 0% to 100% (fixed range, in line with physical definitions). During actual calculation, it is effectively truncated according to the vegetation distribution characteristics of the study area (for example, automatically limited to 0 - 30% in desert areas). The gradient step size is 1%, and it can be refined to 0.5% in high - precision scenarios.

[0110] For wildland patch area, the lower threshold is set to 0.1 square kilometer (fixed value, to avoid interference from too small areas), and the upper limit is dynamically calculated as the largest continuous wildland patch area in the study area, with a step size of 0.1 square kilometer.

[0111] For the buffer zone distance measure, the lower threshold is set to 0.1 km (fixed value, for accuracy requirements), and the upper limit is dynamically set to 10% of the diagonal length of the study area, with a step size of 0.1 km. The formula is:

[0112] ;

[0113] Step 602: Calculate the WUI areas of the hybrid WUI and the interface - type WUI at different thresholds;

[0114] Among them, the calculation process of the WUI area of the hybrid WUI at different thresholds includes:

[0115] Generate candidate areas based on the building density threshold (x), extract all connected areas with a building density ≥ x per square kilometer. Subsequently, overlay the vegetation coverage threshold (y), and within the candidate areas, extract sub - areas with a vegetation coverage ≥ y%;

[0116] Among them, the calculation process of the WUI area of the hybrid WUI at different thresholds includes:

[0117] First, candidate areas sharing the building density threshold (x) with the hybrid - WUI. Subsequently, reverse the vegetation coverage constraint and extract sub - areas with a vegetation coverage < y% within the candidate areas. Thirdly, conduct spatial association analysis of wildland patches, screen wildland patches with an area ≥ m square kilometers, and generate a buffer zone with a width of n kilometers based on the wildland patch boundaries, and retain the interface - type WUI candidate areas that intersect with the buffer zone;

[0118] Step 7: Combine historical burned area data and the WUI area to define the fire exposure index. Use the spatial distribution characteristics of the WUI area as the independent variable of the RF model, and the fire exposure index at different thresholds as the dependent variable of the RF model, and fit to obtain the optimal threshold;

[0119] Specifically, it includes the following steps:

[0120] Step 701: Define the fire exposure index (WEI). The fire exposure index is expressed as a spatial association intensity index between historical fires and the WUI area:

[0121] ;

[0122] Step 702: Uniformly convert the historical burned area data to the same coordinate system (WGS_1984_UTM) as the WUI area, and uniformly resample the spatial resolution to 10 meters;

[0123] Step 703: Use the spatial distribution characteristics of the WUI area generated by the threshold combination as the independent variable (X) of the RF model, and the fire exposure index (WEI) at different thresholds as the dependent variable (Y) of the RF model;

[0124] Adopt a stratified sampling strategy to randomly sample and divide the training set (80%) and the validation set (20%), ensuring the randomness of the selection of the training set and the validation set;

[0125] Define the RF hyperparameter search space as follows:

[0126] 'n_estimators': [100, 200, 300],

[0127] 'learning_rate': [0.01, 0.05, 0.1],

[0128] 'max_depth': [3, 4, 5, 6],

[0129] 'min_samples_split': [2, 3, 4],

[0130] 'min_samples_leaf': [1, 2, 3],

[0131] Adopt a univariate scanning strategy, that is, fix other parameters as the current optimal values, and use the RF model to calculate the partial dependence curve (PDP) under the threshold change for all candidate values generated by the target parameter at a step size within its full range, and take the peak point of the curve as the best threshold of this influencing factor.

[0132] By defining a spatio-temporal weighted fire exposure index (WEI), integrating the spatial alignment and resolution matching mechanism of long-term historical burned area data, the spatial distribution characteristics of fire risk are innovatively dynamically associated with threshold parameters. In the threshold decision-making link, by analyzing the partial dependence curve, the geospatial rationality and practical operability of the threshold scheme are significantly improved. It overcomes the timeliness lag problem of traditional static threshold schemes in different geographical scenarios.

[0133] Example Two

[0134] This embodiment provides an adaptive threshold optimization system for the wildland-urban interface zoning, including:

[0135] A segmentation module, which is used to segment the image of the target area to obtain the building segmentation result;

[0136] A parameter calculation module, which is used to calculate the building density according to the building segmentation result; screen the wildland vegetation types based on the obtained vegetation dataset, and calculate the vegetation coverage, wildland patch area and distance measure respectively;

[0137] A WUI type determination module, which is used to determine the WUI determination rule, and combine the building density, vegetation coverage, wildland patch area, buffer distance measure and determination rule to determine the type of WUI;

[0138] The WUI area determination module is used to set the threshold ranges of building density, vegetation coverage, wildland patch area, and buffer distance measure, and calculate the WUI areas of different types of WUIs under different thresholds.

[0139] The threshold adaptive optimization module is used to combine historical burned area data and the WUI area to define the fire exposure index. It takes the spatial distribution characteristics of the WUI area as the independent variable of the RF model and the fire exposure index under different thresholds as the dependent variable of the RF model, and fits to obtain the optimal threshold.

[0140] It should be noted that the specific implementation of an adaptive threshold optimization system for wildland-urban interface zoning in an embodiment of the present invention is similar to that of an adaptive threshold optimization method for wildland-urban interface zoning in an embodiment of the present invention. For details, please refer to the description in the method section. To avoid redundancy, it will not be elaborated here.

[0141] Embodiment III

[0142] This embodiment provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the steps in an adaptive threshold optimization method for wildland-urban interface zoning as described above.

[0143] Embodiment IV

[0144] This embodiment provides a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements the steps in an adaptive threshold optimization method for wildland-urban interface zoning as described above.

[0145] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An adaptive threshold optimization method for wild city boundary zoning, characterized in that: The steps include: Segment the image of the target area to obtain the building segmentation result; Calculate building density based on building segmentation results; Based on the obtained vegetation dataset, the wildland vegetation types were screened, and the vegetation coverage and wildland patch area and distance measurements were calculated respectively; Determine the WUI determination rules, and determine the type of WUI by combining building density, vegetation coverage, wild patch area, buffer zone distance measurement and determination rules; The threshold ranges of building density, vegetation coverage, wildland patch area and buffer distance measurement were set, and the WUI areas of different types of WUI under different thresholds were calculated; The fire exposure index was defined by combining historical burn site data and the WUI region. The spatial distribution characteristics of the WUI region were used as the independent variable of the RF model, and the fire exposure index under different thresholds was used as the dependent variable of the RF model to obtain the optimal threshold.

2. The adaptive threshold optimization method for wild city junction area demarcation as claimed in claim 1 is characterized in that: The step of segmenting the image of the target area to obtain a building segmentation result includes: Pre-train the ResNet-50 backbone network based on the source domain dataset to obtain universal image features; Transfer the learned general image features to the target domain task; The ResNet-50 backbone network is connected to the corresponding layer of the U-Net decoder through cross-stage feature connection, and the dynamic weighted fusion of multi-scale features is realized through the spatial attention mechanism. The building segmentation result is obtained based on the fused features.

3. The adaptive threshold optimization method for wild city junction area demarcation as claimed in claim 1 is characterized in that: The calculating of building density according to the building segmentation result includes: Convert the building raster data corresponding to the building segmentation result into vector surface data; Generate spatial point features representing buildings based on vector surface data; Based on the spatial point features, spatial density calculation is performed, a buffer zone is constructed with each building point as the center, and the building distribution density around each analysis point is counted.

4. The adaptive threshold optimization method for wild city junction area demarcation as claimed in claim 1 is characterized in that: The method of screening the wildland vegetation type based on the obtained vegetation data set and calculating the vegetation coverage and the wildland patch area and distance measurement respectively includes: The moving window statistical method is used to count the number of wild vegetation pixels in the moving window pixel by pixel, and the wild vegetation pixel number / the total number of pixels in the buffer zone are divided to obtain the regional wild vegetation coverage; The selected wildland raster data is converted into vector surface data through GDAL, the area of ​​each vector patch is calculated, and the wildland patches with an area larger than the set area are selected to obtain the wildland patch area; Generated along wild patch boundaries n The distance measure is obtained by setting the buffer distance and the level threshold.

5. The adaptive threshold optimization method for wild city junction area demarcation as claimed in claim 1 is characterized in that: Threshold ranges for building density, vegetation cover, wildland patch area, and buffer distance measures include: For building density, the threshold range is set from 1 / km2 to the maximum building density, with a step size of 5 / km2; For vegetation cover, the threshold range was set from 0% to 100% with a gradient step of 1%; For wild patch area, the threshold range was set from 0.1 km2 to the maximum patch area, with a step size of 0.1 km2; For the buffer distance measure, the threshold range is 0.1 km to 5.0 km, with a step size of 0.1 km.

6. The adaptive threshold optimization method for wild city junction area demarcation as claimed in claim 1 is characterized in that: Calculate the WUI area for different types of WUI at different thresholds, including: The calculation process of the WUI area under different thresholds for mixed WUI includes: Generate candidate areas based on building density thresholds, extract all connected areas with building density ≥ x / km2, then superimpose vegetation coverage thresholds, and extract sub-areas with vegetation coverage ≥ y% within the candidate areas; The calculation process of the WUI area under different thresholds for mixed WUI includes: First, candidate areas that share the building density threshold with the mixed-WUI are selected. Then, the vegetation coverage constraint is reversed to extract sub-areas with vegetation coverage < y% within the candidate areas. In the third step, spatial association analysis of wildland patches is performed to screen wildland patches with an area ≥ m square kilometers. A strip buffer with a width of n kilometers is generated based on the boundary of the wildland patch, and the interface-type WUI candidate areas that intersect with the buffer are retained.

7. The adaptive threshold optimization method for wild city junction area demarcation as claimed in claim 1 is characterized in that: The fire exposure index WEI is expressed as: 。 8. An adaptive threshold optimization system for wild city boundary zoning, characterized in that: include: A segmentation module is used to segment the image of the target area to obtain a building segmentation result; A parameter calculation module, which is used to calculate the building density according to the building segmentation result; Based on the obtained vegetation dataset, the wildland vegetation types were screened, and the vegetation coverage and wildland patch area and distance measurements were calculated respectively; The WUI type determination module is used to determine the WUI determination rules, and the type of WUI is determined by combining the building density, vegetation coverage, wild patch area, buffer zone distance measurement and determination rules; The WUI area determination module is used to set the threshold range of building density, vegetation coverage, wild patch area and buffer distance measurement, and calculate the WUI area of ​​different types of WUI under different thresholds; The threshold adaptive optimization module is used to define the fire exposure index by combining historical burn site data and the WUI area. The spatial distribution characteristics of the WUI area are used as the independent variable of the RF model, and the fire exposure index under different thresholds is used as the dependent variable of the RF model to obtain the optimal threshold by fitting.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of an adaptive threshold optimization method for wild-urban boundary zoning as described in any one of claims 1 to 7 are implemented.

10. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the adaptive threshold optimization method for wild-city boundary zoning as described in any one of claims 1-7 are implemented.

Images