Traditional cave dwelling settlement remote sensing image intelligent identification and protection planning method and system

Abstract

The application discloses a traditional cave dwelling remote sensing image intelligent recognition and protection planning method and system, belongs to the technical field of remote sensing image intelligent interpretation and cultural heritage protection, and comprises the following steps: acquiring satellite images, unmanned aerial vehicle data and a digital elevation model and generating a slope feature map; adopting a terrain auxiliary multi-scale cave semantic segmentation network to identify three types of caves, i.e., cliff-type caves, sunken-type caves and independent-type caves; extracting a settlement boundary based on spatial correlation degree constraint density clustering; analyzing settlement scale, shape and density characteristics; tracking historical evolution through multi-period image registration; and adopting a multi-factor weighted model to evaluate protection value and delimit three-level protection partitions. The application can realize fine identification of cave types, automatic extraction of a settlement boundary, quantitative tracking of historical evolution and objective evaluation of protection value, and provides intelligent technical support for traditional cave dwelling protection planning.

Description

Technical Field

[0001] This invention relates to the field of intelligent interpretation of remote sensing images and cultural heritage protection technology, and in particular to a method and system for intelligent identification and protection planning of remote sensing images of traditional cave dwelling settlements. Background Technology

[0002] For the investigation and protection of traditional settlements, existing technical solutions mainly rely on manual on-site surveys and expert experience assessments. For example, Chinese patent application CN115393368A discloses a method, system, device, and medium for identifying settlement site selection environmental patterns based on meta-learning. This method acquires remote sensing images and digital elevation model data of the settlement to be identified, inputs them into a trained settlement feature segmentation model to obtain an environmental feature distribution map of the settlement, and then inputs the environmental feature distribution map into a trained settlement site selection environmental pattern identification model to obtain the settlement's site selection environmental pattern type. This technical solution achieves intelligent identification of traditional settlement site selection environmental patterns, improving the efficiency of settlement surveys to a certain extent.

[0003] However, the aforementioned existing technologies have the following shortcomings. First, existing technologies mainly focus on the segmentation and site selection pattern classification of environmental features surrounding settlements, failing to achieve refined type identification for the unique architectural form of cave dwellings, and unable to distinguish the spatial distribution characteristics of different types of cave dwellings such as cliff-side cave dwellings, sunken cave dwellings, and independent cave dwellings. Second, existing technologies lack the ability to automatically extract settlement boundaries based on the distribution density and spatial correlation of cave dwellings, making it difficult to accurately define the boundaries of cave dwelling settlements. Third, existing technologies have not established a multi-period remote sensing image collaborative analysis mechanism, making it impossible to track the historical evolution trajectory of cave dwelling settlements, including the processes of settlement expansion, contraction, and disappearance. Finally, existing technologies lack an objective and quantitative evaluation index system for assessing conservation value, failing to integrate multi-dimensional factors such as settlement scale, morphological integrity, historical continuity, and environmental landscape harmony to output a conservation value level score, and also failing to automatically generate conservation planning zoning recommendations based on the assessment results. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for intelligent identification and protection planning of traditional cave dwelling settlements using remote sensing images, in order to solve the technical problems that existing technologies cannot achieve refined identification of cave dwelling types, automatic extraction of settlement boundaries, tracking of historical evolution, and objective quantitative assessment of protection value.

[0005] To achieve the above objectives, this invention provides a method for intelligent identification and protection planning of traditional cave dwelling settlements using remote sensing images, comprising the following steps: a remote sensing data acquisition and preprocessing step, acquiring high-resolution satellite imagery, UAV oblique photography data, and digital elevation model data of the target area; performing geometric correction and coordinate registration on the multi-source remote sensing data to generate remote sensing image data and slope feature maps; a cave dwelling intelligent identification step, inputting the remote sensing image data and slope feature maps into a terrain-assisted multi-scale cave dwelling semantic segmentation network; fusing terrain information through multi-scale feature extraction and an attention mechanism to output a cave dwelling type distribution map including three types: cliff-side cave dwellings, sunken cave dwellings, and independent cave dwellings; and a settlement boundary extraction step, extracting a spatial point set of cave dwellings based on the cave dwelling type distribution map, and employing an adaptive method that integrates spatial correlation constraints. The process involves several steps: First, a density clustering algorithm delineates settlement boundaries and generates settlement boundary polygons. Second, a settlement morphology analysis step extracts settlement scale, layout, cave density, and courtyard organization pattern indicators based on the settlement boundary polygons and cave dwelling type distribution maps, generating settlement morphology feature vectors. Third, a historical evolution analysis step acquires historical image sequences and performs multi-period collaborative registration with current remote sensing image data. A change detection algorithm tracks the expansion, contraction, and disappearance of cave dwelling settlements, generating evolution trajectory data. Fourth, a conservation planning generation step uses the settlement morphology feature vectors and evolution trajectory data, employing a multi-factor weighted evaluation model to calculate conservation value scores. Based on these scores, core protection zones, construction control zones, and environmental coordination zones are delineated, outputting conservation zone boundary data and a conservation value assessment report.

[0006] This invention also provides a remote sensing image intelligent recognition and protection planning system for traditional cave dwelling settlements, comprising: a remote sensing data acquisition module configured to acquire high-resolution satellite imagery, UAV oblique photography data, and digital elevation model data of the target area, perform geometric correction and coordinate registration processing on the multi-source remote sensing data, and generate remote sensing image data and slope feature maps; a cave dwelling identification module configured to input the remote sensing image data and slope feature maps into a terrain-assisted multi-scale cave dwelling semantic segmentation network, and output a cave dwelling type distribution map; a settlement boundary extraction module configured to generate settlement boundary polygons based on the cave dwelling type distribution map using an adaptive density clustering algorithm that incorporates spatial correlation constraints; a settlement morphology analysis module configured to extract settlement morphology feature vectors based on the settlement boundary polygons and the cave dwelling type distribution map; a historical evolution analysis module configured to generate evolution trajectory data through multi-period image collaborative registration and change detection; and a protection planning generation module configured to calculate protection value scores and delineate protection zone boundaries using a multi-factor weighted evaluation model.

[0007] The beneficial effects of this invention are as follows: First, the terrain-assisted multi-scale cave dwelling semantic segmentation network designed in this invention, by integrating slope feature maps and a multi-scale attention mechanism, can accurately identify and distinguish three types of cave dwellings: cliff-side, sunken, and independent. Its identification accuracy is significantly better than existing general semantic segmentation methods. Second, the adaptive density clustering algorithm proposed in this invention, which integrates spatial correlation constraints, can automatically delineate settlement boundaries based on the distribution characteristics of cave dwellings, effectively solving the problem of irregular and difficult-to-define traditional settlement boundaries. Third, the multi-period image collaborative registration and change detection mechanism established in this invention enables quantitative tracking of the historical evolution of cave dwelling settlements, providing objective temporal analysis basis for protection decisions. Fourth, the multi-factor weighted protection value assessment model constructed in this invention integrates four dimensions: scale, integrity, continuity, and coordination, outputting an objective and quantitative protection value score and automatically generating three-level protection zoning suggestions, providing intelligent technical support for the systematic survey and hierarchical classification protection planning of traditional cave dwelling settlements. Attached Figure Description

[0008] Figure 1 This is a flowchart of the method for intelligent identification and protection planning of traditional cave dwelling settlements using remote sensing images, as described in this invention.

[0009] Figure 2 This is the architecture diagram of the intelligent identification and protection planning system for remote sensing images of traditional cave dwelling settlements according to the present invention. Detailed Implementation

[0010] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments described below are only for explaining the present invention and are not intended to limit the scope of protection of the present invention.

[0011] Reference Figure 1 The method for intelligent identification and protection planning of traditional cave dwelling settlements based on remote sensing images provided in this embodiment of the invention includes the following steps.

[0012] Step S1: Remote sensing data acquisition and preprocessing steps.

[0013] In one embodiment of the present invention, the remote sensing data acquisition and preprocessing step is used to acquire multi-source remote sensing data of the target area and perform standardized processing, laying a data foundation for subsequent cave dwelling identification and analysis. Specifically, high-resolution satellite imagery of the target area is first acquired. Preferably, the spatial resolution of the satellite imagery is not less than 0.5m to ensure that the morphological features of individual cave dwellings can be clearly identified. In a preferred embodiment of the present invention, multispectral imagery covering the typical cave dwelling distribution area of ​​the Loess Plateau is acquired using domestic Gaofen series satellites or commercial satellites. The imagery includes four bands: red, green, blue, and near-infrared. The image size is determined according to the range of the study area, with typical values ​​ranging from 10000×10000 pixels to 50000×50000 pixels.

[0014] Simultaneously, digital elevation model (DEM) data of the target area is acquired. This DEM data is used to calculate terrain slope characteristics and assist in distinguishing the spatial distribution of different types of cave dwellings. Preferably, the spatial resolution of the DEM is no less than 5m, and the elevation accuracy is no less than 2m. In key areas with dense cave dwelling settlements, higher-precision terrain data can also be obtained using UAV oblique photography. The UAV flight altitude is typically set to 100m to 200m, and the ground sampling distance can reach 0.02m to 0.05m.

[0015] After acquiring multi-source remote sensing data, geometric correction and coordinate registration are performed. In one embodiment of the present invention, geometric correction employs a rational polynomial coefficient model for orthorectification to eliminate geometric distortion caused by terrain undulations. Coordinate registration unifies all data to the same geographic coordinate system, preferably using the WGS84 coordinate system or the National Geodetic Coordinate System 2000. Registration accuracy is controlled within one pixel to ensure geometric consistency in subsequent multi-source data fusion analysis.

[0016] Based on the registered digital elevation model data, a slope feature map is generated. The slope calculation employs a third-order inverse distance weighted difference algorithm for any location within the digital elevation model. elevation value at the location Its slope The calculation formula is: ,in, For position The slope value at the location is in degrees (°) and ranges from [0, 90]. This is the elevation value at that location, in meters (m). and These are the elevation change rates along the east-west and north-south directions, respectively, calculated using the elevation values ​​within a 3×3 neighborhood window through a difference algorithm, with units of m / m; It is the arctangent function; The conversion factor for radians to angles is approximately 57.2958. The introduction of slope characteristic maps is significant for identifying cave dwelling types: cliff-side cave dwellings are typically distributed in areas with steep cliff slopes, generally greater than 25°; sunken cave dwellings are located on relatively flat terraces or slope tops, with slopes typically less than 10°; and independent cave dwellings are distributed in gently sloping areas with moderate slopes, generally between 10° and 25°.

[0017] This step ultimately outputs standardized remote sensing image data. and slope feature map ,in and These are the height and width of the image, respectively. This represents the number of spectral bands. This data will serve as input for subsequent cave dwelling identification steps.

[0018] Preferably, the data preprocessing also includes radiometric correction and atmospheric correction. Radiometric correction converts the raw digital quantization values ​​into physical quantity radiance values, eliminating the influence of sensor response differences. Atmospheric correction uses the FLAASH model or 6S model to remove the effects of atmospheric scattering and absorption, obtaining the true surface reflectance. After radiometric and atmospheric correction, image data acquired from different time phases and different sensors are comparable, laying the foundation for subsequent multi-phase image comparison and analysis.

[0019] In one embodiment of the present invention, histogram equalization and contrast enhancement processing are performed on the remote sensing image data to improve the distinguishability between cave dwellings and background features. Specifically, cumulative histograms are calculated for each band of the image, and then nonlinear stretching is performed to make the pixel value distribution more uniform. Experiments show that the contrast-enhanced image can improve the segmentation accuracy in cave dwelling identification tasks by about 3 to 5 percentage points. In addition, for cloud-covered areas, multi-temporal image synthesis or interpolation is used to ensure the integrity of the target area data.

[0020] Step S2: Intelligent identification of cave dwellings.

[0021] In one embodiment of the present invention, the intelligent cave dwelling identification step employs a specially designed terrain-assisted multi-scale cave dwelling semantic segmentation network to achieve refined identification of three types of cave dwellings. This network architecture innovatively improves upon the traditional encoder-decoder structure by introducing terrain feature bypass branches and a multi-scale attention fusion module, significantly enhancing the accuracy and robustness of cave dwelling identification.

[0022] The overall architecture of the terrain-assisted multi-scale cave dwelling semantic segmentation network comprises three core components: a backbone feature extraction network, a terrain feature bypass branch, and a multi-scale attention fusion module. The backbone feature extraction network uses ResNet-101 as the encoder, extracting semantic features from remote sensing images layer by layer through five residual blocks. Each residual block outputs feature maps at different scales, with spatial resolutions of 1 / 2, 1 / 4, 1 / 8, 1 / 16, and 1 / 32 of the original image. Preferably, the backbone network is initialized with weights pre-trained on the ImageNet dataset to accelerate network convergence and improve feature representation capabilities.

[0023] The terrain feature bypass branch is one of the innovations of this invention. This branch is specifically designed to extract terrain semantic information from the slope feature map. In a preferred embodiment of this invention, the terrain feature bypass branch adopts a three-layer convolutional network structure. The first convolutional layer uses a 7×7 kernel with a stride of 2 and 64 output channels; the second convolutional layer uses a 3×3 kernel with a stride of 2 and 128 output channels; and the third convolutional layer uses a 3×3 kernel with a stride of 1 and 256 output channels. Each convolutional layer is followed by a batch normalization layer and a ReLU activation function. The output feature map of the terrain feature bypass branch has the same spatial size as the output feature map of the third residual block of the backbone network, which facilitates subsequent feature fusion.

[0024] The multi-scale attention fusion module is another innovation of this invention. This module combines spatial pyramid pooling with channel attention to effectively fuse feature information at different scales. Specifically, firstly, spatial pyramid pooling is performed on the feature maps output from each layer of the backbone network, with pooling kernel sizes set to 1×1, 2×2, 3×3, and 6×6, respectively, to obtain feature representations with four different receptive fields. Subsequently, the pooled features are upsampled to restore a uniform size and then concatenated with the output of the terrain feature bypass branch.

[0025] For the stitched multi-scale feature map The weight coefficients of each channel are calculated using a channel attention mechanism. Channel attention weights. The calculation formula is: ,in, This is the channel attention weight vector, with values ​​ranging from [0,1], representing the importance of each channel feature; This is the stitched multi-scale feature map; For global average pooling operations, the spatial dimension is compressed to 1×1; and These are the weight matrices for the two fully connected layers, where... The compression ratio is 16, with a preferred value to reduce computational complexity. To modify the activation function of the linear unit; The Sigmoid activation function maps the output value to the [0,1] interval. The channel attention mechanism adaptively emphasizes the feature channels more important for cave dwelling identification while suppressing interference from irrelevant features.

[0026] The feature maps, after channel attention weighting, are fed into the decoder for progressive upsampling. The decoder employs a progressive upsampling strategy, with each upsampling increment being 2. During the upsampling process, skip connection features from the corresponding encoder layer are fused to recover spatial detail information. Finally, the decoder outputs semantic segmentation results of the same size as the original image, containing four categories: background, cliff-side cave dwellings, sunken cave dwellings, and freestanding cave dwellings.

[0027] In one embodiment of the present invention, network training employs a weighted combination of cross-entropy loss function and Dice loss function as the optimization objective to balance pixel-level classification accuracy and region connectivity. Total loss function Defined as: ,in, The total loss value is dimensionless. The cross-entropy loss is used to calculate the difference between the predicted probability distribution and the true label. The Dice loss measures the degree of overlap between the predicted and actual regions. and These are the weighting coefficients for the two types of losses, with the preferred values ​​being... and This ratio optimizes the integrity of region boundaries while ensuring classification accuracy. The network training uses the Adam optimizer with an initial learning rate of 0.0001, a batch size of 8, and 100 training epochs. A cosine annealing strategy is used to dynamically adjust the learning rate.

[0028] This step outputs a map showing the distribution of cave dwelling types. Each pixel value represents the category label for that location: 0 represents the background, 1 represents a cliff-side cave dwelling, 2 represents a sunken cave dwelling, and 3 represents a freestanding cave dwelling. In the test experiments of this invention, the cave dwelling semantic segmentation network achieved an average intersection-union ratio of 88.6% on the self-built Loess Plateau cave dwelling dataset, which is 7.3 percentage points higher than the baseline method using the standard DeepLabV3+ network, verifying the effectiveness of the terrain feature-assisted and multi-scale attention fusion strategy.

[0029] In a preferred embodiment of the present invention, a data augmentation strategy is employed during the network training phase to further enhance the robustness of cave dwelling identification. Data augmentation operations include random horizontal flipping, random vertical flipping, random rotation (rotation angle ranging from -15° to 15°), random scaling (scaling ratio ranging from 0.8 to 1.2), random cropping, Gaussian noise injection, and color dithering. Experiments show that the generalization ability of the model is significantly improved after data augmentation, and the fluctuation in identification accuracy on cave dwelling datasets from different regions is reduced by approximately 4 percentage points.

[0030] Preferably, a test-time enhancement strategy is employed during the model inference stage. Specifically, the same input image undergoes four transformations: original, horizontally flipped, vertically flipped, and rotated 180°. The transformed images are then input into the network to obtain prediction results. The prediction results are then inversely transformed, and the average value is taken as the final output. Test-time enhancement effectively reduces prediction noise in boundary regions, resulting in a smoother and more complete cave dwelling outline. Furthermore, this invention employs a conditional random field as a post-processing step, utilizing the spatial relationships between pixels to optimize the segmentation boundaries, further improving the precision of the segmentation results.

[0031] Step S3: Settlement boundary extraction step.

[0032] In one embodiment of the present invention, the settlement boundary extraction step is based on a cave dwelling type distribution map and employs an adaptive density clustering algorithm that incorporates spatial correlation constraints to automatically delineate the boundaries of cave dwelling settlements. Compared with the traditional DBSCAN clustering algorithm, the improved algorithm proposed in this invention introduces a measurement mechanism for the spatial correlation between cave dwellings, which can more accurately identify cave dwelling groups with functional associations, thereby extracting polygonal ranges that conform to the actual boundaries of the settlements.

[0033] First, a spatial point set of cave dwellings is extracted from the cave dwelling type distribution map. Specifically, a connected component analysis is performed on the cave dwelling type distribution map to extract the centroid coordinates of each independent cave dwelling region as representative points of the cave dwellings. Let the extracted spatial point set of cave dwellings be... ,in Indicates the first Location coordinates and type labels for each cave dwelling and These are geographic coordinates, in meters. These correspond to three types of cave dwellings: cliff-side, sunken, and independent.

[0034] Subsequently, the spatial correlation between the cave dwellings is calculated. In one embodiment of the present invention, the spatial correlation comprehensively considers three factors: Euclidean distance, azimuth difference, and type similarity between the cave dwellings. For any two cave dwellings... and Its spatial correlation The calculation formula is: ,in, cave dwellings and The spatial correlation between the two cave dwellings ranges from [0,1], with a larger value indicating a higher degree of correlation between the two cave dwellings. The Euclidean distance between the two cave dwellings is expressed in meters (m). The formula for its calculation is: ; The angle between the line connecting the two cave dwellings and the principal axis direction, expressed in degrees, is determined by principal component analysis of the local cave dwelling distribution. The distance attenuation parameter is preferably 50m, which is determined based on the spatial scale of a typical cave dwelling courtyard. The azimuth attenuation parameter is preferably 30°. For type similarity functions, when The value is 1.0 when... The value is set to 0.7, reflecting the spatial characteristic that similar cave dwellings tend to cluster together. The introduction of spatial correlation enables the algorithm to identify functionally related groups of cave dwellings, avoiding the erroneous classification of geographically adjacent but functionally independent cave dwellings into the same settlement.

[0035] Based on spatial correlation calculation, adaptive density clustering is performed. In one embodiment of the invention, the improved density clustering algorithm replaces the distance-based neighborhood determination in traditional DBSCAN with a neighborhood determination based on spatial correlation. For cave dwellings... Its associated neighborhood Defined as: ,in, cave dwellings The set of related neighborhoods; The correlation threshold is set to 0.3, with a preferred value of 0.3. This threshold is determined through cross-validation based on the spatial organization characteristics of typical cave dwelling settlements. When the number of cave dwellings in the correlated neighborhood is not less than the minimum point threshold... At that time, the cave dwelling was marked as the core point. Preferably, Setting it to 5 means that a cave dwelling settlement should contain at least 5 cave dwelling buildings.

[0036] After clustering, a settlement boundary polygon is generated for each cluster. In one embodiment of the invention, a concave hull algorithm is used for boundary generation, which can better fit the irregular boundary shape of the cave dwelling settlement compared to the traditional convex hull algorithm. Specifically, using representative points of all cave dwellings within the cluster as input, the AlphaShapes algorithm is used to generate concave polygon boundaries, with the Alpha parameter set to 100m, which achieves a good balance between boundary smoothness and detail preservation. Finally, the generated boundary polygons are buffered outwards, with a buffer distance set to 20m to cover the courtyard land and alleyway space surrounding the cave dwellings.

[0037] This step outputs a set of settlement boundary polygons. ,in For the number of identified settlements, each This represents the vertex coordinate sequence of a closed polygon. In the test experiments of this invention, the intersection-union ratio (IoU) between the extracted settlement boundaries and the manually labeled boundaries reached 82.5%, significantly better than the 68.3% achieved by directly using the DBSCAN algorithm, thus verifying the effectiveness of the spatial correlation constraint mechanism.

[0038] In a preferred embodiment of the present invention, boundary smoothing and topology verification mechanisms are also introduced during the settlement boundary extraction process. Boundary smoothing employs the Douglas-Peucker algorithm to simplify polygon vertices, with a simplification threshold set to 5m. This removes jagged edges while maintaining the overall shape of the boundary, resulting in a more natural and smooth boundary line. Topology verification is used to detect and correct potential topological errors in the boundary polygons, such as self-intersections, holes, and dangling edges, ensuring the geometric validity of the output boundary.

[0039] Preferably, for larger settlements, this invention also supports hierarchical boundary extraction. Specifically, a larger correlation threshold is first used to extract the overall outer boundary of the settlement. Then, a smaller correlation threshold is used within the settlement to identify relatively independent courtyard clusters, thereby obtaining a three-level hierarchical spatial boundary structure of settlement-cluster-courtyard. This hierarchical boundary can more precisely characterize the internal spatial organization features of the settlement, providing richer spatial information for subsequent morphological analysis and conservation planning.

[0040] Furthermore, this invention also provides a boundary editing interface, allowing professionals to manually correct and adjust the automatically extracted boundaries. The results of manual correction can be fed back into the parameter optimization process of the clustering algorithm, continuously improving the accuracy of automatic boundary extraction through iterative learning, thus forming a progressive optimization mechanism of human-machine collaboration.

[0041] Step S4: Settlement morphology analysis.

[0042] In one embodiment of the present invention, the settlement morphology analysis step, based on the settlement boundary polygon and the distribution map of cave dwelling types, systematically extracts the spatial morphological features of the settlement and generates a settlement morphology feature vector for conservation value assessment. This step performs quantitative analysis from four dimensions: settlement size, layout morphology, cave dwelling density, and courtyard organization pattern, comprehensively characterizing the spatial structural features of the cave dwelling settlement.

[0043] First, calculate the settlement size indicators. Settlement size mainly includes three sub-indicators: settlement area, total number of cave dwellings, and number of cave dwellings of each type. For the settlement boundary polygon... Its area Calculated using the polygon area formula, unit: Total number of cave dwellings This represents the number of representative points of cave dwellings falling within the bounded polygon area. Number of cave dwellings of each type. The number of cliff-side, sunken, and independent cave dwellings was counted separately. Preferably, the settlement area was normalized to the [0, 100] interval for easier subsequent comprehensive evaluation. The normalization formula is: ,in, The normalized area value is dimensionless and ranges from [0, 100]. This represents the actual area of ​​the settlement, in units of... ; Set the reference area value to 50000. (i.e., 5 hectares), this value is determined based on the scale distribution characteristics of typical cave dwelling settlements on the Loess Plateau, and settlements exceeding this area receive full marks.

[0044] Secondly, the settlement layout morphology is analyzed. The layout morphology analysis uses three geometric characteristic indicators: shape index, compactness, and fractal dimension. Shape index... The formula reflecting the complexity of settlement boundaries is as follows: ,in, is a shape index, dimensionless, with a value of 1 indicating a perfect circle, and a larger value indicating a more irregular boundary; The perimeter of the settlement boundary polygon, in meters; The area represents the settlement area, in units of... ; Pi, approximately 3.14159. Compactness. Fractal dimension reflects the spatial density of a settlement and is calculated as the ratio of the settlement area to the area of ​​its smallest circumscribed rectangle, with a value ranging from [0,1]. Box counting was used to calculate the self-similarity of settlement boundaries at different scales. The typical fractal dimension of cave dwelling settlements in the Loess Plateau ranges from 1.1 to 1.4.

[0045] Next, calculate the density index of the cave dwellings. Cave dwelling density. Defined as the number of cave dwellings per unit area, the calculation formula is: ,in, Density of cave dwellings, unit: cave dwellings / ; The total number of cave dwellings within the settlement, expressed in units of; The area represents the settlement area, in units of... Multiply by It is to change the unit from individual / Converted to units / Preferably, the distribution density ratio of each type of cave dwelling is also calculated to reflect the compositional characteristics of cave dwelling types within the settlement.

[0046] Finally, the courtyard organization pattern is analyzed. The courtyard organization pattern analysis aims to identify the spatial organization patterns within the cave dwelling settlement. In one embodiment of this invention, a Voronoi diagram analysis method is used to construct the spatial adjacency relationships between cave dwellings, and courtyard units are identified based on these relationships. Specifically, a Voronoi diagram is constructed using representative points of cave dwellings as generators; cave dwellings corresponding to adjacent Voronoi units are considered to have potential courtyard organization relationships. Subsequently, based on the spatial correlation and type characteristics of adjacent cave dwellings, a hierarchical clustering method is used to identify courtyard units. Courtyard organization pattern indicators include average courtyard size (measured by the number of cave dwellings), courtyard spacing (the average distance between the centroids of adjacent courtyards), and courtyard arrangement orientation (the relationship between the main courtyard arrangement direction and the slope aspect).

[0047] This step ultimately outputs a settlement morphology feature vector. This vector contains the quantitative index values ​​of the above dimensions, which serve as input data for subsequent protection value assessment.

[0048] In a preferred embodiment of the present invention, the courtyard organization pattern analysis also includes statistical analysis of courtyard orientation and entrance direction. By analyzing the opening direction of the cave dwellings (calculated based on the slope aspect information from the slope characteristic map) and the main axis direction of the courtyard, the dominant orientation pattern of the settlement can be identified. This feature reflects the adaptive utilization of natural conditions such as sunlight and ventilation by the early inhabitants in the construction of the settlement. In addition, the present invention also calculates the spatial connectivity index between courtyards and analyzes the characteristics of the alleyway network structure within the settlement, including parameters such as the total length of alleyways, average width, number of intersections, and network connectivity. These indicators have important reference value for assessing the integrity of the settlement's spatial structure.

[0049] Preferably, the morphological feature extraction process employs a multi-scale analysis strategy. At the macro-scale, the overall outline and spatial distribution characteristics of the settlement are analyzed; at the meso-scale, the layout patterns and organizational rules of courtyard clusters are analyzed; and at the micro-scale, the morphological characteristics and typological composition of individual cave dwellings are analyzed. The results of the multi-scale analysis are integrated into a unified feature vector format, comprehensively depicting the spatial morphological characteristics of the settlement from the overall to the local levels.

[0050] Step S5: Historical Evolution Analysis Step.

[0051] In one embodiment of the present invention, the historical evolution analysis step acquires and analyzes historical image sequences to trace the spatial evolution of cave dwelling settlements in different historical periods, providing objective temporal evidence for conservation value assessment. This step includes three stages: historical image acquisition, multi-period image collaborative registration, and change detection and analysis.

[0052] First, a historical image sequence is acquired. The sources of historical image data include historical aerial photographs, early satellite imagery, and surveying and mapping data from different periods. Preferably, the historical images should cover a time span of approximately 30 to 50 years, with time intervals of 5 to 10 years. In one embodiment of the invention, for a typical cave dwelling settlement on the Loess Plateau, image data from five periods—the 1980s, 1990s, 2000s, 2010s, and the present—were collected to form a historical image sequence. ,in For the number of image periods, Indicates the first The year in which the images were acquired.

[0053] Subsequently, multi-stage collaborative registration is performed on the historical image sequence. Due to differences in image data sources, resolutions, and imaging conditions across different periods, unified geometric correction and coordinate registration are required. In one embodiment of this invention, the registration process uses the current high-precision remote sensing image as the reference image, employing a combination of feature point matching and affine transformation to register the historical images to a unified coordinate system. Specifically, SIFT feature points are first automatically extracted from both the historical and reference images. Then, the RANSAC algorithm is used to remove mismatched points and estimate affine transformation parameters. Finally, the historical images are resampled to achieve geometric registration. The registration accuracy is controlled within 2 to 3 pixels; for historical aerial photographs with lower resolution, the accuracy requirement can be appropriately relaxed.

[0054] After registration is completed, change detection analysis is performed. In one embodiment of the invention, change detection employs a method combining pixel-level differencing and object-level analysis. First, differencing is performed on adjacent images to extract regions showing significant changes. Then, the differencing results are overlaid with the cave dwelling identification results to identify evolution events such as cave dwelling construction, cave dwelling disappearance, changes in cave dwelling type, and settlement boundary expansion or contraction. For each identified cave dwelling settlement, its evolution trajectory data is constructed. ,in For the settlement in the Area of ​​the period For the number of cave dwellings, Label the evolution type (expansion, stability, contraction, or extinction).

[0055] To quantify the evolutionary characteristics of settlements, this invention defines an evolution rate index and a historical continuity index. Evolution Rate The calculation formula is: ,in, This represents the average annual rate of evolution, expressed as % / year. Positive values ​​indicate settlement expansion, while negative values ​​indicate settlement contraction. This represents the current number of cave dwellings, expressed in units. The number of cave dwellings in the earliest historical period, expressed in units; The time span is in years. Historical continuity indicator. It reflects the degree of continuous existence of settlements in various historical periods and is calculated as the ratio of the number of periods in which the settlements continued to exist to the total number of periods, with a value range of [0,1].

[0056] This step outputs evolutionary trajectory data and evolutionary characteristic indicators, providing a historical dimension for subsequent conservation value assessment. In practical applications of this invention, the evolutionary analysis results can intuitively reveal the historical development of cave dwelling settlements and identify high-value settlements with a long history and good continuity.

[0057] In a preferred embodiment of the present invention, the historical evolution analysis further includes tracking the migration trajectory of the settlement's spatial center of gravity. By calculating the geometric center coordinates of the settlement boundary polygons in each historical period and plotting the migration trajectory of the settlement's spatial center of gravity, the spatial development direction and expansion trend of the settlement in different historical stages can be intuitively reflected. Spatial center of gravity migration analysis has important reference value for understanding the evolutionary patterns of settlements and predicting future development trends.

[0058] Preferably, for settlements with long-term time-series image data, this invention also supports automatic evolution stage segmentation. Specifically, based on the time-series change curves of core indicators such as settlement area and number of cave dwellings, piecewise linear fitting or change point detection algorithms are used to automatically identify key turning points in the settlement's development process, dividing the evolution process into different stages such as the formation stage, development stage, stable stage, and decline stage. The evolution stage segmentation results can help planning decision-makers quickly grasp the historical development context and current evolution status of settlements, providing a scientific basis for formulating targeted protection strategies. In addition, this invention also supports presenting the evolution analysis results in a dynamic visualization manner, generating a time-series animation of the settlement's historical evolution, making it easier for non-professionals to intuitively understand the spatial change process of settlements.

[0059] Step S6: Protection plan generation steps.

[0060] In one embodiment of the present invention, the protection planning generation step is based on settlement morphology feature vectors and evolution trajectory data, employs a multi-factor weighted evaluation model to calculate a protection value score, and automatically delineates three-level protection zones based on the score results, generating protection planning recommendations. This step is the core output of the technical solution of the present invention, providing objective and quantitative technical support for the protection decision-making of traditional cave dwelling settlements.

[0061] First, a multi-factor weighted conservation value assessment model is constructed. This invention decomposes the conservation value assessment into four dimensions: scale score, integrity score, continuity score, and coordination score, and obtains a comprehensive conservation value score through weighted summation. (Comprehensive Conservation Value Score) The calculation formula is: ,in, The comprehensive conservation value score ranges from [0, 100], with higher values ​​indicating greater conservation value. The scale score reflects the spatial size of the settlement and the number of cave dwellings. The integrity score reflects the degree of integrity of the settlement's spatial morphology and the distribution of cave dwellings; The continuity score reflects the historical evolution and temporal continuity of the settlement. The coordination score reflects the degree of harmony between the settlement and its surrounding environment. The weight coefficients for each dimension satisfy... The preferred values ​​are respectively , , and This weighting reflects the central role of integrity in the assessment of conservation value.

[0062] Size rating Taking into account both the settlement area and the number of cave dwellings, the calculation formula is as follows: ,in, This represents the normalized settlement area, with a value range of [0, 100]. This represents the total number of cave dwellings, expressed in units of [number]. The number of cave dwellings is set to 100 for reference purposes. Settlements exceeding this number will receive a perfect score.

[0063] Integrity score An assessment is conducted based on the integrity of the settlement's spatial morphology and the continuity of the cave dwelling distribution. The calculation formula is as follows: ,in, This is the shape index; a larger value indicates a more irregular boundary and lower integrity. For compactness, a larger value indicates a higher degree of spatial compactness and better integrity; The internal vacancy rate is defined as the percentage of the area within the settlement boundary without cave dwellings, ranging from [0,1]. It reflects whether there are significant vacancy or damaged areas within the settlement. The integrity score is limited to the interval [0,100].

[0064] Continuity rating Based on the results of historical evolution analysis, reflecting the temporal continuity characteristics of settlements, the calculation formula is as follows: ,in, It is a historical continuity indicator, with a value range of [0,1]. This represents the average annual rate of change, expressed as % / year. The maximum acceptable evolution rate is set at 5% per year. The design philosophy of this formula is that settlements with higher historical continuity and more stable (close to zero) evolution rates have higher continuity scores.

[0065] Coordination score The degree of harmony between a settlement and its surrounding natural and human environment is assessed. In one embodiment of this invention, the harmony assessment considers three factors: land use type, vegetation cover, and the degree of disturbance from modern buildings surrounding the settlement. Specifically, within a 500m buffer zone outside the settlement boundary, the composition of land use types is analyzed, and the area proportions of different land use types, such as traditional farmland, natural vegetation, and modern buildings, are calculated, thereby calculating a harmony score. The higher the proportion of traditional farmland and natural vegetation and the less disturbance from modern buildings, the higher the harmony score.

[0066] Based on a comprehensive protection value score, three levels of protection zones are delineated. This invention employs a hierarchical threshold method for zone delineation, with the specific rule being: when... At that time, the settlement was designated as a core protected area; when At that time, it was included in the construction control zone; when At that time, it is included in the environmental coordination zone. Preferably, the spatial range of the zone boundary is differentiated and buffered according to the polygon of the settlement boundary: the boundary of the core protection zone is buffered outward by 30m from the settlement boundary; the boundary of the construction control zone is buffered outward by 100m from the core protection zone boundary; and the boundary of the environmental coordination zone is buffered outward by 300m from the construction control zone boundary.

[0067] This step ultimately outputs a conservation value score, conservation zone boundary data, and a conservation value assessment report. The report includes basic settlement information, detailed scores for each dimension, a comprehensive score, recommendations for conservation zone delineation, and targeted conservation measures. In practical application testing, a systematic assessment was conducted on 87 cave dwelling settlements within a county on the Loess Plateau. Of these, 12 settlements were designated as core protected areas, 35 as construction control areas, and 40 as environmental coordination areas. The assessment results showed an 89.7% consistency rate with the expert opinions on cultural relic protection, validating the reliability and practicality of the conservation value assessment model of this invention.

[0068] In a preferred embodiment of the present invention, the protection planning generation step further includes generating differentiated protection strategy recommendations for settlements with different protection levels. For high-value settlements included in the core protection zone, the system automatically generates strict protection strategy recommendations, including prohibiting the demolition of existing cave dwellings, limiting the height and style of new buildings, and maintaining the original alleyway layout and courtyard organization. For medium-value settlements included in the construction control zone, the system generates moderate renewal strategy recommendations, allowing necessary infrastructure improvements and functional updates while maintaining overall landscape harmony. For general settlements included in the environmental coordination zone, the system generates landscape coordination strategy recommendations, focusing on the harmonious control of the surrounding environment of the settlement.

[0069] Preferably, the protection planning generation module also supports multi-scenario simulation. By adjusting the weighting coefficients of each dimension's score and the threshold parameters for zoning, zoning schemes under different protection intensities can be generated for decision-makers to compare and weigh. For example, in a protection-first scenario, the weights of integrity and continuity can be increased, while in a development-coordination scenario, the protection zoning standards can be appropriately lowered, thereby meeting the differentiated protection planning needs of different regions and different stages of development.

[0070] Furthermore, this invention also provides a function for tracking and evaluating the effectiveness of protection measures. By periodically acquiring remote sensing image data of the target area, re-executing the cave dwelling identification and settlement analysis processes, and comparing and analyzing the changes in settlement morphology characteristics and protection value scores before and after the implementation of the protection plan, the effectiveness of the protection measures can be objectively evaluated, providing data support for the dynamic adjustment of the protection plan.

[0071] Reference Figure 2 This invention also provides a remote sensing image intelligent identification and protection planning system for traditional cave dwelling settlements. The system includes a remote sensing data acquisition module 1, a cave dwelling identification module 2, a settlement boundary extraction module 3, a settlement morphology analysis module 4, a historical evolution analysis module 5, and a protection planning generation module 6. The modules are interconnected through a data bus to collaboratively complete the intelligent identification and protection planning tasks of cave dwelling settlements.

[0072] The remote sensing data acquisition module 1 is configured to acquire high-resolution satellite imagery, UAV oblique photography data, and digital elevation model data of the target area, perform geometric correction and coordinate registration processing on the multi-source remote sensing data, and generate remote sensing image data and slope feature maps. Preferably, the remote sensing data acquisition module 1 includes a data interface unit, a geometric correction unit, and a slope calculation unit. The data interface unit is used to interface with the satellite data service platform and the UAV data management system. The geometric correction unit uses a rational polynomial coefficient model to achieve orthorectification. The slope calculation unit generates the slope feature map according to the third-order inverse distance weighted difference algorithm. The remote sensing data acquisition module 1 executes the technical content of step S1 in the method embodiment.

[0073] The cave dwelling identification module 2 is configured to input remote sensing image data and slope feature maps into a terrain-assisted multi-scale cave dwelling semantic segmentation network. It then fuses terrain information through multi-scale feature extraction and an attention mechanism to output a cave dwelling type distribution map. The cave dwelling identification module 2 includes a feature extraction unit, a terrain bypass unit, an attention fusion unit, and a decoding output unit. The feature extraction unit uses a ResNet-101 backbone network to extract multi-scale semantic features; the terrain bypass unit processes the slope feature map to extract terrain semantic information; the attention fusion unit implements channel attention weighting; and the decoding output unit generates semantic segmentation results containing three cave dwelling types. The cave dwelling identification module 2 executes the technical content of step S2 in the method embodiment.

[0074] The settlement boundary extraction module 3 is configured to extract spatial point sets of cave dwellings based on the cave dwelling type distribution map, and to delineate the settlement range using an adaptive density clustering algorithm that integrates spatial correlation constraints, generating settlement boundary polygons. The settlement boundary extraction module 3 includes a point set extraction unit, a correlation calculation unit, a density clustering unit, and a boundary generation unit. The point set extraction unit extracts the centroid coordinates of the cave dwellings through connected component analysis. The correlation calculation unit calculates spatial correlation based on distance, azimuth, and type similarity. The density clustering unit performs neighborhood determination and core point identification based on correlation thresholds. The boundary generation unit uses the AlphaShapes algorithm to generate concave hull boundaries. The settlement boundary extraction module 3 executes the technical content of step S3 in the method embodiment.

[0075] The settlement morphology analysis module 4 is configured to extract settlement scale indicators, layout morphology indicators, cave dwelling density indicators, and courtyard organization pattern indicators based on settlement boundary polygons and cave dwelling type distribution maps, generating settlement morphology feature vectors. The settlement morphology analysis module 4 includes a scale statistics unit, a morphology calculation unit, a density analysis unit, and a courtyard identification unit. Each unit performs feature quantification calculations for its corresponding dimensions and integrates the calculation results into a unified feature vector format. The settlement morphology analysis module 4 executes the technical content of step S4 in the method embodiment.

[0076] The historical evolution analysis module 5 is configured to acquire historical image sequences and perform multi-period collaborative registration with current remote sensing image data. It tracks the expansion, contraction, and disappearance of cave dwelling settlements using a change detection algorithm, generating evolution trajectory data. The historical evolution analysis module 5 includes a historical data interface unit, a multi-period registration unit, and a change detection unit. The historical data interface unit interfaces with a historical image database to acquire remote sensing data from different periods. The multi-period registration unit uses feature point matching and affine transformation to achieve geometric alignment. The change detection unit identifies evolution events through pixel-level difference and object-level analysis. The historical evolution analysis module 5 executes the technical content of step S5 in the method embodiment.

[0077] The protection planning generation module 6 is configured to calculate the protection value score based on settlement morphology feature vectors and evolution trajectory data, using a multi-factor weighted evaluation model. Based on the protection value score, it delineates core protection areas, construction control areas, and environmental coordination areas, outputting protection zone boundary data and a protection value assessment report. The protection planning generation module 6 includes a scoring calculation unit, a zoning unit, and a report generation unit. The scoring calculation unit calculates the scores for each sub-item according to four dimensions: scale, completeness, continuity, and coordination, and then weights and sums them to obtain a comprehensive score. The zoning unit generates three-level protection zone boundaries based on scoring thresholds and buffer rules. The report generation unit integrates the analysis results and outputs a formatted assessment report document. The protection planning generation module 6 executes the technical content of step S6 in the method embodiment.

[0078] During system operation, each module executes processing tasks sequentially according to the data flow. Remote sensing data acquisition module 1 first completes data acquisition and preprocessing, then transmits the generated remote sensing image data and slope feature map to cave dwelling identification module 2. After identifying the cave dwelling type, cave dwelling type distribution map is simultaneously transmitted to settlement boundary extraction module 3 and settlement morphology analysis module 4. The settlement boundary polygons generated by settlement boundary extraction module 3 are transmitted to settlement morphology analysis module 4 and conservation planning generation module 6. Remote sensing data acquisition module 1 also transmits remote sensing image data to historical evolution analysis module 5 for multi-period registration reference. The evolution trajectory data generated by historical evolution analysis module 5 and the feature vectors generated by settlement morphology analysis module 4 serve as input to conservation planning generation module 6, ultimately generating conservation value assessment results and conservation planning recommendations.

[0079] The system provided in this invention can be deployed on servers or workstations equipped with GPU acceleration devices, preferably using NVIDIA Tesla V100 or higher-performance GPUs to support efficient inference of deep learning networks. The system software environment is developed based on the Python language and relies on open-source components such as the PyTorch deep learning framework, the GDAL geographic data processing library, and the GeoPandas spatial analysis library. Under typical configuration, the system takes approximately 2 to 5 minutes to process the complete identification and evaluation process of a single cave dwelling settlement, and has the processing capability to support large-scale census applications at the county or even city level.

[0080] In a preferred embodiment of the present invention, the system further includes a user interaction module and a visualization module. The user interaction module provides a graphical user interface, supporting functions such as parameter configuration, task submission, progress monitoring, and result querying, lowering the barrier to entry for system use and enabling non-technical personnel to easily operate the system to complete the task of identifying and evaluating cave dwelling settlements. The visualization module presents the analysis results in various forms such as maps, charts, and reports, supporting the automatic generation and export of thematic maps such as cave dwelling distribution maps, settlement boundary maps, and protection zoning maps, and supporting interactive display of statistical charts such as radar charts and bar charts for protection value scoring.

[0081] Preferably, the system adopts a modular architecture design, with each functional module exchanging data through standardized interfaces, supporting independent upgrades and replacements of modules. For example, when the deep learning model for cave dwelling identification is updated, the upgrade can be completed simply by replacing the model file in the cave dwelling identification module, without requiring a complete redeployment of the entire system. This modular design improves the system's maintainability and scalability, facilitating functional customization and performance optimization according to actual application needs.

[0082] Furthermore, the system supports a distributed deployment mode. For applications requiring the processing of large-scale regional data, the system can be deployed on a computing cluster, significantly improving processing efficiency through task decomposition and parallel processing. In distributed deployment mode, remote sensing data is divided into multiple sub-tasks according to spatial extent. Each sub-task is executed in parallel on different computing nodes, and finally, the master node aggregates and integrates the processing results of each sub-task. Tests show that with an 8-node cluster configuration, the system's efficiency in processing a provincial-level survey of cave dwelling settlements is approximately 6 times higher than that of a single-node deployment.

[0083] The embodiments of the present invention are not limited to the specific embodiments described above. Those skilled in the art can make various equivalent changes or substitutions based on the technical solutions of the present invention, and all such changes or substitutions should be included within the protection scope of the present invention.

Claims

1. A method for intelligent identification and protection planning of traditional cave dwelling settlements using remote sensing images, characterized in that: Includes the following steps: The remote sensing data acquisition and preprocessing steps involve acquiring high-resolution satellite imagery, UAV oblique photography data, and digital elevation model (DEM) data for the target area. Geometric correction and coordinate registration are performed on the multi-source remote sensing data. A slope feature map is generated based on the DEM data, and the remote sensing image data and slope feature map are output. The intelligent cave dwelling identification step involves inputting the remote sensing image data and slope feature map into a terrain-assisted multi-scale cave dwelling semantic segmentation network. This network includes a backbone feature extraction network, a terrain feature bypass branch, and a multi-scale attention fusion module. By fusing terrain information through multi-scale feature extraction and channel attention mechanisms, a cave dwelling type distribution map is output, including three types: cliff-side cave dwellings, sunken cave dwellings, and independent cave dwellings. The settlement boundary extraction step involves extracting a spatial point set of cave dwellings based on the cave dwelling type distribution map and employing an adaptive density method that integrates spatial correlation constraints. The process involves several steps: First, a density clustering algorithm is used to calculate the spatial correlation between cave dwellings and perform density clustering to generate settlement boundary polygons. Second, a settlement morphology analysis step is performed, based on the settlement boundary polygons and the cave dwelling type distribution map, to extract settlement scale indicators, layout morphology indicators, cave dwelling density indicators, and courtyard organization pattern indicators, generating settlement morphology feature vectors. Third, a historical evolution analysis step is performed, acquiring historical image sequences and co-registering them with remote sensing image data for multiple periods. A change detection algorithm is used to track the expansion, contraction, and disappearance of cave dwelling settlements, generating evolution trajectory data. Fourth, a protection planning generation step is performed, based on the settlement morphology feature vectors and evolution trajectory data, using a multi-factor weighted evaluation model to calculate protection value scores from four dimensions: scale, integrity, continuity, and coordination. Based on the protection value scores, core protection areas, construction control areas, and environmental coordination areas are delineated, and protection zone boundary data and a protection value assessment report are output.

2. The method for intelligent identification and protection planning of traditional cave dwelling settlements using remote sensing images according to claim 1, characterized in that, The spatial resolution of the high-resolution satellite imagery is no less than 0.5m, and the spatial resolution of the digital elevation model data is no less than 5m and the elevation accuracy is no less than 2m.

3. The method for intelligent identification and protection planning of traditional cave dwelling settlements using remote sensing images according to claim 1, characterized in that, The slope feature map is generated using a third-order inverse distance weighted difference algorithm. The slope feature map is used to help distinguish the spatial distribution characteristics of cliff-side cave dwellings, sunken cave dwellings, and independent cave dwellings.

4. The method for intelligent identification and protection planning of traditional cave dwelling settlements using remote sensing images according to claim 1, characterized in that, The training loss function of the terrain-assisted multi-scale cave dwelling semantic segmentation network is a weighted combination of cross-entropy loss and Dice loss, with the weight coefficients of both losses set to 0.

5.

5. The method for intelligent identification and protection planning of traditional cave dwelling settlements using remote sensing images according to claim 1, characterized in that, The spatial correlation degree comprehensively considers three factors: Euclidean distance, azimuth difference, and type similarity between cave dwellings. When the two cave dwellings are of the same type, the type similarity function takes a value of 1.0, and when the cave dwellings are of different types, the value takes a value of 0.

7.

6. The method for intelligent identification and protection planning of traditional cave dwelling settlements using remote sensing images according to claim 1, characterized in that, In the adaptive density clustering algorithm, the correlation threshold is set to 0.3, and the minimum number of points threshold is set to 5.

7. The method for intelligent identification and protection planning of traditional cave dwelling settlements using remote sensing images according to claim 1, characterized in that, The layout morphology indicators include shape index, compactness, and fractal dimension. The shape index reflects the complexity of the settlement boundary, the compactness reflects the degree of concentration of the settlement space, and the fractal dimension is calculated using the box counting method.

8. The method for intelligent identification and protection planning of traditional cave dwelling settlements using remote sensing images according to claim 1, characterized in that, The multi-stage collaborative registration adopts a method that combines feature point matching and affine transformation, and the registration accuracy is controlled within 2 to 3 pixels.

9. The method for intelligent identification and protection planning of traditional cave dwelling settlements using remote sensing images according to claim 1, characterized in that, In the multi-factor weighted evaluation model, the weight coefficients for scale score, integrity score, continuity score, and coordination score are 0.2, 0.3, 0.25, and 0.25, respectively. When the conservation value score is greater than or equal to 70, it is classified as a core protected area; when the score is greater than or equal to 50 and less than 70, it is classified as a construction control area; and when the score is less than 50, it is classified as an environmental coordination area.

10. A remote sensing image intelligent identification and protection planning system for traditional cave dwelling settlements, used to implement the remote sensing image intelligent identification and protection planning method for traditional cave dwelling settlements as described in any one of claims 1-9, characterized in that, include: The remote sensing data acquisition module is configured to acquire high-resolution satellite imagery, UAV oblique photography data, and digital elevation model (DEM) data of the target area; perform geometric correction and coordinate registration processing on the multi-source remote sensing data; calculate and generate a slope feature map based on the DEM data; and output remote sensing imagery data and the slope feature map. The cave dwelling identification module is configured to input the remote sensing imagery data and the slope feature map into a terrain-assisted multi-scale cave dwelling semantic segmentation network; fuse terrain information through multi-scale feature extraction and channel attention mechanisms; and output a cave dwelling type distribution map including three types: cliff-side cave dwellings, sunken cave dwellings, and independent cave dwellings. The settlement boundary extraction module is configured to extract a spatial point set of cave dwellings based on the cave dwelling type distribution map and use spatial fusion... An adaptive density clustering algorithm with correlation constraints generates settlement boundary polygons; a settlement morphology analysis module is configured to extract settlement scale indicators, layout morphology indicators, cave density indicators, and courtyard organization pattern indicators based on the settlement boundary polygons and the cave dwelling type distribution map, and generate settlement morphology feature vectors; a historical evolution analysis module is configured to acquire historical image sequences and perform multi-period collaborative registration with the remote sensing image data, and generate evolution trajectory data through a change detection algorithm; a protection planning generation module is configured to calculate protection value scores using a multi-factor weighted evaluation model based on the settlement morphology feature vectors and the evolution trajectory data, and delineate core protection areas, construction control areas, and environmental coordination areas based on the protection value scores.

Images