A method and system for identifying geographic information based on deep learning

Abstract

The present invention relates to the field of geographic information recognition technology, and in particular to a method and system for identifying geographic information based on deep learning. The method comprises the following steps: removing environmental interference from a regional geographic fusion image to generate a regional geographic interference removal image; performing multi-dimensional feature fusion on the regional geographic interference removal image to generate a regional geographic fusion feature map; using an automatic encoder to perform geographic high-dimensional reconstruction on the regional geographic fusion feature map to generate a high-precision regional geographic feature map; capturing regional geographic dynamic changes on the regional geographic fusion image to generate regional geographic dynamic change information data; and marking dynamic feature areas on the regional geographic fusion image according to the regional geographic dynamic change information data to generate regional geographic dynamic feature change areas. The present invention improves the update speed and accuracy of geographic information recognition through multi-source data integration, environmental interference processing, dynamic change monitoring and terrain simulation.

Description

Technical Field

[0001] The present invention relates to the field of geographic information recognition technology, and in particular to a method and system for recognizing geographic information based on deep learning. Background Art

[0002] Initial remote sensing image analysis relied on hand-crafted feature extractors, such as edge detection and texture analysis. However, these methods are limited by changes and noise in complex environments. With the increase in computing power and the availability of large-scale datasets, deep learning has begun to be applied to the processing of geographic information. Key technical breakthroughs include the introduction of convolutional neural networks (CNNs), which can effectively learn feature representations such as land use type and vegetation cover from the pixel level. Subsequent developments include recurrent neural networks (RNNs) and attention mechanisms, which enable models to process sequence data and spatial information at different scales. Another key factor is the emergence of open datasets and pre-trained models, such as ImageNet and OpenStreetMap data, which provide large-scale geographic information required for training deep learning models. Pre-trained models such as BERT and GPT have further promoted the combination of natural language processing and geographic information disciplines, enabling geographic text and context to be better understood and processed. However, traditional technologies are not accurate enough for environmental interference in images. At the same time, when processing geographic information, it is often difficult to capture and accurately identify geographic dynamic changes (such as building changes, land use changes, etc.), which leads to low update speed and accuracy of geographic information. Summary of the invention

[0003] Based on this, it is necessary to provide a method and system for identifying geographic information based on deep learning to solve at least one of the above technical problems.

[0004] To achieve the above purpose, a method for identifying geographic information based on deep learning is provided, the method comprising the following steps:

[0005] Step S1: Acquire a regional multi-source remote sensing image set; perform spatiotemporal alignment on the regional multi-source remote sensing image set to generate a standard regional multi-source spatial alignment image set and a standard regional multi-source spatial alignment image set; perform image fusion on the standard regional multi-source spatial alignment image set and the standard regional multi-source temporal alignment image set to generate a regional geographic fusion image;

[0006] Step S2: removing environmental interference from the regional geographic fusion image to generate a regional geographic interference removal image; performing multi-dimensional feature fusion on the regional geographic interference removal image to generate a regional geographic fusion feature map; performing geographic high-dimensional reconstruction on the regional geographic fusion feature map using an automatic encoder to generate a high-precision regional geographic feature map;

[0007] Step S3: capturing the regional geographic dynamic changes of the regional geographic fusion image to generate regional geographic dynamic change information data; marking the dynamic feature area of ​​the regional geographic fusion image according to the regional geographic dynamic change information data to generate the regional geographic dynamic feature change area; accurately classifying and labeling the features of the high-precision regional geographic feature map according to the regional geographic dynamic feature change area to generate the regional feature classification map;

[0008] Step S4: Perform depth estimation based on the regional land feature classification map and the high-precision regional geographic feature map to generate a regional geographic information depth map; perform terrain simulation on the regional geographic information depth map to generate a regional geographic simulation terrain model; perform intelligent geographic environment analysis and prediction on the regional geographic simulation terrain model to generate regional geographic environment prediction data.

[0009] The present invention obtains multi-source remote sensing image data, performs spatiotemporal alignment of images to ensure accuracy, generates a geographical fusion image of comprehensive information through image fusion, and provides more comprehensive and accurate regional geographical information. By removing environmental interference, extracting geographical features, generating a high-precision regional geographical feature map through feature fusion and high-dimensional reconstruction, more accurate and detailed geographical information is provided. By analyzing regional geographical dynamic change information, marking the area of ​​ground object change, and classifying and labeling ground objects, detailed information on geographical changes and ground object classification is provided, supporting research and decision-making in the fields of environmental monitoring and urban planning. The depth of geographical information is estimated, a terrain model is generated, and geographical environment analysis and prediction are performed to provide important geographical information for geological exploration, urban planning, environmental assessment and other fields. By the spatiotemporal alignment and fusion of multi-source remote sensing images, a more comprehensive and accurate regional geographical fusion image is generated, which is helpful for the extraction and analysis of comprehensive information. By removing environmental interference from the geographical fusion image and then performing multi-dimensional feature fusion, more representative and useful geographical feature information can be extracted, which is helpful for subsequent in-depth analysis and application. Capturing geographical dynamic change information and marking the area of ​​land objects can generate a map of dynamic land object change areas, which is helpful for real-time monitoring and accurate classification of geographical information. Depth estimation and terrain simulation are performed based on the land object classification map and the high-precision geographical feature map to generate a simulated terrain model, which is helpful for more accurate and realistic terrain analysis and simulation. Using the simulated terrain model to perform intelligent geographical environment analysis and prediction and generate environmental prediction data is helpful for the construction and application of decision support systems. Therefore, the present invention improves the update speed and accuracy of geographical information recognition through multi-source data integration, environmental interference processing, dynamic change monitoring and terrain simulation.

[0010] Preferably, step S1 comprises the following steps:

[0011] Step S11: Acquire a regional multi-source remote sensing image set;

[0012] Step S12: performing image preprocessing on the regional multi-source remote sensing image set to generate a standard regional multi-source remote sensing image set, wherein the image preprocessing includes image pixel brightness enhancement, image grayscale transformation and image smoothing;

[0013] Step S13: spatially aligning the standard area multi-source remote sensing image set to generate a standard area multi-source spatially aligned image set;

[0014] Step S14: performing image time series alignment on the standard area multi-source remote sensing image set to generate a standard area multi-source time aligned image set; performing image fusion on the standard area multi-source spatially aligned image set and the standard area multi-source time aligned image set to generate a regional geographic fusion image.

[0015] The present invention can comprehensively utilize data from different sensors by acquiring a multi-source remote sensing image set, enrich the information source, and improve the diversity and integrity of image data. Image pixel brightness enhancement can improve the contrast of the image, making the details clearer and facilitating subsequent processing and analysis. By adjusting the gray value distribution, the image features are enhanced and the visual effect of the image is improved. The noise in the image is removed, the smoothness of the image is improved, and the image is clearer and continuous. By spatially aligning the multi-source remote sensing image set of the standard area, the spatial consistency of images acquired from different sources and at different times can be ensured, and the spatial error caused by the difference in sensor position and terrain changes can be eliminated. By aligning the images in time series, the changes caused by time differences can be eliminated, which is convenient for comparing and analyzing observations at different time points in the same area and obtaining the time-varying characteristics of the area. The images after spatial alignment and time alignment are fused, and the generated regional geographic fusion image contains more information and features, which can more comprehensively reflect the actual situation of the geographical area and improve the practicality and accuracy of the image.

[0016] Preferably, step S2 comprises the following steps:

[0017] Step S21: performing image environment interference recognition on the regional geographic fusion image to generate regional geographic environment interference recognition data; performing image interference removal on the regional geographic fusion image based on the regional geographic environment interference recognition data to generate a regional geographic interference removal image;

[0018] Step S22: performing image edge detection on the regional geographic interference removal image to generate regional geographic image edge detection data; performing ground feature edge segmentation on the regional geographic interference removal image according to the regional geographic image edge detection data to generate a regional geographic edge feature map;

[0019] Step S23: performing regional geographic texture feature analysis on the regional geographic edge feature map to generate regional geographic texture feature data; performing image texture segmentation on the regional geographic edge feature map according to the regional geographic texture feature data to generate a regional geographic texture feature map;

[0020] Step S24: Use a graph convolutional network to perform feature fusion on the regional geographic edge feature map and the regional geographic texture feature map to generate a regional geographic fusion feature map; use an autoencoder to perform geographic high-dimensional reconstruction on the regional geographic fusion feature map to generate a high-precision regional geographic feature map.

[0021] The present invention generates relevant data by identifying environmental interference (such as clouds, shadows, noise, etc.) in the image, which helps to understand and locate the interference area. Based on the identified data, the environmental interference in the image is removed to improve the clarity and quality of the image and ensure the accuracy of subsequent processing. The image after interference removal is edge detected to generate clear edge detection data, which helps to identify the boundaries and contours of the object. According to the edge detection data, the image is segmented for edge features of the object to generate an edge feature map to clearly display the morphological features of the object. Texture feature analysis is performed on the edge feature map to generate detailed texture feature data to capture the surface details and texture of the object. According to the texture feature data, the image is texture segmented to generate a texture feature map to make the surface features of the object clearer and more detailed. The edge feature map and the texture feature map are feature fused using a graph convolutional network, and the information of the two is comprehensively utilized to improve the comprehensiveness and accuracy of feature extraction. The fused feature map is reconstructed in high dimension using an automatic encoder to generate a high-precision regional geographic feature map to provide richer and more detailed geographic information. The high-precision regional geographic feature map finally generated can provide accurate and detailed geographic information and is widely used in geographic information systems (GIS), environmental monitoring, urban planning, natural disaster prediction and other fields.

[0022] Preferably, step S21 includes the following steps:

[0023] Step S211: performing image stratification on the regional geographic fusion image to generate regional geographic fusion image hierarchical data;

[0024] Step S212: performing frequency domain separation on the regional geographic fusion image according to the regional geographic fusion image level data to generate regional geographic level frequency domain information data;

[0025] Step S213: performing regional geographic level spectrum analysis on the regional geographic level frequency domain information data to generate regional geographic image level spectrum analysis data; performing interference target area segmentation on the regional geographic fusion image through the regional geographic image level spectrum analysis data to generate interference target area segmentation data;

[0026] Step S214: performing image structure attribute analysis on the regional geographic fusion image to generate regional geographic image attribute analysis data; performing effective target area segmentation on the regional geographic fusion image according to the regional geographic image attribute analysis data to generate effective target area segmentation data;

[0027] Step S215: Perform binary mask reconstruction based on the interference target area segmentation data and the effective target area segmentation data to generate regional geographic environment interference identification data; perform image difference on the regional geographic fusion image based on the regional geographic environment interference identification data to generate image interference difference and similarity data; remove interference areas from the regional geographic fusion image based on the image interference difference and similarity data to generate a regional geographic interference removed image.

[0028] The present invention can accurately identify the interference area in the image and improve the accuracy of interference identification through steps such as stratification, frequency domain separation and spectral analysis. Differentiation and removal based on the identified interference area data can effectively remove the interference in the image and improve the clarity and quality of the image. Through structural attribute analysis and effective target area segmentation, the effective target area can be more accurately identified and extracted, reducing the impact of interference on target identification. The entire processing process focuses on maintaining the integrity and details of the image, and through multi-step fine processing, it is ensured that valuable information is not lost while removing interference. The generated interference removal image has a wide range of applicability in various application scenarios, such as geographic information systems (GIS), environmental monitoring, urban planning, and agricultural research, providing high-quality image data for subsequent analysis and application.

[0029] Preferably, using an automatic encoder to perform geographic high-dimensional reconstruction on the regional geographic fusion feature map includes:

[0030] Perform multi-level decomposition on the regional geographic fusion image to generate a multi-modal and multi-resolution regional geographic feature map, where the multi-level decomposition includes spatial resolution decomposition, object classification decomposition, and spatiotemporal feature decomposition;

[0031] The encoder in the autoencoder is used to encode the regional geographic feature map in low dimensional data to generate low dimensional latent variable data of regional geographic features; based on the decoder in the autoencoder, the regional geographic features of the low dimensional latent variable data are enhanced to generate a high-precision regional geographic feature map.

[0032] The present invention can effectively extract and enhance regional geographical features and realize high-precision feature reconstruction through the encoding and decoding process of multi-level decomposition and autoencoder. The multi-level decomposition process can capture geographical features at different spatial resolutions and different scales, and provide more comprehensive and detailed geographical information. The classification decomposition of land objects and the decomposition of spatiotemporal features can integrate the feature information of multiple modalities and spatiotemporal dimensions, and improve the diversity and expression ability of the map. The encoding process of the autoencoder can effectively compress high-dimensional data, extract the most important features, reduce redundant information, and improve processing efficiency. Through the feature enhancement and reconstruction of the decoder, the generated regional geographical feature map has higher accuracy and richer details, and can more accurately reflect the actual geographical situation. The encoder in the autoencoder is used to encode the regional geographical feature map with low-dimensional data to generate low-dimensional latent variable data. This step extracts the most important features and reduces redundant information by compressing high-dimensional data. Based on the decoder in the autoencoder, the low-dimensional latent variable data is feature enhanced and reconstructed to generate a high-precision regional geographical feature map. The decoder can restore and enhance the features so that it has higher accuracy and richer details.

[0033] Preferably, step S3 comprises the following steps:

[0034] Step S31: capturing the regional geographic dynamic changes of the regional geographic fusion image to generate regional geographic dynamic change information data;

[0035] Step S32: calculating the image area change rate of the regional geographic fusion image according to the regional geographic dynamic change information data to obtain image area change rate data;

[0036] Step S33: comparing the image area change rate data with a preset image change threshold value, and when the image area change rate data is greater than or equal to the preset image change threshold value, marking the regional geographic fusion image corresponding to the image area change rate data as a dynamic feature area to generate a regional geographic dynamic feature change area;

[0037] Step S34: when the image area change rate data is less than a preset image change threshold, the regional geographic fusion image corresponding to the image area change rate data is marked with a static feature area, generating a regional geographic static feature change area and removing it; the regional geographic dynamic feature change area is subjected to feature semantic segmentation, generating regional geographic dynamic feature semantic segmentation data;

[0038] Step S35: Use the regional geographic dynamic feature semantic segmentation data to accurately classify and annotate the features on the high-precision regional geographic feature map to generate a regional feature classification map.

[0039] The present invention captures the dynamic changes of regional geographic fusion images to generate regional geographic dynamic change information data. In this step, the regions that have changed in the image are identified through time series analysis. According to the dynamic change information data, the change rate of each region in the image is calculated to obtain the image region change rate data. In this step, basic data is provided for subsequent processing by quantifying the degree of change. The change rate data is compared with a preset change threshold. When the change rate is greater than or equal to the threshold, the corresponding region is marked with a dynamic feature region to generate a regional geographic dynamic feature change region. In this step, the region with significant changes can be identified and marked as a dynamic feature change region. When the change rate is less than the threshold, the corresponding region is marked with a static feature region to generate a static feature change region and remove it. By removing the static region, the focus is on the feature region with significant changes. The dynamic feature change region is semantically segmented to generate dynamic feature semantic segmentation data. Semantic segmentation can classify and mark the feature types in detail to provide fine data for further analysis. Using the dynamic feature semantic segmentation data, the high-precision regional geographic feature map is accurately classified and labeled to generate a regional feature classification map. This step integrates the data from the previous steps to generate a detailed and accurate classification map of land objects. Through time series analysis and change rate calculation, it is possible to accurately capture the dynamic changes in regional geographic images and improve the ability to identify changing areas. By comparing the change rate with the threshold, it is possible to effectively distinguish between areas of significant change and static areas, and focus on areas of dynamic change that need special attention. Through semantic segmentation of land objects, it is possible to classify dynamically changing areas in detail and obtain the specific changes of different types of land objects. The final generated regional land object classification map can provide high-precision and detailed geographic information, and provide reliable data support for various geographic analyses and decisions.

[0040] Preferably, step S31 includes the following steps:

[0041] Step S311: confirming the change timestamp of the regional geographic fusion image to obtain the change timestamp of the regional geographic image;

[0042] Step S312: extracting image pixel features from the regional geographic fusion image using the regional geographic image change timestamp to generate image pixel feature data;

[0043] Step S313: Perform motion trajectory optical flow analysis on the image pixel feature data to generate regional geographic pixel motion coordinate data; perform pixel adjacent timestamp motion feature comparison on the image pixel feature data according to the regional geographic pixel motion coordinate data to generate regional geographic image pixel change data;

[0044] Step S314: Divide the regional geographic fusion image into regions based on a preset standard grid size to generate geographic regional grid division data; monitor regional dynamic pixel changes of the geographic regional grid division data through the regional geographic image pixel change data to generate regional geographic dynamic change information data.

[0045] The present invention obtains the regional geographic image change timestamp by confirming the change timestamp of the regional geographic fusion image. This step determines the specific time when the change occurs by marking the time point of the image and provides a reference for the time dimension. The regional geographic fusion image is subjected to image pixel feature extraction using the change timestamp to generate image pixel feature data. This step extracts the pixel features of the image at each time point to form basic data. The image pixel feature data is subjected to motion trajectory optical flow analysis to generate regional geographic pixel motion coordinate data. Optical flow analysis can track the motion trajectory of pixels at different time points and generate pixel motion coordinate data. According to the pixel motion coordinate data, the pixel feature data is subjected to pixel adjacent timestamp motion feature comparison to generate regional geographic image pixel change data. By comparing the motion features of pixels at different time points, the pixel change is identified. The regional geographic fusion image is divided into regions based on a preset standard grid size to generate geographic region grid division data. By dividing the image into standard grids, management and analysis are facilitated. The geographic region grid division data is subjected to regional dynamic pixel change monitoring using the image pixel change data to generate regional geographic dynamic change information data. Each grid is dynamically monitored using the pixel change data to obtain change information within the region. By confirming the change timestamp, the changes of the image at different time points can be accurately captured, providing a time reference for subsequent analysis. By extracting pixel features, detailed pixel feature data is generated, laying the foundation for optical flow analysis and motion feature comparison. Using optical flow analysis technology, the movement trajectory of pixels can be accurately tracked and the changes of pixels at different time points can be identified. By comparing the motion features of pixels at adjacent time points, pixel change data is generated to identify change information in regional geographic images. Dividing images into standard grids facilitates management and analysis, which helps to refine regional dynamic change monitoring.

[0046] Preferably, step S4 comprises the following steps:

[0047] Step S41: performing depth estimation based on the regional feature classification map and the high-precision regional geographic feature map to generate a regional geographic information depth map;

[0048] Step S42: converting the regional geographic information depth map into a three-dimensional point cloud to generate regional geographic three-dimensional point cloud data; and reconstructing the regional geographic three-dimensional point cloud data into a grid to generate a regional geographic three-dimensional grid model;

[0049] Step S43: performing texture mapping on the regional geographic three-dimensional grid model to generate a regional geographic textured three-dimensional model; performing terrain simulation on the regional geographic textured three-dimensional model to generate a regional geographic simulated terrain model;

[0050] Step S44: Perform intelligent geographic environment analysis and prediction on the regional geographic simulation terrain model to generate regional geographic environment prediction data.

[0051] The present invention generates a regional geographic information depth map by performing depth estimation based on a regional object classification map and a high-precision regional geographic feature map. By comprehensively utilizing the object classification and feature information, the depth data of the terrain is accurately estimated. The regional geographic information depth map is converted into a three-dimensional point cloud to generate regional geographic three-dimensional point cloud data. This step converts the two-dimensional depth information into three-dimensional point cloud data to provide three-dimensional spatial information. The regional geographic three-dimensional point cloud data is grid reconstructed to generate a regional geographic three-dimensional grid model. Through grid reconstruction, the point cloud data is converted into a structured three-dimensional grid model. Texture mapping is performed on the regional geographic three-dimensional grid model to generate a regional geographic textured three-dimensional model. This step maps the actual image texture onto the three-dimensional model to make the model more realistic. Terrain simulation is performed on the regional geographic textured three-dimensional model to generate a regional geographic simulated terrain model. Through terrain simulation, the real terrain is simulated to improve the accuracy and practicality of the model. Intelligent geographic environment analysis and prediction are performed on the regional geographic simulation terrain model to generate regional geographic environment prediction data. This step uses an intelligent algorithm to analyze the terrain model to predict environmental changes and trends. Converting depth maps into 3D point cloud data provides rich spatial information and helps to fully understand the 3D structure of the terrain. Through mesh reconstruction of point cloud data, a structured 3D mesh model is generated, making terrain data easier to process and analyze. Through texture mapping, the actual image texture is applied to the 3D model, making the model more realistic and intuitive, which helps with visual display and analysis.

[0052] Preferably, step S44 includes the following steps:

[0053] Step S441: collecting regional geographic information according to the regional geographic simulation terrain model to obtain regional geographic information data;

[0054] Step S442: Divide the regional geographic information data into data sets to generate a model training set and a model test set; perform model training on the model training set using a random forest algorithm to generate a regional geographic environment training model; perform model testing iterations on the regional geographic environment training model using the model test set to generate a regional geographic environment prediction model;

[0055] Step S443: Importing the regional geographic information data into the regional geographic environment prediction model to perform intelligent geographic environment analysis and prediction to generate regional geographic environment prediction data.

[0056] The present invention obtains regional geographic information data by collecting regional geographic information according to a regional geographic simulation terrain model. In this step, relevant geographic information is collected through a simulation terrain model to form a data basis. The regional geographic information data is divided into a data set to generate a model training set and a model test set. Dividing the data set is helpful for model training and testing. The model training set is trained by a random forest algorithm to generate a regional geographic environment training model. The random forest algorithm can process high-dimensional data and complex nonlinear relationships and is suitable for the prediction of the geographic environment. The regional geographic environment training model is tested and iterated by using a model test set to generate a regional geographic environment prediction model. The prediction accuracy and reliability of the model are ensured by repeated testing and optimization. The regional geographic information data is imported into the regional geographic environment prediction model for intelligent geographic environment analysis and prediction to generate regional geographic environment prediction data. In this step, the trained prediction model is used to intelligently predict future geographic environment changes. Information collection based on the simulation terrain model can obtain comprehensive and accurate regional geographic information data, providing a solid data basis for model training and prediction. The data is divided into a training set and a test set to ensure that the model has sufficient data support during the training and testing processes, thereby improving the training effect and test accuracy of the model. The use of random forest algorithm for model training can handle complex data features and improve the model's prediction ability and accuracy. The model is repeatedly tested and optimized through the test set to ensure that the generated prediction model has high reliability and accuracy and can be effectively applied to actual predictions. The regional geographic environment prediction data finally generated can perform intelligent analysis and prediction of future geographic environment changes, providing important decision-making support for environmental protection, urban planning, disaster warning, etc.

[0057] In this specification, a system for identifying geographic information based on deep learning is provided, which is used to perform the above-mentioned method for identifying geographic information based on deep learning. The system for identifying geographic information based on deep learning includes:

[0058] An image fusion module is used to obtain a regional multi-source remote sensing image set; perform spatiotemporal alignment on the regional multi-source remote sensing image set to generate a standard regional multi-source spatial alignment image set and a standard regional multi-source spatial alignment image set; perform image fusion on the standard regional multi-source spatial alignment image set and the standard regional multi-source temporal alignment image set to generate a regional geographic fusion image;

[0059] The geographic high-dimensional reconstruction module is used to remove environmental interference from the regional geographic fusion image to generate a regional geographic interference removal image; perform multi-dimensional feature fusion on the regional geographic interference removal image to generate a regional geographic fusion feature map; use an automatic encoder to perform geographic high-dimensional reconstruction on the regional geographic fusion feature map to generate a high-precision regional geographic feature map;

[0060] The feature annotation module is used to capture the regional geographic dynamic changes of the regional geographic fusion image and generate regional geographic dynamic change information data; mark the regional geographic fusion image with dynamic feature areas according to the regional geographic dynamic change information data and generate the regional geographic dynamic feature change area; accurately classify and annotate the high-precision regional geographic feature map according to the regional geographic dynamic feature change area and generate the regional feature classification map;

[0061] The intelligent prediction module is used to perform depth estimation based on the regional land feature classification map and the high-precision regional geographic feature map to generate a regional geographic information depth map; perform terrain simulation on the regional geographic information depth map to generate a regional geographic simulation terrain model; perform intelligent geographic environment analysis and prediction on the regional geographic simulation terrain model to generate regional geographic environment prediction data.

[0062] The beneficial effect of the present invention is that by fusing a plurality of different remote sensing images, a more comprehensive and higher-quality regional geographic image can be obtained, thereby improving the accuracy and detail of geographic information. By performing spatiotemporal alignment on remote sensing images of different times and sources, they can be unified into a standard coordinate system, which is convenient for subsequent processing and comparative analysis. By removing interference from the geographical fusion image, environmental noise and interference can be eliminated, real geographic information can be extracted, and the interpretability and usability of the image can be enhanced. By multidimensional feature fusion and geographic high-dimensional reconstruction of the autoencoder, richer and more accurate geographic feature information can be extracted, and a high-precision regional geographic feature map can be generated. By capturing dynamic changes and classifying objects in the geographical fusion image, the geographical dynamic change area in the region can be identified and marked, and accurate object classification can be performed to generate a regional object classification map. By depth estimation based on the object classification map and the high-precision geographic feature map, a regional geographic information depth map can be generated, and the depth map can be used for terrain simulation to simulate the real geographical terrain, providing more accurate data for geographical environment analysis and prediction. Therefore, the present invention improves the updating speed and accuracy of geographic information identification through multi-source data integration, environmental interference processing, dynamic change monitoring and terrain simulation. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 A flowchart of the steps of a method for identifying geographic information based on deep learning;

[0064] Figure 2 for Figure 1 Detailed implementation steps of step S2 in the flowchart;

[0065] Figure 3 for Figure 1 Detailed implementation steps of step S3 in FIG.

[0066] Figure 4 for Figure 1 Detailed implementation steps of step S4 in FIG.

[0067] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0068] The following is a clear and complete description of the technical method of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by technicians in this field without creative work are within the scope of protection of the present invention.

[0069] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.

[0070] It should be understood that, although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are used only to distinguish one unit from another unit. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0071] To achieve this, please refer to Figures 1 to 4 , a method for identifying geographic information based on deep learning, the method comprising the following steps:

[0072] Step S1: Acquire a regional multi-source remote sensing image set; perform spatiotemporal alignment on the regional multi-source remote sensing image set to generate a standard regional multi-source spatial alignment image set and a standard regional multi-source spatial alignment image set; perform image fusion on the standard regional multi-source spatial alignment image set and the standard regional multi-source temporal alignment image set to generate a regional geographic fusion image;

[0073] Step S2: removing environmental interference from the regional geographic fusion image to generate a regional geographic interference removal image; performing multi-dimensional feature fusion on the regional geographic interference removal image to generate a regional geographic fusion feature map; performing geographic high-dimensional reconstruction on the regional geographic fusion feature map using an automatic encoder to generate a high-precision regional geographic feature map;

[0074] Step S3: capturing the regional geographic dynamic changes of the regional geographic fusion image to generate regional geographic dynamic change information data; marking the dynamic feature area of ​​the regional geographic fusion image according to the regional geographic dynamic change information data to generate the regional geographic dynamic feature change area; accurately classifying and labeling the features of the high-precision regional geographic feature map according to the regional geographic dynamic feature change area to generate the regional feature classification map;

[0075] Step S4: Perform depth estimation based on the regional land feature classification map and the high-precision regional geographic feature map to generate a regional geographic information depth map; perform terrain simulation on the regional geographic information depth map to generate a regional geographic simulation terrain model; perform intelligent geographic environment analysis and prediction on the regional geographic simulation terrain model to generate regional geographic environment prediction data.

[0076] The present invention obtains multi-source remote sensing image data, performs spatiotemporal alignment of images to ensure accuracy, generates a geographical fusion image of comprehensive information through image fusion, and provides more comprehensive and accurate regional geographical information. By removing environmental interference, extracting geographical features, generating a high-precision regional geographical feature map through feature fusion and high-dimensional reconstruction, more accurate and detailed geographical information is provided. By analyzing regional geographical dynamic change information, marking the area of ​​ground object change, and classifying and labeling ground objects, detailed information on geographical changes and ground object classification is provided, supporting research and decision-making in the fields of environmental monitoring and urban planning. The depth of geographical information is estimated, a terrain model is generated, and geographical environment analysis and prediction are performed to provide important geographical information for geological exploration, urban planning, environmental assessment and other fields. By the spatiotemporal alignment and fusion of multi-source remote sensing images, a more comprehensive and accurate regional geographical fusion image is generated, which is helpful for the extraction and analysis of comprehensive information. By removing environmental interference from the geographical fusion image and then performing multi-dimensional feature fusion, more representative and useful geographical feature information can be extracted, which is helpful for subsequent in-depth analysis and application. Capturing geographical dynamic change information and marking the area of ​​land objects can generate a map of dynamic land object change areas, which is helpful for real-time monitoring and accurate classification of geographical information. Depth estimation and terrain simulation are performed based on the land object classification map and the high-precision geographical feature map to generate a simulated terrain model, which is helpful for more accurate and realistic terrain analysis and simulation. Using the simulated terrain model to perform intelligent geographical environment analysis and prediction and generate environmental prediction data is helpful for the construction and application of decision support systems. Therefore, the present invention improves the update speed and accuracy of geographical information recognition through multi-source data integration, environmental interference processing, dynamic change monitoring and terrain simulation.

[0077] In the embodiment of the present invention, reference Figure 1 The above is a schematic flow chart of the steps of a method for identifying geographic information based on deep learning of the present invention. In this example, the method for identifying geographic information based on deep learning includes the following steps:

[0078] Step S1: Acquire a regional multi-source remote sensing image set; perform spatiotemporal alignment on the regional multi-source remote sensing image set to generate a standard regional multi-source spatial alignment image set and a standard regional multi-source spatial alignment image set; perform image fusion on the standard regional multi-source spatial alignment image set and the standard regional multi-source temporal alignment image set to generate a regional geographic fusion image;

[0079] In an embodiment of the present invention, a multi-source remote sensing image set in a region is obtained from different remote sensing data sources. The remote sensing data source may include satellite images, aerial images, drone images and other sources. The acquired regional multi-source remote sensing image set is subjected to image brightness enhancement to improve the visual quality and contrast of the image. The image is subjected to grayscale transformation, including color space conversion, hue adjustment, etc., to meet the needs of subsequent processing. Image smoothing techniques (such as Gaussian filtering, median filtering, etc.) are applied to reduce noise and unnecessary details in the image to improve image quality and subsequent processing effects. The pre-processed image set is spatially aligned to ensure the consistency of different source images in geographic space. Spatial alignment can use geographic coordinate conversion, feature point matching and other technologies to align images to the same geographic reference system (such as geographic coordinates or projection coordinates). The spatially aligned image set is time-series aligned so that images at different time points or from different data sources correspond to the same moment in time. Time alignment can be adjusted and matched according to the acquisition time or other timestamp information to ensure that the images can be continuous in time or correspond to the same event. The image set that has been spatially and temporally aligned is merged into a comprehensive regional geographic fusion image. Image fusion can adopt a variety of techniques, such as pixel-level fusion, feature fusion, or deep learning-based fusion methods to integrate information from different sources and at different times to generate more comprehensive and accurate geographic fusion images.

[0080] Step S2: removing environmental interference from the regional geographic fusion image to generate a regional geographic interference removal image; performing multi-dimensional feature fusion on the regional geographic interference removal image to generate a regional geographic fusion feature map; performing geographic high-dimensional reconstruction on the regional geographic fusion feature map using an automatic encoder to generate a high-precision regional geographic feature map;

[0081] In the embodiment of the present invention, various environmental interferences in the image, such as clouds, haze, shadows, reflections, etc., are identified by performing environmental interference recognition on the regional geographic fusion image. Methods based on image segmentation, feature extraction or deep learning can be used to automatically identify environmental interference. Based on the environmental interference recognition result, the image is segmented into a clean background and an interference target area. Interference target area segmentation data can be generated by using techniques such as binary mask reconstruction as a basis for further processing. According to the interference target area segmentation data, the interference area is removed from the regional geographic fusion image. The interference area can be restored to the background or original information by using image restoration, interpolation or traditional pixel replacement technology. Multidimensional feature extraction is performed on the regional geographic interference removal image after environmental interference removal. Various features such as space, spectrum, texture, shape, etc. of the image can be extracted to capture rich information in the image. Features obtained from different feature extraction methods are fused to generate a comprehensive regional geographic fusion feature map. Feature fusion can be achieved by using statistical methods, machine learning models or deep learning networks to enhance the expression ability and accuracy of image features. An autoencoder model is designed and trained to learn the high-dimensional representation of the regional geographic fusion feature map. The autoencoder consists of two parts: the encoder and the decoder. The encoder maps the input data to a low-dimensional latent space, and the decoder maps the low-dimensional representation back to the original feature space. The trained autoencoder is used to reconstruct the high-dimensional geography of the regional geographic fusion feature map. The decoder maps the low-dimensional latent variable data back to the high-dimensional feature space to reconstruct a high-precision regional geographic feature map.

[0082] Step S3: capturing the regional geographic dynamic changes of the regional geographic fusion image to generate regional geographic dynamic change information data; marking the dynamic feature area of ​​the regional geographic fusion image according to the regional geographic dynamic change information data to generate the regional geographic dynamic feature change area; accurately classifying and labeling the features of the high-precision regional geographic feature map according to the regional geographic dynamic feature change area to generate the regional feature classification map;

[0083] In an embodiment of the present invention, the dynamic change information of the regional geographic fusion image is extracted and analyzed. The dynamic change of the geographical features and objects in the image can be detected by using image time series analysis, difference image analysis or machine learning model recognition technology. The dynamic change amount and position in the regional geographic fusion image are calculated based on the dynamic change information data. The pixel-level or object-level change detection method can be used to determine the changed geographical area and the changed object type. The geographical area determined according to the dynamic change information is marked as a dynamic object area. The segmentation algorithm or marking technology can be used to segment or mark the changed area from the original image as a basis for further processing. According to the marking results of the dynamic object change area, the high-precision regional geographic feature map is accurately classified and marked. The objects can be classified and marked by using supervised learning, semi-supervised learning or unsupervised learning methods to generate a classification map of objects.

[0084] Step S4: Perform depth estimation based on the regional land feature classification map and the high-precision regional geographic feature map to generate a regional geographic information depth map; perform terrain simulation on the regional geographic information depth map to generate a regional geographic simulation terrain model; perform intelligent geographic environment analysis and prediction on the regional geographic simulation terrain model to generate regional geographic environment prediction data.

[0085] In an embodiment of the present invention, a depth map of regional geographic information is calculated and generated by using a regional object classification map and a high-precision regional geographic feature map. Depth estimation of objects and geographic features can be performed using visual stereo matching, laser scanning data or a deep learning model. A depth map of regional geographic information is generated based on the depth estimation data. The depth map reflects the distribution and height information of objects and geographic features in three-dimensional space, and is the basic data for terrain simulation and environmental analysis. The depth map of regional geographic information is converted into three-dimensional point cloud data. Point cloud processing software or algorithms can be used to convert pixel information in the depth map into a set of three-dimensional coordinate points. Based on the generated three-dimensional point cloud data, a three-dimensional grid model is reconstructed. A continuous three-dimensional grid model is generated by a surface reconstruction algorithm of a point cloud (such as Poisson reconstruction, Marching Cubes algorithm, etc.). The generated regional geographic simulation terrain model is used for intelligent geographic environment analysis and prediction. The geographical environment can be simulated and predicted in combination with simulation algorithms, physical models or machine learning methods, such as hydrological simulation, meteorological forecasting, and natural disaster risk assessment.

[0086] Preferably, step S1 comprises the following steps:

[0087] Step S11: Acquire a regional multi-source remote sensing image set;

[0088] Step S12: performing image preprocessing on the regional multi-source remote sensing image set to generate a standard regional multi-source remote sensing image set, wherein the image preprocessing includes image pixel brightness enhancement, image grayscale transformation and image smoothing;

[0089] Step S13: spatially aligning the standard area multi-source remote sensing image set to generate a standard area multi-source spatially aligned image set;

[0090] Step S14: performing image time series alignment on the standard area multi-source remote sensing image set to generate a standard area multi-source time aligned image set; performing image fusion on the standard area multi-source spatially aligned image set and the standard area multi-source time aligned image set to generate a regional geographic fusion image.

[0091] In an embodiment of the present invention, remote sensing image data from multiple different sources in the region is obtained. Data sources include: satellite remote sensing data: multispectral, hyperspectral or radar image data obtained from different satellites. Aerial data: high-resolution images obtained by aircraft or drones. Other sensor data: data obtained by ground sensors, laser radar (LiDAR), etc., used to obtain information such as terrain or buildings. Use methods such as histogram equalization and contrast enhancement to adjust the brightness and contrast of the image to enhance the visual effect and information content of the image. These techniques can make the image easier to analyze in display and subsequent processing. Apply linear or nonlinear grayscale transformations, such as logarithmic transformations or power law transformations, to adjust the grayscale dynamic range of the image and improve the visual quality and information expression ability of the image. Use a smoothing filter (such as a Gaussian filter) to reduce noise and smooth the image. In the spatial alignment step, ensure that all images are consistent in the geographic coordinate system, and use, for example, geometric correction methods to correct geometric differences between images, such as rotation, translation and scaling, to ensure that they are aligned to the same geographic coordinate system. As well as geographic reference system (CRS) conversion, convert the image to a unified geographic reference system, such as the UTM coordinate system or geographic coordinate system (latitude and longitude), to facilitate spatial analysis and subsequent geographic information system (GIS) applications. For multi-temporal remote sensing images, ensure that they are aligned in time to analyze changes and trends, including correcting the acquisition date and time of the image to ensure that they can accurately reflect time series changes in the analysis. Fuse multi-band images into a multispectral or hyperspectral image to obtain richer information, combining information obtained from different sensors, such as combining optical images and radar data to extract ground features. Regional geographic fusion images are generated by integrating spatially aligned and temporally aligned image sets and applying image fusion technology.

[0092] As an example of the present invention, refer to Figure 2 As shown, in this example, step S2 includes:

[0093] Step S21: performing image environment interference recognition on the regional geographic fusion image to generate regional geographic environment interference recognition data; performing image interference removal on the regional geographic fusion image based on the regional geographic environment interference recognition data to generate a regional geographic interference removal image;

[0094] Step S22: performing image edge detection on the regional geographic interference removal image to generate regional geographic image edge detection data; performing ground feature edge segmentation on the regional geographic interference removal image according to the regional geographic image edge detection data to generate a regional geographic edge feature map;

[0095] Step S23: performing regional geographic texture feature analysis on the regional geographic edge feature map to generate regional geographic texture feature data; performing image texture segmentation on the regional geographic edge feature map according to the regional geographic texture feature data to generate a regional geographic texture feature map;

[0096] Step S24: Use a graph convolutional network to perform feature fusion on the regional geographic edge feature map and the regional geographic texture feature map to generate a regional geographic fusion feature map; use an autoencoder to perform geographic high-dimensional reconstruction on the regional geographic fusion feature map to generate a high-precision regional geographic feature map.

[0097] In the embodiment of the present invention, various environmental interferences in the image, such as clouds, shadows, haze, etc., are identified and analyzed by using computer vision and image processing technology. This can be achieved by using a pixel-based threshold method, a feature-based classification method (such as a machine learning classifier), a convolutional neural network (CNN)-based method, etc. Based on the interference identification data generated in step S21, appropriate image restoration or interference removal techniques, such as image patching, content filling, deblurring, etc., are used to remove environmental interference in the image. Traditional image processing algorithms and deep learning technologies can be combined to effectively remove interference and retain useful information in the image. Use classic edge detection algorithms such as Sobel, Canny, etc., or edge detection networks based on deep learning (such as CNN-based edge detection models) to extract edge information of objects from the image after interference removal. According to the edge detection data generated in step S22, the image is further processed to segment the edge features of the objects. Image segmentation algorithms, such as region-growing-based and edge-based segmentation methods, and semantic segmentation networks in deep learning (such as U-Net, Mask R-CNN, etc.) can be combined to accurately extract edge features of objects. The regional geographic edge feature map generated in step S22 is subjected to texture feature analysis to reveal the texture features of the surface of the object. Statistical analysis methods (such as grayscale co-occurrence matrix), frequency domain analysis methods (such as wavelet transform), or texture feature extraction networks based on deep learning (such as TextureNet, etc.) can be used to extract and analyze the texture features of the object. According to the geographic texture feature data, a suitable image segmentation algorithm is adopted, such as a texture-based segmentation method, a texture segmentation network based on deep learning (such as TextureNet), or a combination of multi-scale segmentation strategies to achieve texture segmentation of geographic images. The regional geographic edge feature map and the texture feature map are feature fused using a graph convolutional network (GCN) or other suitable deep learning models. GCN can effectively utilize the information and geographical relationships of each position in the image to achieve global fusion and enhancement of features. Use an autoencoder (AE) or other deep learning models to reconstruct the fused feature map in high dimensions. The autoencoder can learn a high-level abstract representation of the feature map, thereby generating a high-precision regional geographic feature map.

[0098] Preferably, step S21 includes the following steps:

[0099] Step S211: performing image stratification on the regional geographic fusion image to generate regional geographic fusion image hierarchical data;

[0100] Step S212: performing frequency domain separation on the regional geographic fusion image according to the regional geographic fusion image level data to generate regional geographic level frequency domain information data;

[0101] Step S213: performing regional geographic level spectrum analysis on the regional geographic level frequency domain information data to generate regional geographic image level spectrum analysis data; performing interference target area segmentation on the regional geographic fusion image through the regional geographic image level spectrum analysis data to generate interference target area segmentation data;

[0102] Step S214: performing image structure attribute analysis on the regional geographic fusion image to generate regional geographic image attribute analysis data; performing effective target area segmentation on the regional geographic fusion image according to the regional geographic image attribute analysis data to generate effective target area segmentation data;

[0103] Step S215: Perform binary mask reconstruction based on the interference target area segmentation data and the effective target area segmentation data to generate regional geographic environment interference identification data; perform image difference on the regional geographic fusion image based on the regional geographic environment interference identification data to generate image interference difference and similarity data; remove interference areas from the regional geographic fusion image based on the image interference difference and similarity data to generate a regional geographic interference removed image.

[0104] In an embodiment of the present invention, the regional geographic fusion image is segmented or divided according to different levels, including stratification according to pixel level or regional level. A segmentation method based on pixel value, texture feature or semantic information can be used to segment the image into different levels for subsequent frequency domain analysis and attribute analysis. According to the regional geographic fusion image level data generated in step S211, frequency domain analysis is performed on the image of each level. The image can be decomposed into different frequency domain components by methods such as fast Fourier transform (FFT) or wavelet transform to obtain information of the image in the frequency domain. Spectral analysis is performed on the frequency domain information data of the regional geographic level after frequency domain separation, including statistical analysis or model fitting of the frequency domain information to extract spectral features of the geographic image level, such as spectral components reflecting different frequencies or bands. Based on the spectral analysis data, the interference target area in the image, such as clouds, shadows, haze, etc., is identified and segmented. Threshold segmentation, regional growth, or target detection algorithm based on deep learning can be used to accurately segment the interference target area. Structural attribute analysis is performed on the regional geographic fusion image, such as extraction and analysis of features such as texture, edge, shape, etc. The structural properties of the image can be extracted using methods such as filters, local binary patterns (LBP), and gradient information. According to the image attribute analysis data, the valid target area in the image, such as objects and buildings, can be identified and segmented. The valid target area can be accurately segmented and identified by combining threshold processing, region growing, or semantic segmentation models in deep learning. Based on the segmentation data of the interference target area and the segmentation data of the effective target area, a binary mask is generated to identify the interference area and the effective area. Using the binary mask, the regional geographic fusion image is differentiated to obtain the interference similarities and differences in the image. According to the image difference data, the interference area in the image, such as occluded objects or noise points, is removed in a targeted manner. The image restoration algorithm, content filling technology, or an image restoration model based on deep learning can be used to generate a clean regional geographic interference removal image.

[0105] Preferably, using an automatic encoder to perform geographic high-dimensional reconstruction on the regional geographic fusion feature map includes:

[0106] Perform multi-level decomposition on the regional geographic fusion image to generate a multi-modal and multi-resolution regional geographic feature map, where the multi-level decomposition includes spatial resolution decomposition, object classification decomposition, and spatiotemporal feature decomposition;

[0107] The encoder in the autoencoder is used to encode the regional geographic feature map in low dimensional data to generate low dimensional latent variable data of regional geographic features; based on the decoder in the autoencoder, the regional geographic features of the low dimensional latent variable data are enhanced to generate a high-precision regional geographic feature map.

[0108] In an embodiment of the present invention, the regional geographic fusion image is decomposed according to different spatial resolutions, including the extraction of high-resolution details and low-resolution overall features. The regional geographic fusion image is classified into different types of objects, such as buildings, vegetation, water bodies, etc. If there are data of multiple time phases, the regional geographic fusion image can be decomposed into time and space features to analyze and extract the changes in geographical features in different time periods. The encoder part of the autoencoder is used to convert the multi-level decomposed regional geographic feature map into a low-dimensional latent variable representation. The encoder is specifically a deep neural network structure, which usually includes multiple convolutional layers or fully connected layers to map the input features to a coding space of smaller dimensions. The decoder part of the autoencoder is used to remap the encoded low-dimensional latent variable data to the original high-dimensional feature space. The decoder usually includes a structure symmetrical to the encoder, and data reconstruction from low dimension to high dimension is realized through a deconvolution layer or a reverse convolution layer. In the decoder stage, the quality and accuracy of the reconstructed regional geographic feature map can be enhanced by adjusting the training strategy of the autoencoder or introducing additional geographic information constraints. This can include using geographic learning constraints, incorporating geographic information system (GIS) data, or feature classification information to optimize the decoding process.

[0109] As an example of the present invention, refer to Figure 3 As shown, in this example, step S3 includes:

[0110] Step S31: capturing the regional geographic dynamic changes of the regional geographic fusion image to generate regional geographic dynamic change information data;

[0111] Step S32: calculating the image area change rate of the regional geographic fusion image according to the regional geographic dynamic change information data to obtain image area change rate data;

[0112] Step S33: comparing the image area change rate data with a preset image change threshold value, and when the image area change rate data is greater than or equal to the preset image change threshold value, marking the regional geographic fusion image corresponding to the image area change rate data as a dynamic feature area to generate a regional geographic dynamic feature change area;

[0113] Step S34: when the image area change rate data is less than a preset image change threshold, the regional geographic fusion image corresponding to the image area change rate data is marked with a static feature area, generating a regional geographic static feature change area and removing it; the regional geographic dynamic feature change area is subjected to feature semantic segmentation, generating regional geographic dynamic feature semantic segmentation data;

[0114] Step S35: Use the regional geographic dynamic feature semantic segmentation data to accurately classify and annotate the features on the high-precision regional geographic feature map to generate a regional feature classification map.

[0115] In an embodiment of the present invention, by analyzing the regional geographic fusion image, the dynamic change information in the geography is identified and captured, such as the addition or removal of buildings, the growth or reduction of vegetation, etc. The dynamic changes can be detected and extracted using image difference, optical flow method, or time series analysis method based on deep learning. Based on the dynamic change information data generated in step S31, the change rate of each regional geographic fusion image is calculated. The change rate can be used to quantify the degree of change of different regions in the geographic image, usually by comparing the differences in features such as brightness, color, texture, etc. of pixels or regions. The image region change rate data is compared with a preset image change threshold. If the change rate of a certain region is greater than or equal to the preset threshold, the region is marked as a dynamic feature change region. For regions with a change rate less than the preset threshold, they are marked as static feature change regions and are removed from subsequent processing. The dynamic feature change region marked in step S33 is further subjected to feature semantic segmentation. Using a deep learning semantic segmentation model (such as FCN, U-Net, etc.), the pixels in the dynamic feature area are accurately classified according to the geographical features and the changing context information, such as identifying and marking newly added buildings, removed vegetation, etc. The dynamic feature semantic segmentation data generated in step S34 is used to accurately classify and annotate the regional geographical feature map. The geographical features in the dynamic feature area are marked according to their categories (such as buildings, roads, water bodies, etc.) to generate a high-precision feature classification map.

[0116] Preferably, step S31 includes the following steps:

[0117] Step S311: confirming the change timestamp of the regional geographic fusion image to obtain the change timestamp of the regional geographic image;

[0118] Step S312: extracting image pixel features from the regional geographic fusion image using the regional geographic image change timestamp to generate image pixel feature data;

[0119] Step S313: Perform motion trajectory optical flow analysis on the image pixel feature data to generate regional geographic pixel motion coordinate data; perform pixel adjacent timestamp motion feature comparison on the image pixel feature data according to the regional geographic pixel motion coordinate data to generate regional geographic image pixel change data;

[0120] Step S314: Divide the regional geographic fusion image into regions based on a preset standard grid size to generate geographic regional grid division data; monitor regional dynamic pixel changes of the geographic regional grid division data through the regional geographic image pixel change data to generate regional geographic dynamic change information data.

[0121] In the embodiment of the present invention, by analyzing the time series data of the regional geographic fusion image, the timestamp information of each image, that is, the acquisition or update time corresponding to each image, is confirmed. According to the timestamp information obtained in step S311, the regional geographic fusion image is subjected to pixel-level feature extraction. Feature data at the level of color information, texture features, brightness, etc. of the pixel can be extracted. The regional geographic fusion image sequence is subjected to pixel-level motion trajectory analysis using optical flow analysis technology. Optical flow analysis can help detect the movement direction and speed change of pixels in the image. The image pixel feature data of adjacent timestamps are compared to analyze the change of pixels between different timestamps. By comparing the feature data of pixels, pixel-level changes, such as moving and changing building outlines, can be detected. According to a preset standard grid size or geographic location division method, the regional geographic fusion image is divided into a plurality of geographic area grids. The size and division of the grid can be adjusted according to the needs of the specific application, and are usually used for regional change monitoring and analysis. The pixel change data generated in step S313 is used to monitor the dynamic changes of pixels in the geographic area grid. When it is detected that the pixel change in a certain grid area exceeds a preset threshold, the grid is marked as a dynamic change area, and regional geographic dynamic change information data is generated.

[0122] As an example of the present invention, refer to Figure 4 As shown, in this example, step S4 includes:

[0123] Step S41: performing depth estimation based on the regional feature classification map and the high-precision regional geographic feature map to generate a regional geographic information depth map;

[0124] Step S42: converting the regional geographic information depth map into a three-dimensional point cloud to generate regional geographic three-dimensional point cloud data; and reconstructing the regional geographic three-dimensional point cloud data into a grid to generate a regional geographic three-dimensional grid model;

[0125] Step S43: performing texture mapping on the regional geographic three-dimensional grid model to generate a regional geographic textured three-dimensional model; performing terrain simulation on the regional geographic textured three-dimensional model to generate a regional geographic simulated terrain model;

[0126] Step S44: Perform intelligent geographic environment analysis and prediction on the regional geographic simulation terrain model to generate regional geographic environment prediction data.

[0127] In an embodiment of the present invention, by utilizing a regional feature classification map and a high-precision regional geographic feature map, combined with deep learning or computer vision technology, depth estimation is performed on objects or terrain in a geographical environment. Depth estimation can be based on monocular or multi-eye vision, and the depth information of each pixel or region can be inferred through a neural network model or a traditional visual geometry method. The geographic information depth map generated in step S41 is converted into three-dimensional point cloud data. The point cloud data in the geographical environment can be reconstructed into a three-dimensional point cloud model using depth information and pixel coordinates. The three-dimensional point cloud data is gridded to generate a regional geographic three-dimensional grid model with continuity and structure. The gridding process can be implemented using point cloud segmentation, surface reconstruction algorithms or finite element methods. The texture information of the regional geography is mapped to the three-dimensional grid model generated in step S42 to generate a regional geographic textured three-dimensional model. Texture mapping can use geographic information images or texture data collected in real time to fit the surface of the three-dimensional model to enhance the realism and visual effect of the model. The textured three-dimensional model is subjected to terrain simulation to simulate and predict changes in terrain features and terrain influences in the geographical environment. Terrain simulation can be combined with terrain analysis, hydrological models or geospatial data to predict geological disasters, hydrological processes, etc. The simulated terrain model and textured three-dimensional model generated in step S43 are used to perform intelligent geographic environment analysis and prediction. Machine learning models, physical simulation models or hybrid models can be used to analyze factors such as change trends, resource utilization, and disaster risks in the geographic environment.

[0128] Preferably, step S44 includes the following steps:

[0129] Step S441: collecting regional geographic information according to the regional geographic simulation terrain model to obtain regional geographic information data;

[0130] Step S442: Divide the regional geographic information data into data sets to generate a model training set and a model test set; perform model training on the model training set using a random forest algorithm to generate a regional geographic environment training model; perform model testing iterations on the regional geographic environment training model using the model test set to generate a regional geographic environment prediction model;

[0131] Step S443: Importing the regional geographic information data into the regional geographic environment prediction model to perform intelligent geographic environment analysis and prediction to generate regional geographic environment prediction data.

[0132] In the embodiment of the present invention, various geographic information data in the region are collected and integrated through the simulated terrain model generated in step S43. The geographic information data may include various types of geographic data such as terrain elevation, vegetation cover, land use, hydrological conditions, and climate data. The regional geographic information data collected in step S441 is divided into a model training set and a model test set. The process of dividing the data set may adopt a random division or a stratified sampling method based on a geographic region. The model training set is trained using a random forest algorithm to generate a regional geographic environment training model. Random forest is an integrated learning method suitable for processing complex geographic environment data and can effectively perform classification and regression tasks. The trained regional geographic environment training model is tested and iterated using a model test set. By evaluating the performance of the model on the test set, parameter tuning and model optimization are performed to ensure that the model has high prediction accuracy and generalization ability. The regional geographic environment prediction model trained in step S442 is deployed in practical applications. The new geographic information data collected in step S441 is imported into the prediction model for intelligent geographic environment analysis and prediction. According to the output of the prediction model, prediction data of the regional geographic environment is generated. Prediction data can include various types of geographic environment analysis results, such as future terrain change trends, natural disaster risk assessments, and urban development planning recommendations.

[0133] In this specification, a system for identifying geographic information based on deep learning is provided, which is used to perform the above-mentioned method for identifying geographic information based on deep learning. The system for identifying geographic information based on deep learning includes:

[0134] An image fusion module is used to obtain a regional multi-source remote sensing image set; perform spatiotemporal alignment on the regional multi-source remote sensing image set to generate a standard regional multi-source spatial alignment image set and a standard regional multi-source spatial alignment image set; perform image fusion on the standard regional multi-source spatial alignment image set and the standard regional multi-source temporal alignment image set to generate a regional geographic fusion image;

[0135] The geographic high-dimensional reconstruction module is used to remove environmental interference from the regional geographic fusion image to generate a regional geographic interference removal image; perform multi-dimensional feature fusion on the regional geographic interference removal image to generate a regional geographic fusion feature map; use an automatic encoder to perform geographic high-dimensional reconstruction on the regional geographic fusion feature map to generate a high-precision regional geographic feature map;

[0136] The feature annotation module is used to capture the regional geographic dynamic changes of the regional geographic fusion image and generate regional geographic dynamic change information data; mark the regional geographic fusion image with dynamic feature areas according to the regional geographic dynamic change information data and generate the regional geographic dynamic feature change area; accurately classify and annotate the high-precision regional geographic feature map according to the regional geographic dynamic feature change area and generate the regional feature classification map;

[0137] The intelligent prediction module is used to perform depth estimation based on the regional land feature classification map and the high-precision regional geographic feature map to generate a regional geographic information depth map; perform terrain simulation on the regional geographic information depth map to generate a regional geographic simulation terrain model; perform intelligent geographic environment analysis and prediction on the regional geographic simulation terrain model to generate regional geographic environment prediction data.

[0138] The beneficial effect of the present invention is that by fusing a plurality of different remote sensing images, a more comprehensive and higher-quality regional geographic image can be obtained, thereby improving the accuracy and detail of geographic information. By performing spatiotemporal alignment on remote sensing images of different times and sources, they can be unified into a standard coordinate system, which is convenient for subsequent processing and comparative analysis. By removing interference from the geographical fusion image, environmental noise and interference can be eliminated, real geographic information can be extracted, and the interpretability and usability of the image can be enhanced. By multidimensional feature fusion and geographic high-dimensional reconstruction of the autoencoder, richer and more accurate geographic feature information can be extracted, and a high-precision regional geographic feature map can be generated. By capturing dynamic changes and classifying objects in the geographical fusion image, the geographical dynamic change area in the region can be identified and marked, and accurate object classification can be performed to generate a regional object classification map. By depth estimation based on the object classification map and the high-precision geographic feature map, a regional geographic information depth map can be generated, and the depth map can be used for terrain simulation to simulate the real geographical terrain, providing more accurate data for geographical environment analysis and prediction. Therefore, the present invention improves the updating speed and accuracy of geographic information identification through multi-source data integration, environmental interference processing, dynamic change monitoring and terrain simulation.

[0139] Therefore, the embodiments should be regarded as illustrative and non-restrictive from all points, and the scope of the present invention is limited by the appended claims rather than the above description, and it is therefore intended that all changes falling within the meaning and range of equivalent elements of the application documents are included in the present invention.

[0140] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A method for identifying geographic information based on deep learning, characterized in that: The following steps are involved: Step S1: Acquire a regional multi-source remote sensing image set; Performing spatiotemporal alignment on the regional multi-source remote sensing image set to generate a standard regional multi-source spatially aligned image set and a standard regional multi-source spatially aligned image set; Perform image fusion on a standard regional multi-source spatially aligned image set and a standard regional multi-source temporally aligned image set to generate a regional geographic fused image; Step S2: removing environmental interference from the regional geographic fusion image to generate a regional geographic interference removed image; Perform multi-dimensional feature fusion on the regional geographic interference removal image to generate a regional geographic fusion feature map; use an automatic encoder to perform geographic high-dimensional reconstruction on the regional geographic fusion feature map to generate a high-precision regional geographic feature map; Step S3: capturing the regional geographic dynamic changes of the regional geographic fusion image to generate regional geographic dynamic change information data; marking the dynamic feature area of ​​the regional geographic fusion image according to the regional geographic dynamic change information data to generate the regional geographic dynamic feature change area; accurately classifying and labeling the features of the high-precision regional geographic feature map according to the regional geographic dynamic feature change area to generate the regional feature classification map; Step S4: performing depth estimation based on the regional feature classification map and the high-precision regional geographic feature map to generate a regional geographic information depth map; Perform terrain simulation on the regional geographic information depth map to generate a regional geographic simulation terrain model; perform intelligent geographic environment analysis and prediction on the regional geographic simulation terrain model to generate regional geographic environment prediction data; Step S1 includes the following steps: Step S11: Acquire a regional multi-source remote sensing image set; Step S12: performing image preprocessing on the regional multi-source remote sensing image set to generate a standard regional multi-source remote sensing image set, wherein the image preprocessing includes image pixel brightness enhancement, image grayscale transformation and image smoothing; Step S13: spatially aligning the standard area multi-source remote sensing image set to generate a standard area multi-source spatially aligned image set; Step S14: performing image time series alignment on the standard area multi-source remote sensing image set to generate a standard area multi-source time aligned image set; performing image fusion on the standard area multi-source spatially aligned image set and the standard area multi-source time aligned image set to generate a regional geographic fusion image; Step S2 includes the following steps: Step S21: performing image environment interference recognition on the regional geographic fusion image to generate regional geographic environment interference recognition data; performing image interference removal on the regional geographic fusion image based on the regional geographic environment interference recognition data to generate a regional geographic interference removal image; Step S22: performing image edge detection on the regional geographic interference removal image to generate regional geographic image edge detection data; performing ground feature edge segmentation on the regional geographic interference removal image according to the regional geographic image edge detection data to generate a regional geographic edge feature map; Step S23: performing regional geographic texture feature analysis on the regional geographic edge feature map to generate regional geographic texture feature data; performing image texture segmentation on the regional geographic edge feature map according to the regional geographic texture feature data to generate a regional geographic texture feature map; Step S24: using a graph convolutional network to perform feature fusion on the regional geographic edge feature map and the regional geographic texture feature map to generate a regional geographic fusion feature map; using an automatic encoder to perform geographic high-dimensional reconstruction on the regional geographic fusion feature map to generate a high-precision regional geographic feature map; Step S21 includes the following steps: Step S211: performing image stratification on the regional geographic fusion image to generate regional geographic fusion image hierarchical data; Step S212: performing frequency domain separation on the regional geographic fusion image according to the regional geographic fusion image level data to generate regional geographic level frequency domain information data; Step S213: performing regional geographic level spectrum analysis on the regional geographic level frequency domain information data to generate regional geographic image level spectrum analysis data; performing interference target area segmentation on the regional geographic fusion image through the regional geographic image level spectrum analysis data to generate interference target area segmentation data; Step S214: performing image structure attribute analysis on the regional geographic fusion image to generate regional geographic image attribute analysis data; performing effective target area segmentation on the regional geographic fusion image according to the regional geographic image attribute analysis data to generate effective target area segmentation data; Step S215: performing binary mask reconstruction based on the interference target area segmentation data and the effective target area segmentation data to generate regional geographic environment interference identification data; performing image difference on the regional geographic fusion image based on the regional geographic environment interference identification data to generate image interference difference and similarity data; performing interference area removal on the regional geographic fusion image based on the image interference difference and similarity data to generate a regional geographic interference removal image; Among them, using the automatic encoder to reconstruct the geographical high-dimensionality of the regional geographical fusion feature map includes: Perform multi-level decomposition on the regional geographic fusion image to generate a multi-modal and multi-resolution regional geographic feature map, where the multi-level decomposition includes spatial resolution decomposition, object classification decomposition, and spatiotemporal feature decomposition; The encoder in the autoencoder is used to encode the regional geographic feature map in low dimensional data to generate low dimensional latent variable data of regional geographic features; the decoder in the autoencoder is used to enhance the regional geographic features of the low dimensional latent variable data to generate a high-precision regional geographic feature map; Step S3 includes the following steps: Step S31: capturing the regional geographic dynamic changes of the regional geographic fusion image to generate regional geographic dynamic change information data; Step S32: calculating the image area change rate of the regional geographic fusion image according to the regional geographic dynamic change information data to obtain image area change rate data; Step S33: comparing the image area change rate data with a preset image change threshold value, and when the image area change rate data is greater than or equal to the preset image change threshold value, marking the regional geographic fusion image corresponding to the image area change rate data as a dynamic feature area to generate a regional geographic dynamic feature change area; Step S34: when the image area change rate data is less than a preset image change threshold, the regional geographic fusion image corresponding to the image area change rate data is marked with a static feature area, generating a regional geographic static feature change area and removing it; the regional geographic dynamic feature change area is subjected to feature semantic segmentation, generating regional geographic dynamic feature semantic segmentation data; Step S35: using the regional geographic dynamic feature semantic segmentation data to accurately classify and annotate the features on the high-precision regional geographic feature map to generate a regional feature classification map; Step S31 includes the following steps: Step S311: confirming the change timestamp of the regional geographic fusion image to obtain the change timestamp of the regional geographic image; Step S312: extracting image pixel features from the regional geographic fusion image using the regional geographic image change timestamp to generate image pixel feature data; Step S313: Perform motion trajectory optical flow analysis on the image pixel feature data to generate regional geographic pixel motion coordinate data; perform pixel adjacent timestamp motion feature comparison on the image pixel feature data according to the regional geographic pixel motion coordinate data to generate regional geographic image pixel change data; Step S314: dividing the regional geographic fusion image into regions based on a preset standard grid size to generate geographic regional grid division data; monitoring the regional dynamic pixel changes of the geographic regional grid division data through the regional geographic image pixel change data to generate regional geographic dynamic change information data; Step S4 includes the following steps: Step S41: performing depth estimation based on the regional feature classification map and the high-precision regional geographic feature map to generate a regional geographic information depth map; Step S42: converting the regional geographic information depth map into a three-dimensional point cloud to generate regional geographic three-dimensional point cloud data; and reconstructing the regional geographic three-dimensional point cloud data into a grid to generate a regional geographic three-dimensional grid model; Step S43: performing texture mapping on the regional geographic three-dimensional grid model to generate a regional geographic textured three-dimensional model; performing terrain simulation on the regional geographic textured three-dimensional model to generate a regional geographic simulated terrain model; Step S44: performing intelligent geographical environment analysis and prediction on the regional geographical simulation terrain model to generate regional geographical environment prediction data; Step S44 includes the following steps: Step S441: collecting regional geographic information according to the regional geographic simulation terrain model to obtain regional geographic information data; Step S442: Divide the regional geographic information data into data sets to generate a model training set and a model test set; perform model training on the model training set using a random forest algorithm to generate a regional geographic environment training model; perform model testing iterations on the regional geographic environment training model using the model test set to generate a regional geographic environment prediction model; Step S443: Importing the regional geographic information data into the regional geographic environment prediction model to perform intelligent geographic environment analysis and prediction to generate regional geographic environment prediction data.

2. A geographic information recognition system based on deep learning, characterized in that: For executing the method for identifying geographic information based on deep learning as claimed in claim 1, the geographic information identification system based on deep learning comprises: An image fusion module is used to obtain a regional multi-source remote sensing image set; perform spatiotemporal alignment on the regional multi-source remote sensing image set to generate a standard regional multi-source spatial alignment image set and a standard regional multi-source spatial alignment image set; perform image fusion on the standard regional multi-source spatial alignment image set and the standard regional multi-source temporal alignment image set to generate a regional geographic fusion image; The geographic high-dimensional reconstruction module is used to remove environmental interference from the regional geographic fusion image to generate a regional geographic interference removal image; perform multi-dimensional feature fusion on the regional geographic interference removal image to generate a regional geographic fusion feature map; use an automatic encoder to perform geographic high-dimensional reconstruction on the regional geographic fusion feature map to generate a high-precision regional geographic feature map; The feature annotation module is used to capture the regional geographic dynamic changes of the regional geographic fusion image and generate regional geographic dynamic change information data; mark the regional geographic fusion image with dynamic feature areas according to the regional geographic dynamic change information data and generate the regional geographic dynamic feature change area; accurately classify and annotate the high-precision regional geographic feature map according to the regional geographic dynamic feature change area and generate the regional feature classification map; The intelligent prediction module is used to perform depth estimation based on the regional land feature classification map and the high-precision regional geographic feature map to generate a regional geographic information depth map; perform terrain simulation on the regional geographic information depth map to generate a regional geographic simulation terrain model; perform intelligent geographic environment analysis and prediction on the regional geographic simulation terrain model to generate regional geographic environment prediction data.

Images