Surveying image feature extraction system, method, device and medium

Abstract

The application provides a surveying and mapping image feature extraction system, method, device and medium, and belongs to the field of surveying and mapping image feature extraction. The method comprises the following steps: carrying out noise suppression on a terrain surveying and mapping image to be processed to obtain a denoising surveying and mapping image; determining irregular image objects in the denoising surveying and mapping image as terrain recognition units, determining a spectral consistency measure and a complexity descriptor of a surveying and mapping boundary form of each terrain recognition unit, and then forming a feature description set corresponding to each terrain recognition unit; according to the sensitivity of different ground object categories in a preset ground object knowledge base to the spectrum and the surveying and mapping boundary form in the surveying and mapping process, performing feature recombination on the feature description set corresponding to each terrain recognition unit to obtain a plurality of structured feature vectors; and identifying a feature descriptor representing the core content of the terrain surveying and mapping image based on all the structured feature vectors. The technical scheme provided by the application can extract surveying and mapping image features with effective distinguishing degrees in complex scenes.

Description

Technical Field

[0001] This application relates to the field of surveying and mapping image feature extraction technology, and more specifically, to a surveying and mapping image feature extraction system, method, device and medium. Background Technology

[0002] With the rapid development of remote sensing technology and geographic information systems, surveying and mapping image data has become characterized by massive volume, high resolution, and multi-source heterogeneity. Traditional manual interpretation can no longer meet the needs of efficient and accurate geospatial information acquisition. In recent years, with the breakthrough of deep learning technology, models such as convolutional neural networks (CNN) have achieved high-precision processing from pixel level (such as land cover classification) to target level (such as building and road extraction) with their powerful feature learning capabilities. This has further expanded the application boundaries of surveying and mapping in fields such as land survey, smart cities, and disaster monitoring, and has become the core support for the intelligent production of geospatial information.

[0003] In existing surveying and mapping image feature extraction, the underlying information such as image grayscale, texture, and geometric shape is used as a basis. The core is to select key information with recognizability and stability (such as edges, corners, textured regions, target outlines, etc.) from massive pixel data. However, existing surveying and mapping image feature extraction methods usually adopt fixed feature combination strategies or single-scale segmentation models, which fail to fully consider the essential differences in spectral response patterns and morphological structures of different land cover categories. This results in insufficient discriminativeness of the extracted surveying and mapping image features in complex scenes. Therefore, how to extract surveying and mapping image features with effective discriminativeness in complex scenes has become a difficult problem for the industry. Summary of the Invention

[0004] This application provides a surveying and mapping image feature extraction system, method, device, and medium, which can extract surveying and mapping image features with effective discriminative power in complex scenes.

[0005] In a first aspect, this application provides a method for extracting features from surveying and mapping images, comprising the following steps:

[0006] Acquire the topographic mapping image to be processed, and perform noise suppression on the topographic mapping image to obtain a denoised mapping image;

[0007] The denoised mapping image is subjected to multi-dimensional edge-aware segmentation, and irregular image objects in the segmentation process are identified as terrain recognition units. The spectral consistency measure and the complexity descriptor of the mapping boundary morphology of each terrain recognition unit are determined, thereby forming a feature description set corresponding to each terrain recognition unit.

[0008] Based on the sensitivity of different land cover categories in the preset land cover knowledge base to the spectrum and mapping boundary morphology during the surveying process, the feature description set corresponding to each terrain identification unit is reorganized to obtain the structured feature vector of each terrain identification unit.

[0009] Based on the structured feature vectors of each terrain identification unit, feature descriptors representing the core content of the terrain mapping image are identified.

[0010] In some embodiments, noise suppression is performed on the topographic mapping image to obtain a denoised mapping image, specifically including:

[0011] The topographic mapping image is subjected to grayscale normalization processing to obtain a grayscale normalized image;

[0012] The grayscale normalized image is decomposed into a multi-scale image group by using the Gaussian pyramid decomposition algorithm.

[0013] Adaptive median filtering is applied to each scale image in the multi-scale image group to obtain a multi-scale filtered image group.

[0014] The Laplacian pyramid reconstruction algorithm is used to reconstruct the multi-scale filtered image group to obtain denoised mapping images.

[0015] In some embodiments, performing multi-dimensional edge-aware segmentation on the denoised mapping image and identifying irregular image objects as terrain recognition units during the segmentation process specifically includes:

[0016] Gray-level gradient features, texture features, and spectral features are extracted from the denoised mapping image, and then a multi-dimensional feature matrix is ​​constructed.

[0017] Principal component analysis algorithm is used to reduce the dimensionality of the multi-dimensional feature matrix to obtain a dimensionality-reduced feature image.

[0018] Extract the edge mask of the dimensionality-reduced feature image;

[0019] Using the edge mask as a constraint, the denoised mapping image is segmented into regions to obtain multiple irregular image objects in the segmentation process, and each irregular image object is used as a terrain recognition unit.

[0020] In some embodiments, determining the spectral consistency measure and the complexity descriptor of the mapping boundary morphology for each terrain identification unit, and thus forming the feature description set corresponding to each terrain identification unit, specifically includes:

[0021] Select a terrain recognition unit as the selected terrain recognition unit, and calculate the Euclidean distance between the spectral reflectance of each pixel in the selected terrain recognition unit and the mean spectral reflectance of the unit to obtain the spectral distance set.

[0022] The standard deviation of the spectral distance set is calculated, and the obtained standard deviation is used as a spectral consistency measure of the selected terrain identification unit;

[0023] Extract the boundary pixel set of the selected terrain recognition unit, calculate the compactness index of the boundary shape of the selected terrain recognition unit based on the boundary pixel set, and use the compactness index as the complexity descriptor of the boundary shape of the selected terrain recognition unit.

[0024] The spectral consistency measure of the selected terrain identification unit is combined with the complexity descriptor of the mapping boundary morphology to form a feature description set of the selected terrain identification unit.

[0025] Continue to determine the feature description set of the remaining terrain identification units.

[0026] In some embodiments, the feature description set corresponding to each terrain identification unit is reorganized based on the sensitivity of different land cover categories in a preset land cover knowledge base to the spectral and boundary morphology during the surveying process, resulting in a structured feature vector for each terrain identification unit. Specifically, this includes:

[0027] Extract the spectral sensitivity and mapping boundary morphology sensitivity of various land features from the pre-set land feature knowledge base, and then construct a sensitivity weight matrix;

[0028] Read the feature description set of each terrain identification unit and separate the spectral consistency measure and the complexity descriptor of the mapping boundary morphology from the feature description set;

[0029] The spectral consistency measure and complexity descriptor in the feature description set of each terrain identification unit are weighted by the spectral sensitivity weight and the mapping boundary morphology sensitivity weight in the sensitivity weight matrix, respectively, to obtain the weighted spectral feature value and weighted mapping boundary morphology feature value corresponding to each terrain identification unit.

[0030] The structured feature vector of each terrain identification unit is determined based on the weighted spectral feature value and the weighted mapping boundary morphology feature value corresponding to each terrain identification unit.

[0031] In some embodiments, topographical mapping images to be processed are acquired using drones.

[0032] Secondly, this application provides a mapping image feature extraction system, comprising:

[0033] The acquisition module is used to acquire the topographic mapping image to be processed and to perform noise suppression on the topographic mapping image to obtain a denoised mapping image.

[0034] The processing module is used to perform multi-dimensional edge-aware segmentation on the denoised mapping image, and to identify irregular image objects in the segmentation process as terrain recognition units. It determines the spectral consistency measure and the complexity descriptor of the mapping boundary morphology of each terrain recognition unit, thereby forming a feature description set corresponding to each terrain recognition unit.

[0035] The processing module is also used to reorganize the feature description set corresponding to each terrain identification unit according to the sensitivity of different land cover categories in the preset land cover knowledge base to the spectrum and mapping boundary morphology in the mapping process, so as to obtain the structured feature vector of each terrain identification unit.

[0036] The execution module is used to identify feature descriptors that represent the core content of topographic mapping images based on the structured feature vectors of each terrain identification unit.

[0037] Thirdly, this application provides a computer device, the computer device including a memory and a processor, the memory storing code, and the processor being configured to acquire the code and execute the above-described mapping image feature extraction method.

[0038] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for extracting features from surveying and mapping images.

[0039] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects:

[0040] The mapping image feature extraction system, method, device, and medium provided in this application first acquire a topographic mapping image to be processed and perform noise suppression on the topographic mapping image to obtain a denoised mapping image; second, perform multi-dimensional edge-aware segmentation on the denoised mapping image, and identify irregular image objects in the segmentation process as topographic recognition units, determine the spectral consistency measure and the complexity descriptor of the mapping boundary morphology of each topographic recognition unit, and thus form a feature description set corresponding to each topographic recognition unit; then, based on the sensitivity of different land cover categories in a preset land cover knowledge base to the spectrum and mapping boundary morphology in the mapping process, perform feature recombination on the feature description set corresponding to each topographic recognition unit to obtain a structured feature vector for each topographic recognition unit; finally, based on the structured feature vectors of each topographic recognition unit, identify feature descriptors representing the core content of the topographic mapping image.

[0041] Therefore, this application extracts highly discriminative mapping image features in complex scenarios. First, it acquires the topographic mapping image to be processed and performs noise suppression to obtain a denoised mapping image, effectively eliminating useless information such as sensor noise and environmental interference, avoiding errors in subsequent segmentation and feature extraction caused by noise, and providing a high-quality image data foundation for the entire process. Second, it performs multi-dimensional edge-aware segmentation on the denoised mapping image and determines the terrain recognition unit, while forming a feature description set including spectral consistency measures and mapping boundary morphological complexity descriptors. This transforms the continuous image space into discrete ground-level analysis units and provides quantifiable feature basis for subsequent feature processing by quantifying spectral uniformity and morphological irregularity, solving the problems of low efficiency and fragmented information from direct pixel data analysis. Then, based on different ground-feature categories in the preset ground-feature knowledge base, it analyzes the spectral data during the mapping process... By reorganizing the feature description set based on the sensitivity to the morphology of the surveying boundary, a structured feature vector is obtained. This effectively enhances the key features of different land cover categories (such as the spectral features of water bodies and the morphological features of built-up areas), reduces the influence of redundant information, makes the feature vector more closely match the essential attributes of land cover, and provides accurate input for subsequent feature optimization. This avoids insufficient discriminative power in the extraction of surveying image features caused by differences in spectral response patterns and morphological structures among different land cover categories. Finally, based on the structured feature vectors of each terrain identification unit, feature descriptors representing the core content of the terrain surveying image are identified. This allows for further screening of key features with high discriminative power and low redundancy, achieving deep extraction of terrain surveying image features and obtaining core surveying features that accurately represent the composition and spatial distribution of land cover. In summary, the technical solution provided in this application can extract surveying image features with effective discriminative power in complex scenarios. Attached Figure Description

[0042] Figure 1 This is a schematic diagram of an application scenario architecture for a mapping image feature extraction method according to some embodiments of this application;

[0043] Figure 2 This is an exemplary flowchart of a mapping image feature extraction method according to some embodiments of this application;

[0044] Figure 3 This is an exemplary flowchart illustrating the determination of terrain identification units according to some embodiments of this application;

[0045] Figure 4 This is a schematic diagram of the structure of a mapping image feature extraction system according to some embodiments of this application;

[0046] Figure 5 This is a schematic diagram of the structure of a computer device for implementing a method for extracting features from surveying and mapping images, according to some embodiments of this application. Detailed Implementation

[0047] To better understand the technical solution of this application, the technical solution of this application will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0048] refer to Figure 1 This figure is a schematic diagram of an application scenario architecture for the mapping image feature extraction method according to some embodiments of this application. The application scenario architecture includes a data acquisition terminal, a communication network, and a server. The data acquisition terminal and the server are directly or indirectly connected through the communication network. The data acquisition terminal acquires the topographic mapping image to be processed and uploads it to the server. The server performs noise suppression on the topographic mapping image to obtain a denoised mapping image. Multi-dimensional edge-aware segmentation is performed on the denoised mapping image, and irregular image objects in the segmentation process are identified as topographic recognition units. The spectral consistency measure and the complexity descriptor of the mapping boundary morphology of each topographic recognition unit are determined, thereby forming a feature description set corresponding to each topographic recognition unit. Based on the sensitivity of different land cover categories in a preset land cover knowledge base to the spectrum and mapping boundary morphology during the mapping process, the feature description set corresponding to each topographic recognition unit is reorganized to obtain a structured feature vector for each topographic recognition unit. Based on the structured feature vectors of each topographic recognition unit, feature descriptors representing the core content of the topographic mapping image are identified.

[0049] refer to Figure 2 The figure is an exemplary flowchart of a surveying image feature extraction method according to some embodiments of this application. The surveying image feature extraction method mainly includes the following steps:

[0050] In step 101, the topographic mapping image to be processed is acquired, and noise suppression is performed on the topographic mapping image to obtain a denoised mapping image.

[0051] In practice, the topographic mapping images to be processed are acquired by drones. The topographic mapping images refer to the visual images obtained by taking pictures of the target topographic area. The topographic mapping images reflect the topography, landforms, distribution of land features and spatial relationships of the target topographic area.

[0052] In some embodiments, noise suppression is performed on the topographic mapping image to obtain a denoised mapping image, which is specifically achieved through the following steps:

[0053] The topographic mapping image is subjected to grayscale normalization processing to obtain a grayscale normalized image;

[0054] The grayscale normalized image is decomposed into a multi-scale image group by using the Gaussian pyramid decomposition algorithm.

[0055] Adaptive median filtering is applied to each scale image in the multi-scale image group to obtain a multi-scale filtered image group.

[0056] The Laplacian pyramid reconstruction algorithm is used to reconstruct the multi-scale filtered image group to obtain denoised mapping images.

[0057] In specific implementation, firstly, grayscale normalization is adopted. Existing linear transformations are used to map the grayscale values ​​of the topographic mapping image to the standard range of [0, 255], eliminating grayscale deviations caused by different lighting conditions, resulting in a grayscale normalized image. This grayscale normalized image refers to the image after grayscale normalization. Secondly, the Gaussian pyramid decomposition algorithm is used to sequentially perform Gaussian blurring and downsampling operations on the grayscale normalized image. Each operation generates a lower-resolution scale image, ultimately forming a multi-scale image group containing information at different scales. This multi-scale image group refers to a collection of topographic mapping images of different scales obtained from the original topographic mapping image through multiple Gaussian blurring and downsampling processes. When sequentially performing Gaussian blurring and downsampling operations on the grayscale normalized image, downsampling limits can be set according to actual needs; no specific limit is imposed here. Then, targeting... For each scale image in the multi-scale image group, an existing adaptive median filter is applied. This adaptive median filter removes impulse noise while preserving the scale image edge information by adjusting the filter window size (which can be set to 3*3 or 9*9 according to actual needs, without limitation here). The processed scale image is then used as the scale-filtered image. All scale-filtered images are then combined into a multi-scale filtered image group, which refers to the image set formed after each scale image in the multi-scale image group has undergone adaptive median filtering. Finally, the Laplacian pyramid reconstruction algorithm is used to upsample each scale image in the multi-scale filtered image group and subtract it from the corresponding scale Gaussian pyramid image. All difference images are then superimposed on the highest resolution image to complete the image reconstruction. The image obtained from the image reconstruction is used as the denoised mapping image.

[0058] It should be noted that the denoised mapping image in this application refers to the topographic mapping image after noise removal. By identifying the denoised mapping image, useless interference information introduced by sensor noise, environmental interference (such as atmospheric scattering and uneven illumination) in the original topographic mapping image can be effectively eliminated, providing a high-quality image data foundation for the subsequent extraction of topographic image features.

[0059] In step 102, the denoised mapping image is subjected to multi-dimensional edge-aware segmentation, and irregular image objects in the segmentation process are identified as terrain recognition units. The spectral consistency measure and the complexity descriptor of the mapping boundary morphology of each terrain recognition unit are determined, thereby forming a feature description set corresponding to each terrain recognition unit.

[0060] In some embodiments, reference Figure 3 As shown, this figure is an exemplary flowchart of determining terrain recognition units according to some embodiments of this application. In this embodiment, multi-dimensional edge-aware segmentation of the denoised mapping image and determination of irregular image objects in the segmentation process as terrain recognition units can be achieved by the following steps:

[0061] First, in step 1021, grayscale gradient features, texture features and spectral features are extracted from the denoised mapping image, and then a multi-dimensional feature matrix is ​​constructed.

[0062] Secondly, in step 1022, the multidimensional feature matrix is ​​reduced in dimensionality using principal component analysis to obtain a dimensionality-reduced feature image.

[0063] Then, in step 1023, the edge mask of the dimension-reduced feature image is extracted;

[0064] Finally, in step 1024, the denoised mapping image is segmented into regions using the edge mask as a constraint to obtain multiple irregular image objects in the segmentation process, and each irregular image object is used as a terrain recognition unit.

[0065] In specific implementation, firstly, for the denoised mapping image, the Sobel operator is used to calculate the gray-level gradient magnitude at each pixel as the gray-level gradient feature. The contrast corresponding to each pixel is calculated using the gray-level co-occurrence matrix as the texture feature. The band reflectance value of each pixel in the denoised mapping image is directly read as the spectral feature. Then, the gray-level gradient feature, texture feature, and spectral feature are combined according to the pixel position to construct a multi-dimensional feature matrix. The multi-dimensional feature matrix refers to a high-dimensional data matrix that integrates the gray-level gradient, texture, and spectral features of the denoised mapping image and forms a matrix with pixels as the basic unit. Secondly, the principal component analysis algorithm is used to perform mean centering on the multi-dimensional feature matrix to eliminate noise. To analyze the correlation between features, the covariance matrix is ​​calculated to determine the influence of the principal component analysis (PCA). Then, the eigenvalues ​​and eigenvectors of the PCA are solved, and the eigenvectors corresponding to the first K eigenvalues ​​are selected to construct a projection space (where the first K eigenvalues ​​are principal components, and the number K is set according to actual needs, without limitation here). The multi-dimensional feature matrix is ​​then projected onto this projection space to complete dimensionality reduction, resulting in a dimensionality-reduced feature image. This dimensionality-reduced feature image refers to a channel image whose features are represented by principal component information after reducing the feature dimension through PCA. Next, the Canny edge detection operator in image processing is used to extract the edge mask of the dimensionality-reduced feature image. Specifically, the noise of the dimensionality-reduced feature image is first smoothed by Gaussian blurring, and then the dimensionality reduction features are calculated. The gradient magnitude and gradient direction in the feature image are used to locate edge candidate points (i.e., for each pixel in the dimensionality-reduced feature image, based on its gradient direction, it is determined whether the gradient magnitude of the pixel is greater than the gradient magnitude of its two adjacent pixels along the gradient direction. If the gradient magnitude of the pixel is the maximum of these three pixels, the pixel is regarded as an edge candidate point and its pixel value is set to 1; otherwise, the pixel value is set to 0 and it is determined to be a non-edge pixel). A double thresholding method is used to filter strong edges (i.e., edge candidate points in the dimensionality-reduced feature image whose gradient magnitude is greater than a preset high threshold) and weak edges (i.e., edge candidate points whose gradient magnitude is between a preset low threshold and a preset high threshold). Then, discontinuous weak edges are connected with strong edges to form complete edges. An edge mask is obtained, which is an image mask presented in the form of a binary image. In the image mask, edge pixels are marked as foreground (e.g., pixel value 1), and non-edge pixels are marked as background (e.g., pixel value 0). Finally, using the edge mask as a constraint, the watershed segmentation algorithm is used to segment the denoised mapping image into regions. First, the edge mask is used as a segmentation boundary to avoid the segmentation process from crossing the real edge. Then, the gray-level water accumulation simulation is calculated on the denoised mapping image. The pixel gray-level value is regarded as the altitude. Through the "water accumulation" process, the denoised mapping image is divided into multiple unconnected regions, resulting in multiple irregular image objects in the segmentation process. Subsequently, each irregular image object is defined as a terrain recognition unit.

[0066] It should be noted that the terrain recognition unit in this application refers to the basic analysis unit used for terrain recognition. Specifically, it is the basic analysis unit corresponding to irregular image objects. By defining the irregular image objects obtained after multi-dimensional edge perception segmentation of the denoised mapping image as terrain recognition units, the continuous image space can be transformed into discrete analysis units with clear spatial boundaries and attribute associations. This transforms the image pixel set into a concrete carrier that can correspond to actual terrain features, providing a clear analysis object for subsequent targeted extraction of mapping image features.

[0067] In some embodiments, determining the spectral consistency measure and the complexity descriptor of the mapping boundary morphology for each terrain identification unit, and thus forming the feature description set corresponding to each terrain identification unit, is specifically achieved through the following steps:

[0068] Select a terrain recognition unit as the selected terrain recognition unit, and calculate the Euclidean distance between the spectral reflectance of each pixel in the selected terrain recognition unit and the mean spectral reflectance of the unit to obtain the spectral distance set.

[0069] The standard deviation of the spectral distance set is calculated, and the obtained standard deviation is used as a spectral consistency measure of the selected terrain identification unit;

[0070] Extract the boundary pixel set of the selected terrain recognition unit, calculate the compactness index of the boundary shape of the selected terrain recognition unit based on the boundary pixel set, and use the compactness index as the complexity descriptor of the boundary shape of the selected terrain recognition unit.

[0071] The spectral consistency measure of the selected terrain identification unit is combined with the complexity descriptor of the mapping boundary morphology to form a feature description set of the selected terrain identification unit.

[0072] Continue to determine the feature description set of the remaining terrain identification units.

[0073] It should be noted that, in this application, the boundary morphology of the surveying refers to the geometric features and structural state of the spatial outline of topographic features (such as farmland, forest land, water bodies, building areas, etc.) in the context of topographic surveying and remote sensing image analysis. It is the core element representing the spatial morphological attributes of land features.

[0074] In specific implementation, firstly, one terrain recognition unit is selected from all terrain recognition units. The spectral reflectance of each pixel within the selected unit is read using existing image processing tools (such as OpenCV) to extract spectral reflectance data. Then, the arithmetic mean method is used to calculate the average spectral reflectance data to obtain the mean spectral reflectance within the unit. Subsequently, the distance between the spectral reflectance of each pixel and the mean spectral reflectance within the unit is calculated using the Euclidean distance formula, obtaining all spectral distances. These spectral distances are then combined to form a spectral distance set, which refers to the data set consisting of the Euclidean distances between the spectral reflectance of each pixel within the selected terrain recognition unit and the mean spectral reflectance within the unit. Secondly, the standard deviation formula is used to calculate the dispersion of all spectral distances in the spectral distance set. The obtained standard deviation result is used as a spectral consistency measure of the selected terrain recognition unit. The spectral consistency measure is a quantitative index used to characterize the uniformity of the spectral reflectance distribution of pixels within the terrain recognition unit; a smaller spectral consistency measure indicates a more uniform spectrum within the unit. Further, an edge detection algorithm (such as the Canny algorithm) is used to identify the boundary of the selected terrain recognition unit. The boundary contour is defined, and the coordinates of each pixel on the boundary contour are extracted to form a boundary pixel set. Based on the boundary pixel set, the perimeter calculation formula (i.e., calculating the distance between boundary pixels point by point and accumulating the distances to obtain the boundary perimeter) and the minimum bounding rectangle area calculation formula (i.e., finding the maximum and minimum values ​​of the boundary pixels in the horizontal and vertical directions to determine the rectangle range, and then calculating the rectangle area) are used to determine the compactness index of the boundary morphology of the selected terrain recognition unit. This compactness index is determined as the complexity descriptor of the boundary morphology. The complexity descriptor is a parameter used to quantify the degree of irregularity of the boundary shape of a terrain identification unit. The larger the complexity descriptor, the more complex the mapping boundary shape. Then, the spectral consistency measure of the selected terrain identification unit and the complexity descriptor of the mapping boundary shape are integrated in a preset order (such as spectral consistency measure first and complexity descriptor later) to form a feature description set for the selected terrain identification unit. Finally, all the steps from selecting terrain identification units to forming feature description sets are repeated to process all the remaining terrain identification units in sequence, thus completing the determination of the feature description set corresponding to each terrain identification unit.

[0075] It should be noted that, in this application, the feature description set refers to a data combination that includes the spectral consistency measure of the terrain identification unit and the complexity descriptor of the mapping boundary morphology. By integrating the spectral consistency measure of the terrain identification unit and the complexity descriptor of the mapping boundary morphology, the feature description set transforms the spectral distribution uniformity and spatial morphological characteristics of pixels within the unit into clear quantitative indicators, giving the terrain identification unit a unified feature analysis dimension. This effectively solves the problem of difficulty in quantifying spatial information in images. In addition, the determination of the feature description set lays the foundation for outputting feature descriptors that characterize the core content of terrain mapping images and provides data basis that can accurately reflect the essential attributes of ground features for the application of mapping results.

[0076] In step 103, based on the sensitivity of different land cover categories in the preset land cover knowledge base to the spectrum and mapping boundary morphology during the surveying process, the feature description set corresponding to each terrain identification unit is reorganized to obtain the structured feature vector of each terrain identification unit.

[0077] In some embodiments, the feature description set corresponding to each terrain identification unit is reorganized according to the sensitivity of different land cover categories in the preset land cover knowledge base to the spectral and boundary morphology during the surveying process, so as to obtain the structured feature vector of each terrain identification unit. This is specifically achieved through the following steps:

[0078] Extract the spectral sensitivity and mapping boundary morphology sensitivity of various land features from the pre-set land feature knowledge base, and then construct a sensitivity weight matrix;

[0079] Read the feature description set of each terrain identification unit and separate the spectral consistency measure and the complexity descriptor of the mapping boundary morphology from the feature description set;

[0080] The spectral consistency measure and complexity descriptor in the feature description set of each terrain identification unit are weighted by the spectral sensitivity weight and the mapping boundary morphology sensitivity weight in the sensitivity weight matrix, respectively, to obtain the weighted spectral feature value and weighted mapping boundary morphology feature value corresponding to each terrain identification unit.

[0081] The structured feature vector of each terrain identification unit is determined based on the weighted spectral feature value and the weighted mapping boundary morphology feature value corresponding to each terrain identification unit.

[0082] In practice, the first step is to call a pre-defined geographic feature knowledge base (this knowledge base is constructed in advance through field surveys, historical image data statistics, and geographic feature attribute analysis, including but not limited to cultivated land (with corresponding spectral sensitivity and mapping boundary morphology sensitivity of 0.5 and 0.5 respectively), forest land (with corresponding spectral sensitivity and mapping boundary morphology sensitivity of 0.4 and 0.6 respectively), water bodies (with corresponding spectral sensitivity and mapping boundary morphology sensitivity of 0.7 and 0.3 respectively), and built-up areas (with corresponding spectral sensitivity and mapping boundary morphology sensitivity of 0.3 and 0.7 respectively) with corresponding spectral sensitivity and mapping boundary morphology sensitivity. A database of boundary morphology sensitivity parameters is used. Data query tools (such as Python) are employed to extract the spectral sensitivity (a parameter characterizing the dependence of land cover category distinctions on spectral features) and the mapping boundary morphology sensitivity (a parameter characterizing the dependence of land cover category distinctions on mapping boundary morphology features) of each land cover category from the land cover knowledge base. A two-dimensional data matrix is ​​established according to land cover category and sensitivity type, and then a sensitivity weight matrix is ​​constructed. This sensitivity weight matrix is ​​a matrix that stores the corresponding sensitivity parameters for each land cover category, with land cover category as the row and spectral sensitivity and mapping boundary morphology sensitivity as the column. Sensitivity and mapping boundary morphology sensitivity are used as weighting factors. Next, the feature description set of each terrain identification unit is read, and the spectral consistency measure and the complexity descriptor of the mapping boundary morphology are separated from the feature description set. Then, a weighted operation method (i.e., multiplication) is used to multiply the spectral consistency measure of each terrain identification unit with the spectral sensitivity weight of the corresponding land cover category in the sensitivity weight matrix to obtain the weighted spectral feature value of the corresponding terrain identification unit. Simultaneously, the complexity descriptor of the mapping boundary morphology of the terrain identification unit is multiplied by the spectral sensitivity weight of the corresponding land cover category in the sensitivity weight matrix. The sensitivity weights are multiplied to obtain the weighted mapping boundary morphological feature value of the corresponding terrain identification unit. The weighted spectral feature value refers to the spectral feature quantification value of the spectral consistency measure after adjustment by the ground object spectral sensitivity weight, and the weighted mapping boundary morphological feature value refers to the morphological feature quantification value of the mapping boundary morphological sensitivity weight after adjustment by the complexity descriptor. Finally, a feature vector construction method is used to take the weighted spectral feature value of each terrain identification unit as the first element and the weighted mapping boundary morphological feature value as the subsequent elements, and arrange them to form a one-dimensional numerical vector, thereby determining the structured feature vector of the corresponding terrain identification unit.

[0083] It should be noted that the structured feature vector in this application refers to vector data that characterizes the land cover attributes of terrain identification units. Through land cover sensitivity weight optimization, the structured feature vector can make the key features of different land cover categories more prominent. That is, the spectral and morphological feature weights of cultivated land are balanced, which can effectively distinguish similar cultivated land from bare land. The high spectral sensitivity weight of water bodies can avoid shadow confusion caused by infrared light. The high morphological sensitivity weight of built-up areas can highlight the regular boundaries of blocks and reduce misjudgment with dense cultivated land. This provides accurate feature basis for feature extraction of subsequent surveying and mapping image data and reduces the workload and error of manual interpretation.

[0084] In step 104, feature descriptors representing the core content of the topographic mapping image are identified based on the structured feature vectors of each terrain identification unit.

[0085] In some embodiments, the identification of feature descriptors representing the core content of topographic mapping images based on the structured feature vectors of each terrain identification unit is specifically achieved through the following steps:

[0086] Read the structured feature vectors of each terrain recognition unit and construct a feature vector sequence by sorting them according to the spatial location of the terrain mapping image;

[0087] The feature vector sequence is input into a pre-trained feature optimization network, and the feature optimization network extracts the statistical distribution parameters of each component in the structured feature vector;

[0088] The feature optimization network calculates the mutual information value between each component based on the statistical distribution parameters, and then constructs the feature component mutual information matrix.

[0089] Based on the mutual information matrix, terrain mapping features are selected to obtain candidate feature descriptors for each terrain identification unit.

[0090] By associating candidate feature descriptors of all terrain identification units with the spatial location of the topographic mapping image, feature descriptors representing the core content of the topographic mapping image are obtained.

[0091] It should be noted that the feature optimization network in this embodiment adopts a convolutional neural network with an encoder-decoder architecture. The encoder part contains three convolutional layers (using 64, 128, and 256 3×3 convolutional kernels respectively) and a max pooling layer to extract deep representations of structured feature vectors. The decoder part reconstructs the feature distribution through fully connected layers and deconvolutional layers. This feature optimization network is trained using a dataset containing 100,000 sets of labeled topographic mapping images and their corresponding structured feature vectors. During training, a mean squared error loss function and an adaptive moment estimation optimizer are used. The statistical distribution parameters (mean, variance, and skewness) of the feature vectors are calculated through forward propagation, and the correlation between feature components is calculated using a mutual information layer. Finally, the network weights are adjusted through backpropagation, enabling the network to effectively identify feature combinations with high discriminative power. After training, the network can achieve automated optimization and selection of features for topographic recognition units.

[0092] In specific implementation, firstly, the structured feature vectors of all terrain identification units are obtained. Based on the row and column coordinates of each terrain identification unit in the topographic mapping image, the spatial order is determined. The structured feature vectors are arranged from left to right and from top to bottom to construct a feature vector sequence. This feature vector sequence refers to the set of structured feature vectors of all terrain identification units arranged in order of spatial position in the topographic mapping image. Secondly, a pre-trained feature optimization network is invoked (this feature optimization network is pre-trained using a large amount of labeled topographic mapping image structured feature vectors and corresponding land cover category data, and already possesses feature analysis and selection capabilities, which will not be elaborated here). The pre-trained weight parameters and structure configuration file of the feature optimization network are loaded to complete initialization. Then, the constructed feature vector sequence is input into the initialized network input layer. The feature optimization network calculates the mean, variance, and standard deviation of each component of each structured feature vector through its internal statistical analysis module, obtaining the statistical distribution parameters of each component in the structured feature vector. These statistical distribution parameters are quantitative indicators characterizing the data distribution features of each component of the structured feature vector. Then, based on the extracted statistical distribution parameters, the feature optimization network uses a mutual information calculation method (i.e., ... Based on the definition of mutual information in information theory, the mutual information value between any two feature components is calculated by calculating the relative entropy of the joint probability distribution and marginal probability distribution of two random variables (i.e., feature components). All mutual information values ​​are then used to construct a matrix corresponding to the rows and columns of the feature components, resulting in the feature component mutual information matrix. This matrix, with feature components as rows and columns and elements representing the mutual information values ​​of corresponding two feature components, reflects the degree of information correlation between feature components. Next, the feature optimization network selects topographic mapping features based on the feature component mutual information matrix. By setting a mutual information threshold, feature components with mutual information values ​​higher than the threshold are selected to form candidate feature descriptors for each topographic identification unit. These sub-descriptors are the set of feature components that reflect the core landform attributes of the topographic identification unit after feature selection. Finally, based on the spatial location information of the topographic mapping image, all candidate feature descriptors for the topographic identification unit are associated according to their original spatial location (e.g., corresponding according to the row and column order of the topographic mapping image), ensuring that the spatial distribution of the sub-descriptors is consistent with the spatial distribution of the actual topographic features, thus obtaining feature descriptors representing the core content of the topographic mapping image.

[0093] It should be noted that, in this application, the feature descriptor refers to a set of features that reflect the composition and spatial distribution of land features in topographic mapping images. By integrating the feature sub-descriptors of each topographic identification unit after feature selection and retaining their spatial location associations, the feature descriptor can completely and concisely reflect the core attributes and spatial distribution of land features in the image. It can directly provide key feature differences and location information of different land features (arable land, forest land, water bodies, etc.), avoid the interference of redundant data in the original features on the recognition results, greatly improve the efficiency and accuracy of land feature classification, reduce the need for manual interpretation of massive image pixels one by one, and ensure the accurate extraction of mapping image features.

[0094] In another aspect, in some embodiments, this application provides a mapping image feature extraction system, referring to... Figure 4 The figure is a schematic diagram of the structure of a surveying image feature extraction system according to some embodiments of this application. The surveying image feature extraction system includes: an acquisition module 201, a processing module 202, and an execution module 203, which are described below:

[0095] The acquisition module 201 in this application is mainly used to acquire the topographic mapping image to be processed, and to suppress noise in the topographic mapping image to obtain a denoised mapping image.

[0096] Processing module 202, in this application, is mainly used to perform multi-dimensional edge-aware segmentation on the denoised mapping image, and to identify irregular image objects in the segmentation process as terrain recognition units, determine the spectral consistency measure and the complexity descriptor of the mapping boundary morphology of each terrain recognition unit, and then form a feature description set corresponding to each terrain recognition unit.

[0097] The processing module 202 is further configured to perform feature recombination on the feature description set corresponding to each terrain identification unit based on the sensitivity of different land cover categories in the preset land cover knowledge base to the spectrum and mapping boundary morphology during the mapping process, so as to obtain the structured feature vector of each terrain identification unit.

[0098] The execution module 203 in this application is mainly used to identify feature descriptors that represent the core content of topographic mapping images based on the structured feature vectors of each terrain identification unit.

[0099] In addition, this application also provides a computer device, the computer device including a memory and a processor, the memory storing code, and the processor being configured to acquire the code and execute the above-described surveying image feature extraction method.

[0100] In some embodiments, reference Figure 5The figure is a schematic diagram of the structure of a computer device implementing a surveying image feature extraction method according to some embodiments of this application. The surveying image feature extraction method in the above embodiments can be implemented through... Figure 5 The computer device shown is used to implement this, and the computer device includes at least one processor 301, a communication bus 302, a memory 303, and at least one communication interface 304.

[0101] The processor 301 may be a general-purpose central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more devices used to control the execution of the mapping image feature extraction method in this application.

[0102] The communication bus 302 can be used to transmit information between the aforementioned components.

[0103] The memory 303 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disks or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. The memory 303 may exist independently and be connected to the processor 301 via the communication bus 302. The memory 303 may also be integrated with the processor 301.

[0104] The memory 303 stores program code for executing the scheme of this application, and its execution is controlled by the processor 301. The processor 301 executes the program code stored in the memory 303. The program code may include one or more software modules. In the above embodiments, the determination of the mapping image feature extraction method can be implemented by the processor 301 and one or more software modules in the program code in the memory 303.

[0105] Communication interface 304 uses any transceiver-like device for communicating with other devices or communication networks, such as Ethernet, radio access network (RAN), wireless local area networks (WLAN), etc.

[0106] In a specific implementation, as one example, a computer device may include multiple processors, each of which may be a single-core (single-CPU) processor or a multi-core (multi-CPU) processor. Here, a processor may refer to one or more devices, circuits, and / or processing cores used to process data (e.g., computer program instructions).

[0107] The aforementioned computer device can be a general-purpose computer device or a special-purpose computer device. In specific implementations, the computer device can be a desktop computer, a portable computer, a network server, a handheld digital assistant (PDA), a mobile phone, a tablet computer, a wireless terminal device, a communication device, or an embedded device. This application does not limit the type of computer device.

[0108] In addition, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for extracting features from surveying and mapping images.

[0109] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0110] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for extracting features from surveying images, characterized in that, Includes the following steps: Acquire the topographic mapping image to be processed, and perform noise suppression on the topographic mapping image to obtain a denoised mapping image; The denoised mapping image is subjected to multi-dimensional edge-aware segmentation, and irregular image objects in the segmentation process are identified as terrain recognition units. The spectral consistency measure and the complexity descriptor of the mapping boundary morphology of each terrain recognition unit are determined, thereby forming a feature description set corresponding to each terrain recognition unit. Based on the sensitivity of different land cover categories in the preset land cover knowledge base to the spectrum and mapping boundary morphology during the surveying process, the feature description set corresponding to each terrain identification unit is reorganized to obtain the structured feature vector of each terrain identification unit. Based on the structured feature vectors of each terrain identification unit, feature descriptors representing the core content of the terrain mapping image are identified; Specifically, determining the spectral consistency measure and the complexity descriptor of the mapping boundary morphology for each terrain identification unit, and thus forming the feature description set corresponding to each terrain identification unit, includes: Select a terrain recognition unit as the selected terrain recognition unit, and calculate the Euclidean distance between the spectral reflectance of each pixel in the selected terrain recognition unit and the mean spectral reflectance of the unit to obtain the spectral distance set. The standard deviation of the spectral distance set is calculated, and the obtained standard deviation is used as a spectral consistency measure of the selected terrain identification unit; Extract the boundary pixel set of the selected terrain recognition unit, calculate the compactness index of the boundary shape of the selected terrain recognition unit based on the boundary pixel set, and use the compactness index as the complexity descriptor of the boundary shape of the selected terrain recognition unit. The spectral consistency measure of the selected terrain identification unit is combined with the complexity descriptor of the mapping boundary morphology to form a feature description set of the selected terrain identification unit. Continue to determine the feature description set of the remaining terrain identification units; Specifically, based on the sensitivity of different land cover categories in the preset land cover knowledge base to the spectral and boundary morphology during the surveying process, the feature description set corresponding to each terrain identification unit is reorganized to obtain the structured feature vector of each terrain identification unit, which specifically includes: Extract the spectral sensitivity and mapping boundary morphology sensitivity of various land features from the pre-set land feature knowledge base, and then construct a sensitivity weight matrix; Read the feature description set of each terrain identification unit and separate the spectral consistency measure and the complexity descriptor of the mapping boundary morphology from the feature description set; The spectral consistency measure and complexity descriptor in the feature description set of each terrain identification unit are weighted by the spectral sensitivity weight and the mapping boundary morphology sensitivity weight in the sensitivity weight matrix, respectively, to obtain the weighted spectral feature value and weighted mapping boundary morphology feature value corresponding to each terrain identification unit. The structured feature vector of each terrain identification unit is determined based on the weighted spectral feature value and the weighted mapping boundary morphology feature value corresponding to each terrain identification unit.

2. The method as described in claim 1, characterized in that, The process of noise suppression for the topographic mapping image to obtain a denoised mapping image specifically includes: The topographic mapping image is subjected to grayscale normalization processing to obtain a grayscale normalized image; The grayscale normalized image is decomposed into a multi-scale image group by using the Gaussian pyramid decomposition algorithm. Adaptive median filtering is applied to each scale image in the multi-scale image group to obtain a multi-scale filtered image group. The Laplacian pyramid reconstruction algorithm is used to reconstruct the multi-scale filtered image group to obtain denoised mapping images.

3. The method as described in claim 1, characterized in that, Performing multi-dimensional edge-aware segmentation on the denoised mapping image and identifying irregular image objects as terrain recognition units during the segmentation process specifically includes: Gray-level gradient features, texture features, and spectral features are extracted from the denoised mapping image, and then a multi-dimensional feature matrix is ​​constructed. Principal component analysis algorithm is used to reduce the dimensionality of the multi-dimensional feature matrix to obtain a dimensionality-reduced feature image. Extract the edge mask of the dimensionality-reduced feature image; Using the edge mask as a constraint, the denoised mapping image is segmented into regions to obtain multiple irregular image objects in the segmentation process, and each irregular image object is used as a terrain recognition unit.

4. The method as described in claim 1, characterized in that, Topographical mapping images to be processed are acquired using drones.

5. A mapping image feature extraction system, which performs mapping image feature extraction using the method described in any one of claims 1 to 4, characterized in that, The system includes: The acquisition module is used to acquire the topographic mapping image to be processed and to perform noise suppression on the topographic mapping image to obtain a denoised mapping image. The processing module is used to perform multi-dimensional edge-aware segmentation on the denoised mapping image, and to identify irregular image objects in the segmentation process as terrain recognition units. It determines the spectral consistency measure and the complexity descriptor of the mapping boundary morphology of each terrain recognition unit, thereby forming a feature description set corresponding to each terrain recognition unit. The processing module is also used to reorganize the feature description set corresponding to each terrain identification unit according to the sensitivity of different land cover categories in the preset land cover knowledge base to the spectrum and mapping boundary morphology in the mapping process, so as to obtain the structured feature vector of each terrain identification unit. The execution module is used to identify feature descriptors that represent the core content of topographic mapping images based on the structured feature vectors of each terrain identification unit.

6. A computer device, characterized in that, The computer device includes a memory and a processor, the memory storing code, and the processor being configured to acquire the code and execute the mapping image feature extraction method as described in any one of claims 1 to 4.

7. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the mapping image feature extraction method as described in any one of claims 1 to 4.

Images