Surveying and mapping geographic information analysis method and system based on machine vision and medium

Abstract

The invention relates to the technical field of machine vision, and discloses a surveying and mapping geographic information analysis method and system based on machine vision and a medium. The method comprises the following steps: preprocessing multi-source remote sensing data to obtain a standardized surveying and mapping data set, and extracting ground feature edge and texture information to obtain a multi-scale spatial feature and a semantic segmentation result; carrying out multi-source data integration to obtain a unified geographic element representation diagram, and carrying out geographic element identification on the unified geographic element representation diagram to obtain a surveying and mapping geographic element set; performing Bayesian inference of spatial uncertainty quantization and parameter uncertainty quantization on the surveying and mapping geographic element set to obtain a comprehensive uncertainty evaluation result; and constructing vectorized geographic information based on the surveying and mapping geographic element set and the comprehensive uncertainty evaluation result. According to the method and the device, the edge information of the geographic elements is effectively reserved, the identification precision of the geographic elements is improved, the geographic information expression is more structured, and automatic updating is realized.

Description

Technical Field

[0001] This application relates to the field of machine vision technology, and particularly to a method, system, and medium for analyzing surveying and mapping geographic information based on machine vision. Background Art

[0002] With the rapid development of remote sensing technology, the acquisition of multi-source remote sensing data (optical images, radar data, point cloud data, etc.) has become increasingly convenient, providing a rich data basis for the analysis of surveying and mapping geographic information. However, these multi-source remote sensing data have characteristics such as strong heterogeneity, high dimensionality, and complex noise. Traditional geographic feature extraction methods mainly rely on manually designed features and rules, making it difficult to effectively utilize the complementary information of multi-source data and resulting in low recognition accuracy under complex terrain conditions. Especially in complex scenarios such as shadow areas, cloud cover, and urban-rural junctions, existing methods often have misjudgments or missed judgments, unable to meet the requirements of high-precision surveying and mapping.

[0003] Another key challenge faced by current surveying and mapping geographic information analysis is the problem of uncertainty assessment. Traditional geographic information extraction methods only output deterministic results and lack a quantitative assessment of the reliability of the recognition results, which makes subsequent decision-making lack a basis for risk assessment. Most existing uncertainty assessment methods are based on simple statistical models and cannot accurately describe the spatial correlation and parameter uncertainty in complex geographic scenarios. Especially when environmental factors such as lighting conditions, seasonal changes, and image resolution change, the comprehensive uncertainty assessment results are often unreliable, limiting the promotion and use of surveying and mapping results in key application fields. Summary of the Invention

[0004] This application provides a method, system, and medium for analyzing surveying and mapping geographic information based on machine vision. This application effectively retains the edge information of geographic features, improves the recognition accuracy of geographic features, makes the expression of geographic information more structured, and realizes automatic update.

[0005] In a first aspect, this application provides a method for analyzing surveying and mapping geographic information based on machine vision. The method for analyzing surveying and mapping geographic information based on machine vision includes: Preprocess multi-source remote sensing data to obtain a standardized surveying and mapping data set, and extract ground object edge and texture information from the standardized surveying and mapping data set to obtain multi-scale spatial features and semantic segmentation results; Integrate multi-source data based on the multi-scale spatial features and the semantic segmentation results to obtain a unified geographic feature representation map, and identify geographic features from the unified geographic feature representation map to obtain a set of surveying and mapping geographic features; Perform Bayesian inference of spatial uncertainty quantification and parameter uncertainty quantification on the set of surveying and mapping geographic features to obtain a comprehensive uncertainty assessment result; Construct vectorized geographic information based on the set of surveyed geographical elements and the comprehensive uncertainty assessment result.

[0006] In a second aspect, the present application provides a surveyed geographical information analysis system based on machine vision, and the surveyed geographical information analysis system based on machine vision includes: A preprocessing module, configured to preprocess multi-source remote sensing data to obtain a standardized surveyed data set, and extract ground object edge and texture information from the standardized surveyed data set to obtain multi-scale spatial features and semantic segmentation results; A geographical element recognition module, configured to perform multi-source data integration based on the multi-scale spatial features and the semantic segmentation results to obtain a unified geographical element representation map, and perform geographical element recognition on the unified geographical element representation map to obtain a set of surveyed geographical elements; A Bayesian inference module, configured to perform Bayesian inference on spatial uncertainty quantification and parameter uncertainty quantification on the set of surveyed geographical elements to obtain a comprehensive uncertainty assessment result; A construction module, configured to construct vectorized geographic information based on the set of surveyed geographical elements and the comprehensive uncertainty assessment result.

[0007] In a third aspect, a computer-readable storage medium is provided, and instructions are stored in the computer-readable storage medium. When the instructions run on a computer, the computer is caused to execute the above-mentioned surveyed geographical information analysis method based on machine vision.

[0008] In the technical solution provided by the present application, through the adaptive preprocessing of multi-source remote sensing data and multi-scale feature extraction, the technical problems such as low recognition accuracy, lack of uncertainty assessment, and difficulty in information update in the traditional method in complex scenarios are effectively solved. The dynamic parameter adjustment noise reduction algorithm and the image enhancement technology based on deep learning introduced in the present application achieve the effective retention of the edge information of geographical elements; the multi-scale feature extraction network with a five-layer pyramid structure combined with the double-branch processing mechanism of direction sensitivity type and region aggregation type significantly enhances the recognition ability of different types of geographical elements; the feature transfer network of the "feature-space-semantic" three-layer transfer framework effectively solves the problem of unbalanced feature expression in multi-modal data fusion; the geographical feature perception residual block and the three-branch network architecture improve the recognition accuracy of geographical elements; the Bayesian deep learning method combined with variational inference realizes the accurate quantification of the uncertainty of the recognition result; the geographical element relationship reasoning based on the graph neural network and the change detection technology of the deep similarity measurement network make the geographical information expression more structured and realize automatic update. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the accompanying drawings required for the description of the embodiments. Obviously, the accompanying drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can also be obtained based on these drawings.

[0010] Figure 1 It is a schematic diagram of an embodiment of the method for analyzing surveying and mapping geographic information based on machine vision in the embodiments of the present application; Figure 2 It is a schematic diagram of an embodiment of the system for analyzing surveying and mapping geographic information based on machine vision in the embodiments of the present application. Detailed implementation manners

[0011] The embodiments of the present application provide a method, a system and a medium for analyzing surveying and mapping geographic information based on machine vision. The terms "first", "second", "third", "fourth", etc. (if any) in the specification, claims and above-mentioned accompanying drawings of the present application are used to distinguish similar objects, and do not necessarily need to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments described here can be implemented in an order different from that shown or described here. In addition, the term "including" or "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device including a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0012] For ease of understanding, the following describes the specific process of the embodiments of the present application. Please refer to Figure 1 , an embodiment of the method for analyzing surveying and mapping geographic information based on machine vision in the embodiments of the present application includes: Step S101: Preprocess the multi-source remote sensing data to obtain a standardized surveying and mapping data set, and extract the ground object edge and texture information from the standardized surveying and mapping data set to obtain multi-scale spatial features and semantic segmentation results; It can be understood that the execution subject of the present application can be a system for analyzing surveying and mapping geographic information based on machine vision, or a terminal or a server. Specifically, it is not limited here. The embodiments of the present application take the server as the execution subject as an example for illustration.

[0013] Specifically, for the automated identification of sensor parameters of multi-source remote sensing data, the data source type is classified and feature parameters are extracted by automatically parsing the data element information and combining with a deep learning model. These data include high-resolution optical images, radar data, and lidar point cloud data. The spatial resolution, band range, and imaging characteristics of each data source are calibrated to generate the parameter calibration results of each data source. According to the calibration results, the color distribution histogram of the optical image is calculated. By analyzing the color distribution characteristics of the pixels in the image, the corresponding histogram mapping relationship is established to obtain the image characteristic analysis data. These analysis data are used for the noise reduction processing of the optical image. The noise reduction process uses an adaptive filtering algorithm that combines Gaussian filtering and bilateral filtering, automatically adjusts the filtering parameters according to the image complexity, and achieves the noise reduction effect of edge preservation, so as to suppress noise interference to the greatest extent while maintaining the clarity of the ground object boundaries. At the same time, the outlier detection and elimination processing are performed on the point cloud data. The outlier detection algorithm based on density clustering is used to eliminate the noise points and isolated points in the low-density area to obtain a more accurate topographic feature expression. The compensated processing of the cloud cover and shadow areas is performed on the denoised optical image. The generative adversarial network based on deep learning is used for image enhancement. By identifying the cloud cover and shadow areas and automatically generating the compensated pixel information, the enhanced processing image is formed. The enhanced optical image, the point cloud data set after outlier elimination, and the radar data are subjected to resampling and registration processing of the spatial resolution. The image registration algorithm based on the multi-scale pyramid is used, and the spatial position is calibrated in combination with the geographic coordinate system to ensure the precise correspondence of the multi-source data in the spatial position, and the standardized mapping data set is obtained. The standardized mapping data set is input into the multi-scale feature extractor for the extraction of ground object edge and texture information. The feature extractor uses a five-layer pyramid feature extraction network, and different sizes of convolutional kernels (such as 3×3, 5×5, 7×7, 9×9, 11×11) are used for convolutional operations at different levels to capture multi-scale spatial features, and the extraction effect of different scale features is optimized through the adaptive receptive field adjustment mechanism. During the feature extraction process, specific direction-sensitive convolutional modules and region aggregation convolutional modules are used to detect the edge information of linear ground objects (such as roads, rivers) and regional ground objects (such as buildings, water bodies), and at the same time, local response normalization operations are introduced to enhance high-frequency information, and refined ground object edge features and texture information are obtained.

[0014] The standardized mapping data set is input into a multi-scale feature extractor for convolution operations. A multi-scale convolutional network with a five-layer pyramid structure is adopted. At each layer, convolution kernels of different sizes (such as 3×3, 5×5, 7×7, 9×9, 11×11) are used to perform convolution calculations on the input data, so as to extract initial feature maps of different scales. Due to the significant differences in the distribution of feature objects at different scales, the multi-scale convolution mechanism can effectively capture the spatial characteristics of linear feature objects (such as roads and rivers) and regional feature objects (such as buildings and water bodies). The initial feature maps are respectively input into a direction-sensitive convolutional unit and a region-aggregating convolutional unit for the extraction of specific features. The direction-sensitive convolutional unit adopts a convolutional kernel with direction selectivity. By performing convolution operations in different directions such as horizontal, vertical, and diagonal, it effectively captures the edge information and direction characteristics of linear feature objects, and prominently shows the boundary features of linear feature objects such as roads and rivers in the feature map. The region-aggregating convolutional unit, on the other hand, combines a larger convolutional receptive field with an adaptive weight allocation mechanism to achieve the overall feature aggregation of regional feature objects, enhancing the feature expression ability of regional feature objects such as buildings and water bodies. These two convolutional units work together, enabling the linear feature object features and regional feature object features to be strengthened and finely expressed respectively. The extracted linear feature object features and regional feature object features are input into a local response normalization layer for enhancing high-frequency information. Local response normalization effectively highlights the object boundary and texture features and improves the resolution ability of edge details by suppressing the low-frequency components in the feature map and enhancing the high-frequency details, generating an enhanced edge and texture feature map. On the basis of local response normalization, channel attention weighting processing is performed on the enhanced edge and texture feature map. The channel attention mechanism adaptively adjusts the influence degree of different channel information by calculating the weight contributions of different spectral channels in feature extraction, making the recognition of different object materials more targeted. The multi-scale spatial features weighted by channel attention and the enhanced edge and texture feature map are jointly input into a fully convolutional neural network for semantic segmentation. The semantic segmentation network is based on the pre-trained U-Net architecture and performs pixel-by-pixel classification in combination with multi-scale features to achieve accurate recognition of object categories and obtain the semantic segmentation result.

[0015] Step S102: Integrate multi-source data based on the multi-scale spatial features and the semantic segmentation result to obtain a unified geographical feature representation map, and perform geographical feature recognition on the unified geographical feature representation map to obtain a mapping geographical feature set; Specifically, the multi-scale spatial features and semantic segmentation results are encoded. In the encoding process, a multi-layer convolutional neural network is used to represent the features in a high-dimensional manner, and batch normalization and the ReLU activation function are combined to enhance the non-linear expression ability of the features, obtaining the encoded feature representation. The encoded feature representation is separated into a core representation transmitted through an identity mapping and a variable representation adjusted through an adaptive residual connection. The core representation retains the inherent characteristics of geographical elements and stably transmits basic information, while the variable representation dynamically adjusts the features according to the environmental changes of multi-source data by introducing an adaptive residual connection to adapt to different data characteristics. These two feature forms jointly form a dual-path feature flow. Cross-modal feature interaction is performed on the dual-path feature flow. The information flow between different data modalities is controlled through a bidirectional gating mechanism, the correlation between features of different modalities is calculated, and the correlated features are integrated to generate a preliminary fusion feature. Based on the preliminary fusion feature, the feature expression is enhanced by calculating a spatial attention map and a channel attention map. The spatial attention map is used to capture the importance information of geographical elements in terms of spatial location, while the channel attention map is used to measure the weight contribution of different spectral channels or data modalities in the feature extraction process. A feature attention weight matrix is constructed through these two attention maps. According to the feature attention weight matrix, the preliminary fusion feature is adaptively re-weighted. In this process, the semantic segmentation result is input into the local context aggregation unit as prior knowledge for feature guidance. Through the guidance of context information, the expression of features in semantically related regions becomes more accurate, generating context-enhanced features. A cross-scale residual fusion operation is performed on the context-enhanced features. By introducing a multi-scale feature fusion mechanism, the context-enhanced features of different scales are residually connected to achieve adaptive supplementation and information enhancement of cross-scale features, generating a unified geographical element representation map. The unified geographical element representation map is input into a geographical element recognition model, and a convolutional neural network based on an improved ResNet architecture is used for refined recognition. This recognition model combines a feature selection gating mechanism and a multi-task branch network to simultaneously recognize and regress the boundaries, categories, and attributes of geographical elements, thereby obtaining a set of surveying and mapping geographical elements, including accurate geographical element boundaries, detailed category labels, and attribute parameters (such as height, width, area, etc.), forming a structured geographical information representation.

[0016] The unified geographical feature representation map is input into the backbone network, which adopts an improved ResNet architecture and combines geographical feature-aware residual blocks for feature extraction. The geographical feature-aware residual blocks expand the receptive field by introducing a dilated convolution sequence, and the dilation rates are set to 1, 2, 4, and 8 in sequence, enabling the network to capture the feature information of geographical elements at different scales and obtaining multi-scale receptive field mapping features. The contour features of geographical elements are extracted from the multi-scale receptive field mapping features. Through a depthwise separable convolutional network combined with a boundary attention mechanism, the boundary information of geographical elements is captured, and a boundary mask of geographical elements is generated. This boundary mask can effectively identify the contours of different geographical elements, such as building contours, road boundaries, and water system boundaries. At the same time, the spatial distribution relationship of geographical elements is analyzed from the multi-scale receptive field mapping features. Through a context attention module, the spatial correlation of ground objects is captured, and a class probability map of geographical elements is generated in combination with a fully convolutional neural network. This probability map can predict the probability that each pixel belongs to different ground object classes. Based on the multi-scale receptive field mapping features, the attribute information of geographical elements is extracted. Through a fully connected layer combined with an attention mechanism, regression calculations are performed on the geometric and physical attributes of ground objects to obtain the attribute parameter values of geographical elements, such as the height of buildings, the width of roads, and the area of water bodies. The boundary mask of geographical elements is subjected to boundary refinement processing. An edge optimization algorithm based on conditional random fields is used to optimize the rough boundary at the pixel level and, in combination with a class label analysis module, refine the class information in the boundary area to generate the precise boundary and accurate class information of geographical elements. This process can effectively avoid the problem of blurred boundaries caused by resolution differences or noise interference, thereby improving the accuracy of element recognition. After completing the boundary refinement and class label analysis, the obtained precise boundaries, class information, class probability maps, and attribute parameter values of geographical elements are associated and integrated. Through a multi-dimensional feature fusion module, information from different sources is cross-validated and weighted to ensure the integrity and consistency of the recognition results, and finally a collection of mapping geographical elements including building contours, road networks, and water system boundaries is formed.

[0017] Step S103: Perform Bayesian inference on the collection of mapping geographical elements for spatial uncertainty quantification and parameter uncertainty quantification to obtain a comprehensive uncertainty assessment result; Specifically, probability distribution data of the boundary positions and attribute parameters of geographical features are calculated based on the set of surveyed and mapped geographical features. Among them, the probability distribution of the boundary positions is obtained by performing multiple forward propagations during the network inference stage through Monte Carlo rejection sampling, while the probability distribution of the attribute parameters is estimated by introducing the variational inference method into the Bayesian neural network. These probability distribution data provide basic information for subsequent uncertainty quantification. Based on the probability distribution data, the coefficient of variation of the boundary positions of geographical features is calculated. By statistically analyzing the probability distributions of the boundary positions at different spatial locations, the variance and mean of the boundary positions are obtained, and the coefficient of variation is calculated using these data. This coefficient of variation reflects the degree of uncertainty of the boundary positions. Adaptive kernel density estimation is performed according to the coefficient of variation. By dynamically adjusting the kernel function parameters, the kernel density function can more accurately fit the boundary position distributions of different types of geographical features, resulting in a more refined spatial uncertainty quantification result. At the same time, for the attribute parameters of geographical features, probability estimation of the attribute parameters of geographical features is performed through a Bayesian neural network. The variational inference method is used to approximate the posterior distribution, and a Bayesian loss function is constructed by combining the negative log-likelihood and KL divergence terms. This loss function can effectively balance the prediction accuracy and distribution fitting ability of the model, resulting in a parameter uncertainty quantification result. Based on the spatial uncertainty quantification result and the parameter uncertainty quantification result, the two are jointly modeled. By constructing a Bayesian loss function for spatial correlation modeling, a variational graph autoencoder is introduced to model the spatial correlation of the geographical feature distribution, capture the implicit spatial associations between different ground objects, and enhance the accuracy of uncertainty assessment at the spatial feature level, resulting in an uncertainty assessment index with enhanced spatial correlation. Environmental factor sensitivity analysis is performed on the uncertainty assessment index with enhanced spatial correlation. The sensitivity analysis module simulates the influence of different environmental factors on the recognition accuracy of geographical features by introducing environmental factor parameters (such as lighting conditions, seasonal changes, image resolution, etc.), combines with the uncertainty assessment index with enhanced spatial correlation, quantifies the influence degree of environmental factor changes on the recognition result, and generates a sensitivity map and an uncertainty change curve, forming a comprehensive uncertainty assessment result including the class probability distribution, position uncertainty, and attribute parameter probability distribution of geographical features.

[0018] Step S104: Construct vectorized geographic information based on the set of surveyed and mapped geographical features and the comprehensive uncertainty assessment result.

[0019] Specifically, a multi-level geographical feature data structure is created based on the collection of surveying and mapping geographical features. This data structure consists of four levels: the point feature layer, the line feature layer, the surface feature layer, and the composite feature layer. The point feature layer records the inflection points and intersection points at key positions in the geographical space. The line feature layer contains information about linear features such as roads and rivers. The surface feature layer stores regional features such as buildings and water bodies. The composite feature layer is used to represent complex geographical units, such as functional areas like residential areas and industrial areas. Based on the comprehensive uncertainty assessment results, the threshold parameters of the features in the line feature layer are adjusted. By calculating the position variation coefficient of the line feature boundaries in the uncertainty quantification results, the boundary segments with lower accuracy are automatically removed, and the boundary is simplified using the adaptive Douglas-Peucker algorithm to obtain the refined line feature data. After generating the multi-level geographical feature data structure, the geographical features are regarded as network nodes, and the spatial relationships between the features (such as proximity relationships, connectivity relationships, inclusion relationships, etc.) are regarded as network edges. Corresponding attribute feature vectors are assigned to each node and edge, and in this way, a geographical feature relationship network is constructed. This relationship network can capture the spatial topological characteristics between geographical features, and through the assignment of feature vectors, effectively associate the functional attributes and spatial connections between different features. On this basis, element relationship learning is carried out on the geographical feature relationship network. A graph neural network is introduced for network embedding, and the feature interaction relationship between nodes is captured through convolutional operations to automatically identify element groups and their functional attributes, forming cluster features and functional patterns of different features in terms of spatial relationships. The identification of element groups can achieve the automatic clustering of similar features and infer regions with specific functional characteristics. For example, residential areas can be identified by the arrangement of buildings and the structure of the road network, and industrial areas can be identified by the distribution of industrial plants and logistics channels. The element groups and their functional attributes are compared with historical data, combined with the comprehensive uncertainty assessment results, and the degree of change between the collection of surveying and mapping geographical features and historical data is calculated through a deep similarity measurement network to identify the change information of geographical features and generate change detection results. According to the geographical feature change information, the refined line feature data, and the functional attributes of the element groups, combined with the spatial relationship and time dimension change characteristics, vectorized geographical information is generated through a topological correction and data fusion mechanism. The vectorized output includes the boundary information of basic geographical features such as building outlines, road networks, and water system boundaries, as well as feature attributes, functional characteristics, and feature change states.

[0020] In the embodiments of the present application, through the adaptive preprocessing of multi-source remote sensing data and multi-scale feature extraction, technical problems such as low recognition accuracy, lack of uncertainty assessment, and difficulty in information update in traditional methods in complex scenarios are effectively solved. The dynamic parameter adjustment noise reduction algorithm and the image enhancement technology based on deep learning introduced in the present application achieve the effective retention of the edge information of geographical elements; the multi-scale feature extraction network with a five-layer pyramid structure combined with a double-branch processing mechanism of direction-sensitive type and region-aggregation type significantly enhances the recognition ability of different types of geographical elements; the feature transfer network of the "feature-space-semantics" three-layer transfer framework effectively solves the problem of unbalanced feature expression in multi-modal data fusion; the geographical feature-aware residual block and the three-branch network architecture improve the recognition accuracy of geographical elements; the precise quantification of the uncertainty of the recognition result is achieved through the Bayesian deep learning method combined with variational inference; the change detection technology of geographical element relationship reasoning based on graph neural network and deep similarity measurement network makes the geographical information expression more structured and realizes automatic update.

[0021] In a specific embodiment, the process of executing step S101 may specifically include the following steps: Automatically identify the sensor parameters of the multi-source remote sensing data to obtain the parameter calibration results of each data source. The multi-source remote sensing data includes optical images, radar data, and point cloud data; Calculate the color distribution histogram of the optical image according to the parameter calibration results to obtain the image characteristic analysis data; Perform noise reduction processing on the optical image based on the image characteristic analysis data to obtain the edge-preserving noise-reduced image. At the same time, perform outlier detection on the point cloud data to obtain the point cloud data set after removing outliers; Perform compensation processing on the edge-preserving noise-reduced image for cloud coverage and shadow areas to obtain the enhanced processing image; Perform spatial resolution resampling and registration processing on the enhanced processing image, radar data, and the point cloud data set after removing outliers to obtain the standardized mapping data set; Input the standardized mapping data set into the multi-scale feature extractor to extract the ground object edge and texture information, and obtain the multi-scale spatial features and semantic segmentation results.

[0022] Specifically, a multi-source data automatic recognition and calibration module is constructed. This module automatically recognizes different data sources based on a strategy that combines metadata parsing, deep learning classification models, and data feature analysis, and performs parameter calibration according to data types and sensor characteristics. For optical images, the sensor model, shooting angle, spectral band information, and resolution parameters are obtained by parsing the image metadata. At the same time, a convolutional neural network is combined for image content classification verification to ensure the consistency between the metadata and the image content. For radar data, a synthetic aperture radar metadata parsing and feature matching algorithm is adopted. The parameters such as polarization mode, incident angle, and echo intensity of the image are analyzed for automatic calibration, and feature extraction is performed on different polarization channels to verify the accuracy of the parameters. For point cloud data, by combining point cloud metadata parsing with density distribution analysis, parameters such as the scanning mode, point cloud density, scanning angle, and distance resolution of the lidar are automatically determined to achieve accurate point cloud data calibration. After completing the automatic recognition and parameter calibration, the parameter calibration results of each data source are uniformly formatted to obtain the parameter calibration results of each data source. According to the parameter calibration results, the color distribution histogram of the optical image is calculated. The optical image is grayscaled and split into RGB channels, and the pixel intensity histograms of each channel are calculated respectively. Through normalization processing, a standardized color distribution histogram is obtained, which reflects the color characteristics, brightness distribution, and image contrast characteristics of the image. Based on the analysis of the image characteristics, data, noise reduction processing is performed on the optical image. An edge-preserving noise reduction method that combines adaptive Gaussian filtering and bilateral filtering is adopted. Gaussian filtering can effectively remove the random noise of the image, while bilateral filtering combines the information in the spatial domain and the pixel intensity domain to retain the edge details of the image and avoid the occurrence of blurred boundaries, resulting in an edge-preserving noise-reduced image. While performing noise reduction processing, outlier detection is performed on the point cloud data. By analyzing the spatial density distribution of the point cloud data through a density clustering algorithm, outliers in the low-density area are identified and removed, obtaining a point cloud data set after removing outliers. This process effectively avoids terrain feature distortion caused by isolated points or noise points in the point cloud data. Compensation processing for cloud cover and shadow areas is performed on the edge-preserving noise-reduced image. An image restoration technology based on a generative adversarial network is adopted. By inputting the image into a pre-trained adversarial network, semantic filling is performed on the cloud cover area, and adaptive pixel compensation is combined with the texture characteristics of the adjacent area. At the same time, for the shadow area, the spectral characteristics of the occluded area are restored through a brightness correction and color transfer algorithm, obtaining an enhanced processed image. Spatial resolution resampling and registration processing are performed on the enhanced optical image, radar data, and the point cloud data set after removing outliers.The nearest neighbor interpolation method and bicubic interpolation method are used to match the spatial resolution of optical images, and the radar data and point cloud data are accurately registered with the optical images in terms of spatial position through the multi-scale pyramid matching algorithm. During this process, a similarity metric function (such as mutual information or mean square error) is introduced to optimize and adjust the registration results, ensuring the accurate correspondence of multi-source data in spatial position and generating a standardized mapping dataset. When the standardized mapping dataset is input into the multi-scale feature extractor for extracting ground object edge and texture information, a multi-scale feature extraction network with a five-layer pyramid structure is constructed. Each layer uses convolution kernels of different sizes (such as 3×3, 5×5, 7×7, 9×9, 11×11) for convolution operations to capture geographical feature characteristics at different scales. This feature extractor combines two mechanisms: direction-sensitive convolution and region-aggregation convolution. Among them, direction-sensitive convolution extracts features for linear ground objects (such as roads and rivers). Through convolution operations in horizontal, vertical, and diagonal directions, the boundary characteristics of linear ground objects are captured. While region-aggregation convolution aggregates features for regional ground objects (such as buildings and water bodies), and optimizes the extraction of regional features through an adaptive receptive field adjustment mechanism. After multi-scale feature extraction, the feature maps are input into the local response normalization layer for high-frequency information enhancement processing, so that the edge and texture features of ground objects are further retained. The extracted feature maps are weighted and optimized through a channel attention module, and the features are adaptively re-weighted according to the importance of different data channels to highlight key features. Then, the enhanced feature maps are input into a fully convolutional neural network for semantic segmentation. This network is based on the U-Net architecture, restores the spatial resolution layer by layer through an encoder-decoder structure, and realizes pixel-by-pixel semantic classification to obtain multi-scale spatial features and semantic segmentation results, and finally forms a complete feature expression including ground object categories, boundary information, and texture features.

[0023] In a specific embodiment, the process of performing the step of inputting the standardized mapping dataset into the multi-scale feature extractor to extract ground object edge and texture information and obtaining multi-scale spatial features and semantic segmentation results may specifically include the following steps: Input the standardized mapping dataset into the multi-scale feature extractor for convolution operations to obtain multiple initial feature maps; Input the multiple initial feature maps into the direction-sensitive convolution unit and the region-aggregation convolution unit respectively for processing to obtain linear ground object features and regional ground object features; Input the linear ground object features and regional ground object features into the local response normalization layer for high-frequency information enhancement processing to obtain enhanced edge and texture feature maps; Perform channel attention weighting on the enhanced edge and texture feature maps to obtain multi-scale spatial features, and input the enhanced edge and texture feature maps into a fully convolutional neural network for semantic segmentation to obtain semantic segmentation results.

[0024] Specifically, a feature extractor with multi-scale feature extraction capabilities is constructed. This extractor adopts a five-layer pyramid structure, and each layer performs convolution operations using convolution kernels of different sizes. The sizes of these convolution kernels are 3×3, 5×5, 7×7, 9×9, and 11×11 respectively. Each convolution kernel performs convolution calculations on the input data in the spatial receptive fields of different scales, thereby capturing the edges, textures, and spatial distribution characteristics of geographical features at different resolutions. Taking the standardized mapping dataset as the input, through the layer-by-layer convolution operations of the feature extractor, multiple initial feature maps are formed. These feature maps contain spatial information at different scales and retain the spatial consistency among multi-source data. Since different geographical features have different spatial scale characteristics in remote sensing images, for example, linear features such as roads and rivers can be captured within a smaller receptive field, while regional features such as buildings and water bodies require a larger receptive field to be completely recognized. Therefore, multi-scale convolution can effectively cover the characteristics of features at different scales and improve the integrity and accuracy of feature expression. The multiple initial feature maps are respectively input into the direction-sensitive convolution unit and the region-aggregation convolution unit for feature processing. The direction-sensitive convolution unit adopts multi-directional convolution kernels. By performing convolution calculations in the horizontal direction, vertical direction, 45-degree direction, and 135-degree direction, it captures the direction information of linear features, extracts the boundary features of linear features such as roads and rivers, and performs direction-aware feature enhancement on these boundaries. At the same time, the region-aggregation convolution unit captures the overall features of regional features through the introduction of an adaptive aggregation mechanism, aggregating regional features at different scales in space and capturing the shape, area, and texture characteristics of regional features such as buildings and water bodies under a larger receptive field. The combination of the direction-sensitive convolution unit and the region-aggregation convolution unit enables the linear feature and the regional feature to be respectively strengthened in the feature map. The linear feature and the regional feature are input into the local response normalization layer for enhancing high-frequency information. The local response normalization layer effectively suppresses the influence of low-frequency components in the feature map by normalizing the activation responses of each convolution kernel, while enhancing the expression of high-frequency information, thereby highlighting the edge and texture characteristics of the features. By adaptively adjusting the normalization parameters, local response normalization dynamically adjusts the feature weights in feature maps of different scales to ensure the balanced expression of boundary information and regional features at different spatial scales, thereby generating enhanced edge and texture feature maps. Channel attention weighting processing is performed on the enhanced edge and texture feature maps. The channel attention mechanism dynamically adjusts the weight contributions of each channel in feature expression by calculating the importance of different feature channels, thereby highlighting key feature information and improving the feature selection ability of the model. During the channel attention weighting process, the spatial information of the feature map is compressed through global average pooling, mapping the spatial information of the feature map into a global feature vector. Then, the weights of each channel are calculated through a two-layer fully connected network and remapped to the original feature map to achieve channel-level adaptive weighting.Through the channel attention mechanism, the attention to important feature features is enhanced, and the interference of redundant features is suppressed, so as to obtain multi-scale spatial features with better feature expression ability. The enhanced edge and texture feature maps are input into a fully convolutional neural network for semantic segmentation. This network adopts an improved version of the U-Net architecture, and decodes the feature maps layer by layer through an encoder-decoder structure, gradually restoring the spatial resolution, and combining multi-scale features to achieve pixel-by-pixel classification, thereby identifying the feature categories. In the encoding stage, multi-scale convolution is used to capture feature features of different scales, and residual connections are used to retain key boundary information and texture features; in the decoding stage, deconvolution operations are used to restore spatial information, and skip connections are combined to introduce the features of the encoding stage into the decoding path, so as to ensure the complete fusion of semantic information and spatial information. Through the semantic segmentation process of the fully convolutional neural network, basic feature categories such as roads, rivers, buildings, and water bodies are identified, and different types of features in complex scenes are accurately distinguished, realizing the refined expression of multi-scale spatial features.

[0025] In a specific embodiment, the process of executing step S102 may specifically include the following steps: Perform encoding processing on the multi-scale spatial features and the semantic segmentation results to obtain an encoded feature representation; Separate the encoded feature representation into a core representation passed through an identity mapping and a variable representation adjusted through an adaptive residual connection, and use the core representation and the variable representation as a two-path feature flow; Perform cross-modal feature interaction on the two-path feature flow to obtain a preliminary fusion feature; Calculate a spatial attention map and a channel attention map based on the preliminary fusion feature, and construct a feature attention weight matrix for the spatial attention map and the channel attention map; Adaptive reweight the preliminary fusion feature according to the feature attention weight matrix, and input the semantic segmentation result as prior knowledge into the local context aggregation unit for feature guidance to obtain a context-enhanced feature; Perform a cross-scale residual fusion operation on the context-enhanced feature to obtain a unified geographic feature representation map; Perform geographic feature recognition on the unified geographic feature representation map to obtain a set of surveying and mapping geographic features.

[0026] Specifically, the multi-scale spatial features and the semantic segmentation results are used as inputs, and a feature encoder based on a convolutional neural network is used to perform feature compression and high-dimensional feature mapping on them. The feature encoder consists of multiple convolutional units and batch normalization modules. By performing convolution layer by layer, the spatial dimension of the feature map is gradually compressed, while the feature dimension is increased. The ReLU activation function is combined to enhance the non-linear expression ability of the features. The convolutional units capture the local information of the ground object features through convolutional kernels of different sizes and gradually form the high-dimensional mapping results of the multi-scale spatial features. The semantic segmentation results are used as auxiliary information during the feature encoding process and participate in feature reshaping together with the multi-scale features. After this encoding process, the encoded feature representation is obtained. The encoded feature representation is separated into a core representation passed through an identity mapping and a variable representation adjusted through an adaptive residual connection. The core representation retains the basic spatial and semantic information of the features through a direct identity mapping path. This information remains stable throughout the feature processing process and is not affected by changes in the external environment, thus ensuring that the model can maintain feature consistency at different scales. The variable representation is adjusted through an adaptive residual connection. The residual connection uses feature maps of different scales and performs weighted fusion of the feature information of different layers through a weight dynamic adjustment mechanism, enabling the variable representation to flexibly adapt to the changes of ground object features in different modal data, thereby improving the generalization ability of the features. The core representation and the variable representation serve as a dual-path feature flow. One path stably transmits the core information, and the other path dynamically adjusts the feature weights through the residual mechanism. The two together constitute the feature expression structure. Cross-modal feature interaction is performed on the dual-path feature flow through a bidirectional gating unit to control the information interaction between different modal feature flows. The bidirectional gating unit contains a parameterized gating mechanism for adaptively controlling the information flow of different modal features during the interaction process. By calculating the feature attention weights on each path, the transmission ratio of the features is dynamically adjusted, thereby effectively maintaining the information integrity of different modal features when fusing the features. Through cross-modal feature interaction, information fusion from different data sources and different perception scales is achieved, and preliminary fusion features are formed. Based on the preliminary fusion features, a spatial attention map and a channel attention map are calculated, and a feature attention weight matrix is constructed. The spatial attention map calculates the feature importance of each pixel position through an adaptive convolutional kernel and dynamically adjusts the feature weights according to the feature complexity of different spatial regions, thereby enhancing the expression ability of key ground object features. The channel attention map performs importance ranking on different feature channels through global pooling and weight calculation, and highlights the contribution of different channels in feature fusion through an adaptive weighting mechanism, thereby optimizing the channel information of the feature expression. Through the feature weighting calculation that combines the spatial attention map and the channel attention map, a feature attention weight matrix is constructed. This matrix can adaptively perform weighted adjustment on the preliminary fusion features, enabling key features to be enhanced at different scales and modalities, thereby improving the accuracy of feature expression.The preliminary fusion features are adaptively reweighted according to the feature attention weight matrix, and the semantic segmentation result is used as prior knowledge and input into the local context aggregation unit for feature guidance. The local context aggregation unit captures local context information in the feature space, models the spatial correlation of features, and uses the semantic segmentation result as guidance information to guide the feature fusion direction in different spatial regions, thereby optimizing the feature discrimination between different ground object categories. Through the context feature guidance mechanism, the accuracy of the ground object boundary features is effectively enhanced, and the aggregation ability of the features in the semantically related regions is improved, obtaining context-enhanced features. A cross-scale residual fusion operation is performed on the context-enhanced features. By introducing the residual connection mechanism of different-scale features, cross-scale information fusion of the context-enhanced features at different scales is carried out. Cross-scale residual fusion dynamically adjusts the fusion ratio of different-scale features through the superposition and weighting of multi-scale feature maps, so as to ensure the consistency and integrity of the ground object features at different spatial scales. Through cross-scale feature fusion, the expression ability of the ground object features at different scales is effectively enhanced, and the robustness of the features is maintained in complex ground object scenes, generating a unified geographical feature representation map. The unified geographical feature representation map is input into the geographical feature recognition model for accurate recognition of the ground object category, boundary, and attributes. The geographical feature recognition model adopts an improved ResNet architecture, combined with an atrous convolution sequence and a multi-scale feature perception module, captures the boundary information of the ground object within different receptive fields, and extracts the boundary, category, and attribute features of the ground object through multi-task branches respectively. The boundary recognition branch realizes the precise positioning of the ground object contour through a depthwise separable convolutional network, the category recognition branch performs category mapping of the spatial features through a global attention mechanism, and the attribute regression branch realizes the parameter estimation of the ground object attributes through a fully connected layer. After feature recognition and fusion processing, a set of surveying and mapping geographical features is finally obtained. This set includes structured information such as building contours, road networks, and water system boundaries, and forms a complete geographical feature expression in combination with attribute parameters and spatial relationships.

[0027] In a specific embodiment, the process of performing the step of geographical feature recognition on the unified geographical feature representation map to obtain the set of surveying and mapping geographical features may specifically include the following steps: The unified geographical feature representation map is input into the geographical feature perception residual block of the backbone network for processing to obtain multi-scale receptive field mapping features; Extract the geographical feature contour features from the multi-scale receptive field mapping features to obtain the boundary mask of the geographical features; Analyze the spatial distribution relationship of the geographical features from the multi-scale receptive field mapping features to obtain the category probability map of the geographical features; Extract the attribute information of the geographical features from the multi-scale receptive field mapping features to obtain the attribute parameter values of the geographical features; Perform boundary refinement and class label analysis on the boundary mask of geographical features to obtain the precise boundaries of geographical features and the class information of geographical features; Associate and integrate the precise boundaries of geographical features, the class information of geographical features, the class probability map of geographical features, and the attribute parameter values of geographical features to obtain a set of surveyed geographical features including building outlines, road networks, and water system boundaries.

[0028] Specifically, the unified geographical feature representation map is input into a backbone network with multi-scale feature capture capabilities. This network adopts an improved ResNet architecture and introduces geographical feature-aware residual blocks to enhance the network's multi-scale perception ability of geographical feature characteristics. The geographical feature-aware residual blocks expand the receptive field by combining dilated convolution with convolutional kernels of different scales, and dynamically adjust the weight distribution of features at different scales by introducing an attention mechanism to achieve precise capture of object features at different spatial scales. The dilation rates of the dilated convolution are set to 1, 2, 4, and 8 respectively to ensure that the network captures boundary, texture, and spatial distribution information at different scales within different receptive field ranges, generating multi-scale receptive field mapping features. Geographical feature contour feature extraction is performed on the multi-scale receptive field mapping features. Through the boundary extraction network, depthwise separable convolution is combined with the boundary attention mechanism. By performing layer-by-layer convolution on feature maps of different scales, the boundary information of objects is captured, and the weights of pixels in the boundary region are optimized through the boundary attention mechanism to generate the boundary mask of geographical features. At the same time, during the process of obtaining the boundary mask, the boundary extraction network automatically suppresses the interference of background noise and irrelevant regions, thereby improving the accuracy and robustness of boundary detection. At the same time, the spatial distribution relationship of geographical features is analyzed for the multi-scale receptive field mapping features to generate the class probability map of geographical features. A semantic feature distribution network is introduced, adopting a fully convolutional neural network architecture and combining a spatial attention mechanism. By dynamically weighting the feature responses at different spatial positions, precise modeling of the spatial distribution of object classes is achieved. During the spatial distribution relationship analysis process, the spatial correlation of different object classes is modeled using multi-scale feature information, and the spatial connection between adjacent pixels is captured through the context aggregation module to generate an accurate class probability map, which reflects the probability distribution of each pixel point belonging to different object classes. Attribute information extraction of geographical features is performed on the multi-scale receptive field mapping features to obtain the attribute parameter values of geographical features. This part is achieved through an attribute regression network, which combines a fully connected layer with a feature pooling module. By performing global pooling on the spatial dimension of the feature map and combining the weighted mechanism of feature channels, regression calculations are performed on different types of object attributes to obtain the attribute parameter values of geographical features, such as the height of buildings, the width of roads, the area of water bodies, etc. Boundary refinement and class label analysis are performed on the boundary mask of geographical features to obtain the precise boundary and class information of geographical features. Boundary refinement is achieved by introducing a conditional random field for pixel-level boundary optimization. The conditional random field models the spatial consistency of neighboring pixels and performs refined correction on the classification results in the boundary region, thereby eliminating the fuzzy phenomenon in the boundary region and improving the accuracy of boundary localization. The class label analysis, through a multi-task learning mechanism, jointly inputs the class probability map and the boundary mask into the class recognition network, and uses the attention-guided feature weighting mechanism to achieve precise classification of object classes, generating the precise class information of geographical features.Integrate the precise boundaries of geographical features, the category information of geographical features, the category probability maps of geographical features, and the attribute parameter values of geographical features. Through a feature fusion network, align these information spatially and fuse the features to form a collection of surveyed and mapped geographical features. The feature fusion network constructs a spatial relationship graph of geographical features by introducing a graph neural network, treats different features as nodes, takes the spatial relationships between features as edges, and learns and optimizes the features of nodes and edges through a graph convolutional network to realize the spatial correlation modeling between different features. Under the comprehensive calculation of the feature fusion network, a collection of surveyed and mapped geographical features including building outlines, road networks, and water system boundaries is generated.

[0029] In a specific embodiment, the process of executing step S103 may specifically include the following steps: Calculate the probability distribution data of the geographical feature boundary positions and geographical feature attribute parameters based on the collection of surveyed and mapped geographical features; Calculate the coefficient of variation of the geographical feature boundary positions based on the probability distribution data, and perform adaptive kernel density estimation according to the coefficient of variation to obtain the spatial uncertainty quantification result; Use a Bayesian neural network to calculate the probability estimation of the geographical feature attribute parameters to obtain the parameter uncertainty quantification result, and construct a Bayesian loss function based on the spatial uncertainty quantification result and the parameter uncertainty quantification result; Based on the Bayesian loss function, perform spatial correlation modeling on the collection of surveyed and mapped geographical features to obtain an uncertainty evaluation index with enhanced spatial correlation; Perform environmental factor sensitivity analysis on the uncertainty evaluation index with enhanced spatial correlation to generate a comprehensive uncertainty evaluation result including the category probability distribution of geographical features, position uncertainty, and attribute parameter probability distribution.

[0030] Specifically, taking the collection of surveying and mapping geographic elements as the input, the Monte Carlo dropout technique with multiple forward propagations is used to model the uncertainty of the boundary positions and attribute parameters of geographic elements. By introducing a random dropout layer during the inference process, a different feature representation is obtained for each forward propagation, thus forming a probability distribution data regarding the boundary positions and attribute parameters. After multiple forward propagations, the mean and variance distributions of the boundary positions of each geographic element are obtained, and the regression results of the feature parameters of ground objects (such as building height, road width, water area, etc.) are statistically analyzed to form a probability distribution matrix. Based on the probability distribution data, the coefficient of variation of the boundary positions of geographic elements is calculated. The magnitude of uncertainty is measured by calculating the ratio of the standard deviation to the mean of the boundary positions. For each boundary point of a geographic element, the position changes that occur during multiple forward propagations are calculated, and the uncertainty level of the boundary is measured by the ratio of variance to mean. The larger the boundary coefficient of variation, the higher the uncertainty of the boundary position; conversely, it indicates that the boundary position is stable. According to the calculated coefficient of variation, adaptive kernel density estimation is performed. Kernel density estimation forms a spatial probability density function by fitting the probability distribution of the boundary positions. This function dynamically adjusts the width of the kernel function according to the magnitude of the coefficient of variation to more accurately fit the uncertainty distribution of the boundary positions and generate a spatial uncertainty quantification result. The key to adaptive kernel density estimation lies in adjusting the smoothing parameter of the kernel function according to the variation characteristics of different ground objects, so that the kernel density function can accurately reflect the variation law of different ground object boundaries in different spatial regions and provide a more reliable uncertainty quantification result for subsequent spatial correlation modeling. At the same time, in order to quantify the uncertainty of the attribute parameters of geographic elements, a Bayesian neural network is used to calculate the probability estimation of the attribute parameters of geographic elements. The Bayesian neural network models the uncertainty of parameters by introducing a probability distribution on the network weights. During the training process, the posterior distribution is approximated as a set of parameterized distributions through variational inference, and the model is optimized by maximizing the evidence lower bound. During the inference process, the mean and variance of the attribute parameters are obtained through multiple samplings to form a probability distribution regarding the feature parameters of ground objects, which reflects the fluctuations of the feature parameters of ground objects under different environmental conditions. Based on the spatial uncertainty quantification result and the parameter uncertainty quantification result, a Bayesian loss function is constructed. This loss function consists of a negative log-likelihood term and a KL divergence term. The former ensures the fitting accuracy of the model to the observed data, and the latter prevents the model from overfitting by restricting the distance between the posterior distribution and the prior distribution. The goal of the Bayesian loss function is to improve the prediction reliability and generalization ability of the model by optimizing the uncertainty quantification result. Based on the Bayesian loss function, spatial correlation modeling is performed on the collection of surveying and mapping geographic elements to obtain an uncertainty evaluation index with enhanced spatial correlation.Spatial correlation modeling is accomplished using a variational graph autoencoder. This model constructs geographical features and their spatial relationships into an undirected graph, where nodes represent geographical features and edges represent the spatial relationships between features. The node features are encoded through a graph convolutional network, and the uncertainty distributions of nodes and edges are captured through variational inference to generate an uncertainty assessment index with enhanced spatial correlation. Through variational graph autoencoder modeling, the spatial correlation of geographical features is captured, and potential associations between different features are identified through graph structure learning, thus significantly improving the accuracy of uncertainty assessment. Sensitivity analysis of environmental factors is performed on the uncertainty assessment index with enhanced spatial correlation. The sensitivity analysis simulates multi-dimensional perturbations of the uncertainty assessment results by introducing environmental factor variables (such as lighting conditions, seasonal changes, resolution changes, etc.), and quantifies the influence degree of different environmental factors on the geographical feature recognition results through variational sensitivity analysis (VSA). During the sensitivity analysis process, parameter perturbations are performed on different environmental factor variables, and the changes in the uncertainty assessment results are observed. By calculating the correlation coefficient between the change in uncertainty and the change in environmental factors, an environmental factor sensitivity map is generated, reflecting the sensitivity degree of different feature categories, boundary positions, and attribute parameters to environmental factor changes. Based on the results generated by the sensitivity analysis, a comprehensive uncertainty assessment result including the category probability distribution of geographical features, position uncertainty, and attribute parameter probability distribution is obtained. This assessment result reflects the recognition accuracy of features under different scales and environmental conditions, and quantifies the uncertainty levels of boundary positions and attribute parameters under different conditions.

[0031] In a specific embodiment, the process of executing step S104 may specifically include the following steps: Create a multi-level geographical feature data structure including a point feature layer, a line feature layer, a surface feature layer, and a composite feature layer based on the collection of surveyed and mapped geographical features. The point feature layer includes inflection points and intersection points, the line feature layer includes roads and rivers, the surface feature layer includes buildings and water bodies, and the composite feature layer includes residential areas and industrial areas; Adjust the threshold parameters of the features in the line feature layer based on the comprehensive uncertainty assessment result to obtain refined line feature data; Based on the multi-level geographical feature data structure, use geographical features as nodes and the spatial relationships between features as edges, and assign attribute feature vectors to each node and edge to obtain a geographical feature relationship network; Perform feature relationship learning on the geographical feature relationship network to obtain feature groups and their functional attributes; Compare the collection of surveyed and mapped geographical features with historical data based on the feature groups and their functional attributes and the comprehensive uncertainty assessment result to obtain geographical feature change information; Generate vectorized geographical information based on the geographical feature change information, the refined line feature data, and the feature groups and their functional attributes.

[0032] Specifically, the collection of surveying and mapping geographical elements is processed in layers. Different types of ground feature characteristics are hierarchically divided according to spatial attributes, morphological features, and functional attributes to create a multi-level geographical element data model with a four-layer structure. The point element layer is used to represent key point information in space, including inflection points and intersection points. These point elements are located at road intersections, river confluences, or building boundary corners and are important nodes in the spatial structure of ground features. The line element layer contains linear ground feature characteristics, such as roads and rivers, and forms a complete line element representation by connecting point elements. The road element records the spatial path of the transportation network, while the river element records the spatial flow direction of the water system. The surface element layer is used to represent ground features with area characteristics, including buildings and water bodies, and expresses the boundaries and spatial extents of surface features through closed polygons. The building element provides information on the spatial occupancy of ground features, while the water body element records the distribution and area characteristics of water areas. The composite element layer is used to represent areas with complex functions and containing multiple basic elements, including residential areas and industrial areas. These areas are complexes composed of multiple buildings, roads, and infrastructure, and by spatially aggregating and functionally associating the basic elements, a composite element layer with specific functional attributes is formed. Based on the comprehensive uncertainty assessment results, the threshold parameters of the elements in the line element layer are adjusted to obtain refined line element data. The threshold parameter adjustment mainly targets linear ground features such as roads and rivers. According to the spatial uncertainty quantification results and the coefficient of variation, the retention thresholds of different line elements are dynamically adjusted. Through the adaptive optimization of the Douglas-Peucker algorithm, while ensuring the geometric accuracy of the line elements, redundant points are removed and the expression accuracy of the boundaries is optimized. When this algorithm performs line element simplification, it focuses on retaining areas with a large change rate according to the uncertainty assessment results, while simplifying the point set in areas with small changes or noise interference, thereby effectively reducing the data volume and improving the expression efficiency of line elements. After the threshold parameter adjustment, the refined line element data can maintain the spatial structure characteristics of the line elements and accurately express the ground feature boundaries while significantly reducing the data volume. Based on the multi-level geographical element data structure, geographical elements are regarded as nodes, and the spatial relationships between elements are regarded as edges. Attribute feature vectors are assigned to each node and edge to generate a geographical element relationship network. Point elements, line elements, surface elements, and composite elements are respectively used as nodes in the network. The spatial adjacency relationships, topological connections, and functional dependencies between different elements are used as edges in the network. Through the node feature embedding mechanism, the attribute information, spatial coordinates, and functional characteristics of each element are mapped into high-dimensional feature vectors. The attribute feature vectors assigned to the edges include spatial distance, topological adjacency weights, and spatial direction information, forming a multi-attribute, multi-dimensional geographical element relationship network. Based on the geographical element relationship network, element relationship learning is carried out on the element relationship network to obtain element groups and their functional attributes.Element relationship learning uses a graph neural network to model the spatial topological structure, and through graph convolution and attention mechanism, the features of nodes and edges are iteratively updated in multiple rounds to capture the high-order spatial correlation and functional attribute relationship between elements. During the element relationship learning process, by continuously updating the feature embeddings of nodes and edges, the association relationships between elements gradually emerge, and element groups with similar spatial features and functional attributes are automatically formed. These element groups can reflect the topological connections of spatially adjacent elements and reveal the functional attribute patterns between different feature elements. After completing the element relationship learning, based on the element groups and their functional attributes, and combined with the comprehensive uncertainty assessment results, the collection of surveying and mapping geographic elements is compared with historical data to identify the change information of geographic elements. Change detection uses a deep similarity measurement network to compare the features of current data and historical data, performs multi-scale feature matching in the spatial, semantic, and attribute dimensions, and calculates the change matrix of geographic elements. Through threshold analysis, the change matrix is classified and judged to identify the added, disappeared, and changed feature categories and attribute information. During the change detection process, the functional attribute information of the element groups is used to optimize the change detection results. By combining the functional features of the element groups, false changes caused by noise interference or environmental changes are effectively excluded, thereby improving the accuracy and reliability of change detection. Based on the geographic element change information, the refined line element data, and the element groups and their functional attributes, complete vectorized geographic information is generated. By fusing the change information with the refined line element data, the accurate update of the feature boundary and attribute information is realized. At the same time, the functional attribute information of the element groups is used to reconstruct the composite feature layer to update the spatial structure and functional attributes of complex areas such as residential areas and industrial areas. The generated vectorized geographic information contains the accurate boundary information of points, lines, surfaces, and composite features, and integrates attribute parameters, spatial relationships, and functional attributes.

[0033] The above describes the machine vision-based surveying and mapping geographic information analysis method in the embodiments of the present application. Next, the machine vision-based surveying and mapping geographic information analysis system in the embodiments of the present application will be described. Please refer to Figure 2 , an embodiment of the machine vision-based surveying and mapping geographic information analysis system in the embodiments of the present application includes: A preprocessing module 201, configured to preprocess multi-source remote sensing data to obtain a standardized surveying and mapping data set, and extract ground object edge and texture information from the standardized surveying and mapping data set to obtain multi-scale spatial features and semantic segmentation results; A geographic element recognition module 202, configured to perform multi-source data integration based on the multi-scale spatial features and semantic segmentation results to obtain a unified geographic element representation map, and perform geographic element recognition on the unified geographic element representation map to obtain a collection of surveying and mapping geographic elements; The Bayesian inference module 203 is used to perform Bayesian inference on the collection of surveying and mapping geographic elements to quantify spatial uncertainty and parameter uncertainty, and obtain a comprehensive uncertainty assessment result; The construction module 204 is used to construct vectorized geographic information based on the collection of surveying and mapping geographic elements and the comprehensive uncertainty assessment result.

[0034] Through the collaborative cooperation of the above-mentioned various components, through multi-source remote sensing data adaptive preprocessing and multi-scale feature extraction, the technical problems such as low recognition accuracy, lack of uncertainty assessment, and difficulty in information update in traditional methods in complex scenarios are effectively solved. The dynamic parameter adjustment noise reduction algorithm and the deep learning-based image enhancement technology introduced in this application effectively retain the edge information of geographic elements; the multi-scale feature extraction network with a five-layer pyramid structure combined with a double-branch processing mechanism of direction sensitivity type and region aggregation type significantly enhances the recognition ability of different types of geographic elements; the feature transfer network of the "feature-space-semantics" three-layer transfer framework effectively solves the problem of unbalanced feature expression in multi-modal data fusion; the geographic feature-aware residual block and the three-branch network architecture improve the recognition accuracy of geographic elements; the Bayesian deep learning method combined with variational inference realizes the accurate quantification of the uncertainty of the recognition result; the change detection technology based on the graph neural network for geographic element relationship reasoning and the deep similarity metric network makes the geographic information expression more structured and realizes automatic update.

[0035] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions run on a computer, the computer is made to execute the steps of the method for analyzing surveying and mapping geographic information based on machine vision.

[0036] Those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working processes of the above-described system, system, and unit can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0037] When the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to enable a machine vision-based mapping geographic information analysis device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The foregoing storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs.

[0038] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for analyzing surveying and mapping geographic information based on machine vision, characterized in that, Including: Preprocessing multi-source remote sensing data to obtain a standardized mapping data set, and extracting ground object edge and texture information from the standardized mapping data set to obtain multi-scale spatial features and semantic segmentation results; Integrating multi-source data based on the multi-scale spatial features and the semantic segmentation results to obtain a unified geographic feature representation map, and performing geographic feature recognition on the unified geographic feature representation map to obtain a mapping geographic feature set; Performing Bayesian inference of spatial uncertainty quantification and parameter uncertainty quantification on the mapping geographic feature set to obtain a comprehensive uncertainty evaluation result; Constructing vectorized geographic information based on the mapping geographic feature set and the comprehensive uncertainty evaluation result.

2. The method for analyzing surveying and mapping geographic information based on machine vision according to claim 1, wherein The preprocessing of multi-source remote sensing data to obtain a standardized mapping data set, and extracting ground object edge and texture information from the standardized mapping data set to obtain multi-scale spatial features and semantic segmentation results includes: Automatically identifying sensor parameters of multi-source remote sensing data to obtain parameter calibration results of each data source, where the multi-source remote sensing data includes optical images, radar data, and point cloud data; Calculating a color distribution histogram of the optical image according to the parameter calibration results to obtain image characteristic analysis data; Performing noise reduction processing on the optical image based on the image characteristic analysis data to obtain an edge-preserving denoised image, and simultaneously performing outlier detection on the point cloud data to obtain a point cloud data set after removing outliers; Performing compensation processing on cloud cover and shadow areas of the edge-preserving denoised image to obtain an enhanced processing image; Performing spatial resolution resampling and registration processing on the enhanced processing image, the radar data, and the point cloud data set after removing outliers to obtain a standardized mapping data set; Inputting the standardized mapping data set into a multi-scale feature extractor to extract ground object edge and texture information to obtain multi-scale spatial features and semantic segmentation results.

3. The method for analyzing surveying and mapping geographic information based on machine vision according to claim 2, wherein The inputting the standardized mapping data set into a multi-scale feature extractor to extract ground object edge and texture information to obtain multi-scale spatial features and semantic segmentation results includes: Inputting the standardized mapping data set into a multi-scale feature extractor for convolution operations to obtain multiple initial feature maps; Inputting the multiple initial feature maps into a direction-sensitive convolutional unit and a region-aggregating convolutional unit for processing respectively to obtain linear ground object features and region ground object features; Inputting the linear ground object features and the region ground object features into a local response normalization layer for high-frequency information enhancement processing to obtain an enhanced edge and texture feature map; Performing channel attention weighting on the enhanced edge and texture feature map to obtain multi-scale spatial features, and inputting the enhanced edge and texture feature map into a fully convolutional neural network for semantic segmentation to obtain semantic segmentation results.

4. The method for analyzing surveying and mapping geographic information based on machine vision according to claim 1, wherein The integrating multi-source data based on the multi-scale spatial features and the semantic segmentation results to obtain a unified geographic feature representation map, and performing geographic feature recognition on the unified geographic feature representation map to obtain a mapping geographic feature set includes: Perform an encoding process on the multi-scale spatial features and the semantic segmentation result to obtain an encoded feature representation; Separate the encoded feature representation into a core representation passed through an identity mapping and a variable representation adjusted through an adaptive residual connection, and use the core representation and the variable representation as a dual-path feature stream; Perform cross-modal feature interaction on the dual-path feature stream to obtain a preliminary fusion feature; Calculate a spatial attention map and a channel attention map based on the preliminary fusion feature, and construct a feature attention weight matrix of the spatial attention map and the channel attention map; Perform adaptive re-weighting on the preliminary fusion feature according to the feature attention weight matrix, and use the semantic segmentation result as prior knowledge to input into a local context aggregation unit for feature guidance to obtain a context-enhanced feature; Perform a cross-scale residual fusion operation on the context-enhanced feature to obtain a unified geographical feature representation map; Perform geographical feature recognition on the unified geographical feature representation map to obtain a set of surveying and mapping geographical features; 5. The method for analyzing surveying and mapping geographic information based on machine vision according to claim 4, wherein, The performing geographical feature recognition on the unified geographical feature representation map to obtain a set of surveying and mapping geographical features includes: Input the unified geographical feature representation map into a geographical feature perception residual block of a backbone network for processing to obtain a multi-scale receptive field mapping feature; Extract geographical feature contour features from the multi-scale receptive field mapping feature to obtain a boundary mask of the geographical feature; Analyze the spatial distribution relationship of the geographical feature from the multi-scale receptive field mapping feature to obtain a class probability map of the geographical feature; Extract attribute information of the geographical feature from the multi-scale receptive field mapping feature to obtain an attribute parameter value of the geographical feature; Perform boundary refinement and class label analysis on the boundary mask of the geographical feature to obtain an accurate boundary of the geographical feature and class information of the geographical feature; Associate and integrate the accurate boundary of the geographical feature, the class information of the geographical feature, the class probability map of the geographical feature, and the attribute parameter value of the geographical feature to obtain a set of surveying and mapping geographical features including building outlines, road networks, and water system boundaries; 6. The method for analyzing surveying and mapping geographic information based on machine vision according to claim 1, wherein The performing Bayesian inference of spatial uncertainty quantification and parameter uncertainty quantification on the set of surveying and mapping geographical features to obtain a comprehensive uncertainty assessment result includes: Calculate probability distribution data of the geographical feature boundary position and geographical feature attribute parameters based on the set of surveying and mapping geographical features; Calculate the coefficient of variation of the geographical feature boundary position based on the probability distribution data, and perform adaptive kernel density estimation according to the coefficient of variation to obtain a spatial uncertainty quantification result; Use a Bayesian neural network to calculate the probability estimation of the geographical feature attribute parameters to obtain a parameter uncertainty quantification result, and construct a Bayesian loss function based on the spatial uncertainty quantification result and the parameter uncertainty quantification result; Based on the Bayesian loss function, perform spatial correlation modeling on the set of surveying and mapping geographical features to obtain a spatial correlation-enhanced uncertainty assessment index; Perform environmental factor sensitivity analysis on the uncertainty evaluation index with enhanced spatial correlation, and generate a comprehensive uncertainty evaluation result including the category probability distribution of geographical elements, location uncertainty, and attribute parameter probability distribution.

7. The method for analyzing surveying and mapping geographic information based on machine vision according to claim 1, wherein Construct a four-layer element expression model of points, lines, surfaces, and composites based on the set of surveyed geographical elements and the comprehensive uncertainty evaluation result, and output vectorized geographical information, including: Create a multi-level geographical element data structure including a point element layer, a line element layer, a surface element layer, and a composite element layer based on the set of surveyed geographical elements. The point element layer includes inflection points and intersection points, the line element layer includes roads and rivers, the surface element layer includes buildings and water bodies, and the composite element layer includes residential areas and industrial areas; Adjust the threshold parameters of the elements in the line element layer based on the comprehensive uncertainty evaluation result to obtain refined line element data; Based on the multi-level geographical element data structure, use geographical elements as nodes and the spatial relationships between elements as edges, and assign attribute feature vectors to each node and edge to obtain a geographical element relationship network; Perform element relationship learning on the geographical element relationship network to obtain element groups and their functional attributes; Compare the set of surveyed geographical elements with historical data based on the element groups and their functional attributes and the comprehensive uncertainty evaluation result to obtain geographical element change information; Generate vectorized geographical information based on the geographical element change information, the refined line element data, and the element groups and their functional attributes.

8. A surveying and mapping geographic information analysis system based on machine vision, characterized in that, For implementing the machine vision-based surveyed geographical information analysis method according to any one of claims 1-7, the machine vision-based surveyed geographical information analysis system includes: A preprocessing module for preprocessing multi-source remote sensing data to obtain a standardized surveyed data set, and extracting ground object edge and texture information from the standardized surveyed data set to obtain multi-scale spatial features and semantic segmentation results; A geographical element recognition module for integrating multi-source data based on the multi-scale spatial features and the semantic segmentation results to obtain a unified geographical element representation map, and performing geographical element recognition on the unified geographical element representation map to obtain a set of surveyed geographical elements; A Bayesian inference module for performing Bayesian inference of spatial uncertainty quantification and parameter uncertainty quantification on the set of surveyed geographical elements to obtain a comprehensive uncertainty evaluation result; A construction module for constructing vectorized geographical information based on the set of surveyed geographical elements and the comprehensive uncertainty evaluation result.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is run by a processor, the processor is caused to execute the machine vision-based surveyed geographical information analysis method according to any one of claims 1 to 7.

Images