Automatic Recognition Method and System for Geographical Elements of Remote Sensing Images Based on Large Model Analysis

Abstract

The present invention discloses a method and system for automatically identifying geographical elements in remote sensing images based on large model analysis, which relates to the technical field of image processing. The method includes: obtaining remote sensing image data based on multiple remote sensing data sources according to remote sensing data processing rules; obtaining a geographical element identification instruction; performing multi-level image segmentation constraint registration to establish an image segmentation constraint tree; segmenting the remote sensing image data according to the SAM image segmentation large model to obtain a multi-level remote sensing image segmentation result; optimizing the segmentation loss to obtain a multi-level remote sensing image segmentation optimization result; performing vectorized element identification to obtain a multi-level geographical element identification result. It solves the technical problems of insufficient accuracy, low efficiency in identifying multi-level geographical elements in remote sensing images and inaccurate image segmentation in the prior art, realizes the efficient and accurate identification of geographical elements, and achieves the technical effect of improving the efficiency and accuracy of geographical element identification.

Description

Technical Field

[0001] This application relates to the technical field of image processing, and specifically relates to a method and system for automatically identifying geographic elements of remote sensing images based on large model analysis. Background Art

[0002] The identification of geographic elements in remote sensing images is an important application of remote sensing technology in Geographic Information System (GIS), and is widely used in fields such as urban planning, agricultural monitoring, environmental monitoring, and resource exploration. As an important means of obtaining ground information, remote sensing images collect image data of the Earth's surface through sensors such as satellites and aerial platforms, and can provide rich information for the identification of geographic elements. However, due to the fact that remote sensing images often have complex geographical backgrounds, different land cover types, and different spatial resolutions, traditional methods for identifying geographic elements in remote sensing images mostly rely on pixel-based image classification methods. Generally, by extracting features from the image data, such as spectral features, texture features, shape features, etc., to achieve the classification and identification of geographic elements, it is difficult to process complex image features, especially in the case of poor image quality or the presence of noise, and it is unable to make full use of the spatial context information of remote sensing images, resulting in low classification accuracy in complex scenarios. Moreover, for complex multi-level and multi-scale geographic element information, such as urban buildings, farmland, water bodies, forests, etc., traditional geographic element identification methods are unable to process such diverse and complex remote sensing images. For example, it is difficult to automatically identify and accurately locate geographic elements in areas such as nature reserves and post-disaster assessment areas.

[0003] Therefore, in the current related technologies, there are technical problems such as insufficient accuracy, low efficiency in identifying multi-level geographic elements in remote sensing images, and inaccurate image segmentation. Summary of the Invention

[0004] This application provides a method and system for automatically identifying geographic elements of remote sensing images based on large model analysis, solves the technical problems of insufficient accuracy, low efficiency in identifying multi-level geographic elements in remote sensing images, and inaccurate image segmentation existing in the prior art, realizes the efficient and accurate identification of geographic elements, and achieves the technical effect of improving the efficiency and accuracy of geographic element identification.

[0005] The present application provides a method for automatically identifying geographic elements in remote sensing images based on large model analysis. The method includes: obtaining remote sensing image data based on multiple remote sensing data sources according to remote sensing data processing rules; obtaining a geographic element identification instruction, where the geographic element identification instruction includes multi-level element identification targets, and the multi-level element identification targets include full-scale elements, prompt word elements, and point of interest elements; performing multi-level image segmentation constraint registration on the remote sensing image data according to the multi-level element identification targets to establish an image segmentation constraint tree; based on the image segmentation constraint tree, performing segmentation processing on the remote sensing image data according to the SAM image segmentation large model to obtain a multi-level remote sensing image segmentation result; optimizing the segmentation loss of the multi-level remote sensing image segmentation result according to an image segmentation loss detection model to obtain a multi-level remote sensing image segmentation optimization result; and performing vectorized element identification according to the multi-level remote sensing image segmentation optimization result to obtain a multi-level geographic element identification result.

[0006] In a possible implementation manner, the method for automatically identifying geographic elements in remote sensing images based on large model analysis further performs the following processing: parsing image segmentation constraint parameters for the remote sensing image data according to the full-scale elements to obtain a first image segmentation constraint condition; parsing image segmentation constraint parameters for the remote sensing image data according to the prompt word elements to obtain a second image segmentation constraint condition; parsing image segmentation constraint parameters for the remote sensing image data according to the point of interest elements to obtain a third image segmentation constraint condition; using the full-scale elements as the first node for geographic element identification, using the prompt word elements as the second node for geographic element identification, and using the point of interest elements as the third node for geographic element identification; constructing an element identification node tree according to the first node for geographic element identification, the second node for geographic element identification, and the third node for geographic element identification; and performing image segmentation constraint configuration on the element identification node tree according to the first image segmentation constraint condition, the second image segmentation constraint condition, and the third image segmentation constraint condition to generate the image segmentation constraint tree.

[0007] In a possible implementation manner, the method for automatically identifying geographic elements in remote sensing images based on large model analysis further performs the following processing: performing twin data retrieval on the remote sensing image data to obtain Q remote sensing image twin samples, where Q is a positive integer greater than 1;

[0008] Collect the historical parameters of the full - scale element image segmentation constraints corresponding to the Q remote sensing image twin samples to obtain Q full - scale element segmentation constraint groups. Among them, each full - scale element segmentation constraint group includes M full - scale element image segmentation constraint samples corresponding to each remote sensing image twin sample, and M is a positive integer greater than 1; Based on the global confidence evaluation channel, conduct global confidence evaluation on the Q full - scale element segmentation constraint groups to obtain Q global confidences of the constraint groups; Based on the Q global confidences of the constraint groups, optimize and select the Q full - scale element segmentation constraint groups according to the predetermined global confidence to establish a full - scale element segmentation constraint space; Calculate the central value according to the full - scale element segmentation constraint space to generate the first constraint condition for image segmentation, where the first constraint condition for image segmentation includes M central values of image segmentation constraints.

[0009] In a possible implementation, the method for automatically identifying geographical elements of remote sensing images based on large - model analysis further performs the following processing: The global confidence evaluation channel includes a global confidence evaluation formula, and the global confidence evaluation formula is:

[0010] ;

[0011] Among them, represents the q - th global confidence of the constraint group corresponding to the q - th full - scale element segmentation constraint group, q is a positive integer, 1 ≤ q ≤ Q, represents the support degree of the q - th constraint group corresponding to the q - th full - scale element segmentation constraint group, represents the support degree of the m - th full - scale element image segmentation constraint sample in the q - th full - scale element segmentation constraint group.

[0012] In a possible implementation, the method for automatically identifying geographical elements of remote sensing images based on large - model analysis further performs the following processing: The prompt - word elements include multiple geographical - element prompt words; Expand the large - model according to the prompt words to perform synonym expansion on the multiple geographical - element prompt words to obtain multiple prompt - word clusters; Input the multiple prompt - word clusters and the remote sensing image data into an image target detection model to obtain multiple sets of target - element annotation frames corresponding to the multiple prompt - word clusters; Perform coupling optimization according to the multiple sets of target - element annotation frames to generate the second constraint condition for image segmentation.

[0013] In a possible implementation, the method for automatically identifying geographical elements of remote sensing images based on large - model analysis further performs the following processing: The point - of - interest elements include multiple point - of - interest geographical locations; Construct a coordinate system according to the remote sensing image data to obtain a remote - sensing image coordinate system; Perform coordinate transformation on the multiple point - of - interest geographical locations based on the remote - sensing image coordinate system to obtain element point - of - interest spatial data; Perform CSV conversion according to the element point - of - interest spatial data to generate the third constraint condition for image segmentation.

[0014] In a possible implementation, the automatic recognition method for remote sensing image geographical elements based on large model analysis further performs the following processing: detecting the segmentation loss of the multi-level remote sensing image segmentation result according to the image segmentation loss detection model to obtain each image segmentation loss detection result, where the image segmentation loss detection model includes multi-dimensional indexes for image segmentation loss detection, and the multi-dimensional indexes for image segmentation loss detection include segmentation accuracy, over-segmentation rate, and under-segmentation rate; determining whether each image segmentation loss detection result satisfies the multi-dimensional constraints for segmentation loss detection; if each image segmentation loss detection result does not satisfy the multi-dimensional constraints for segmentation loss detection, generating a segmentation loss optimization instruction; and based on the segmentation loss optimization instruction, correcting the segmentation loss of the multi-level remote sensing image segmentation result according to the multi-dimensional constraints for segmentation loss detection to obtain an optimized result of the multi-level remote sensing image segmentation.

[0015] In a possible implementation, the automatic recognition method for remote sensing image geographical elements based on large model analysis further performs the following processing: the remote sensing data processing rules include remote sensing data quality constraints, a predetermined remote sensing data format, and remote sensing data processing factors, and the remote sensing data processing factors include radiometric correction, geometric correction, and image enhancement; loading first remote sensing image data according to the multiple remote sensing data sources; filtering the first remote sensing image data according to the remote sensing data quality constraints to obtain second remote sensing image data; performing format conversion on the second remote sensing image data according to the predetermined remote sensing data format to obtain third remote sensing image data; and preprocessing the third remote sensing image data according to the remote sensing data processing factors to generate the remote sensing image data.

[0016] The present application also provides an automatic recognition system for geographical elements of remote sensing images based on large model analysis, including: a remote sensing image data acquisition module for acquiring remote sensing image data based on remote sensing data processing rules and multiple remote sensing data sources; a geographical element recognition instruction acquisition module for acquiring geographical element recognition instructions, where the geographical element recognition instructions include multi-level element recognition targets, and the multi-level element recognition targets include full-scale elements, prompt word elements, and point of interest elements; an image segmentation constraint tree establishment module for performing multi-level image segmentation constraint registration on the remote sensing image data according to the multi-level element recognition targets and establishing an image segmentation constraint tree; a remote sensing image segmentation result acquisition module for segmenting the remote sensing image data based on the image segmentation constraint tree according to the SAM image segmentation large model to obtain multi-level remote sensing image segmentation results; a remote sensing image segmentation optimization result acquisition module for optimizing the segmentation loss of the multi-level remote sensing image segmentation results according to an image segmentation loss detection model to obtain multi-level remote sensing image segmentation optimization results; and a geographical element recognition result acquisition module for performing vectorized element recognition according to the multi-level remote sensing image segmentation optimization results to obtain multi-level geographical element recognition results.

[0017] It is intended to obtain remote sensing image data based on remote sensing data processing rules and multiple remote sensing data sources through the automatic recognition method and system for geographical elements of remote sensing images based on large model analysis proposed in this application; obtain geographical element recognition instructions; perform multi-level image segmentation constraint registration to establish an image segmentation constraint tree; segment the remote sensing image data according to the SAM image segmentation large model to obtain multi-level remote sensing image segmentation results; optimize the segmentation loss to obtain multi-level remote sensing image segmentation optimization results; and perform vectorized element recognition to obtain multi-level geographical element recognition results. This solves the technical problems of insufficient accuracy, low efficiency in multi-level geographical element recognition of remote sensing images, and inaccurate image segmentation in the prior art, realizes efficient and accurate recognition of geographical elements, and achieves the technical effect of improving the efficiency and accuracy of geographical element recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the accompanying drawings of the embodiments of the present disclosure will be briefly introduced below. Flowcharts are used in this application to illustrate the operations performed by the system according to the embodiments of the application. It should be understood that the operations before or below do not necessarily need to be executed precisely in sequence. On the contrary, according to the need, they can be executed in reverse order or simultaneously. At the same time, other operations can also be added to these processes, or one or several operations can be removed from these processes.

[0019] Figure 1 It is a schematic flowchart of the automatic recognition method for geographical elements of remote sensing images based on large model analysis provided by the embodiments of the present application.

[0020] Figure 2 This is a schematic structural diagram of a remote sensing image geographic feature automatic recognition system based on large model analysis provided by an embodiment of the present application.

[0021] Explanation of reference numerals: Remote sensing image data acquisition module 10, geographic feature recognition instruction acquisition module 20, image segmentation constraint tree establishment module 30, remote sensing image segmentation result acquisition module 40, remote sensing image segmentation optimization result acquisition module 50, geographic feature recognition result acquisition module 60. Specific implementation manners

[0022] The above description is only an overview of the technical solution of the present application. In order to be able to understand the technical means of the present application more clearly, it can be implemented according to the content of the specification. And in order to make the above and other purposes, features and advantages of the present application more obvious and understandable, the specific implementation manners of the present application are specifically exemplified below.

[0023] In order to make the purpose, technical solution and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limitations of the present application. All other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present application.

[0024] In the following description, "some embodiments" are involved, which describe a subset of all possible embodiments. However, it can be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments, and can be combined with each other without conflict. The terms "first" and "second" involved are only used to distinguish similar objects and do not represent a specific order for the objects. The terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or server including a series of steps or units does not necessarily limit to those clearly listed steps or units, but may include other steps or modules not clearly listed or inherent to these processes, methods, products or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which the present application belongs. The terms used herein are only for the purpose of describing the embodiments of the present application.

[0025] An embodiment of the present application provides a method for automatically recognizing geographic features of remote sensing images based on large model analysis, as Figure 1 shown, the method includes:

[0026] Step S100, based on remote sensing data processing rules, obtain remote sensing image data according to multiple remote sensing data sources.

[0027] Step S100 further includes step S110. The remote sensing data processing rules include remote sensing data quality constraints, a predetermined remote sensing data format, and remote sensing data processing factors. The remote sensing data processing factors include radiometric correction, geometric correction, and image enhancement. Step S120: Load the first remote sensing image data according to the multiple remote sensing data sources. Step S130: Filter the first remote sensing image data according to the remote sensing data quality constraints to obtain the second remote sensing image data. Step S140: Perform format conversion on the second remote sensing image data according to the predetermined remote sensing data format to obtain the third remote sensing image data. Step S150: Preprocess the third remote sensing image data according to the remote sensing data processing factors to generate the remote sensing image data.

[0028] Preferably, based on the remote sensing data processing rules, multiple remote sensing data sources are processed to obtain remote sensing image data. Among them, the remote sensing data processing rules refer to the rules for acquiring, processing, and analyzing remote sensing data, including remote sensing data quality constraints, a predetermined remote sensing data format, and remote sensing data processing factors. The remote sensing data quality constraints define how to evaluate and ensure the quality of remote sensing data. The quality of remote sensing data may be affected by various factors, including the performance of sensors, atmospheric conditions, cloud cover, severe blurring, etc. By setting some quality standards (such as signal-to-noise ratio, resolution requirements, etc.) to screen and evaluate the effectiveness of data. The predetermined remote sensing data format refers to the pre-determined data format requirements to ensure that remote sensing image data from different data sources can be processed and analyzed consistently. Common remote sensing data formats include GeoTIFF, HDF, NetCDF, etc., to ensure the standardized processing of spatial information and attribute data of the data. The remote sensing data processing factors refer to the operation techniques that need to be applied during the preprocessing process. The remote sensing data processing factors include radiometric correction, geometric correction, and image enhancement. Specifically, the radiometric values of remote sensing images may be affected by factors such as sensors, time, and ground conditions. Radiometric correction eliminates these effects by adjusting the radiometric values in the image, making the ground information reflected by the image more accurate. Geometric distortion may occur in the process of acquiring remote sensing images due to factors such as the angle of the sensor and the curvature of the earth. Geometric correction corrects these distortions by performing geometric transformations (such as planar projection) on the image, making the geographic coordinates of the image consistent with the true ground coordinates. The visual effect of the image is improved or the features of certain ground objects are highlighted through enhancement techniques. For example, the contrast and brightness of the image are enhanced or noise is removed through filtering.

[0029] Preferably, the remote sensing data sources include remote sensing image data from different platforms and sensors, which may include images from satellite remote sensing platforms, such as image data from satellites like Landsat, Sentinel, WorldView, etc., with different resolutions, band combinations, and imaging methods; images from aerial remote sensing platforms, remote sensing image data captured by airplanes or drones, which usually have a high spatial resolution and are suitable for fine-grained regional analysis; lidar data (LiDAR), where the lidar system scans the ground with lasers to obtain high-precision three-dimensional terrain data; synthetic aperture radar (SAR) data, which uses radar bands to obtain ground reflection signals and can penetrate clouds and lighting conditions, and is particularly suitable for post-disaster monitoring or surface deformation monitoring; these remote sensing data sources are processed into a standardized data format through remote sensing data processing rules to obtain remote sensing image data, that is, data after the original remote sensing data has been preliminarily processed (such as radiometric correction, atmospheric correction, etc.), and is used for the identification, classification, and analysis of geographical elements.

[0030] Preferably, select and load the first remote sensing image data from multiple remote sensing data sources. The first remote sensing image data may be sourced from satellite images, aerial images, etc., and has not undergone any quality control or format conversion. According to the remote sensing data quality constraints (such as signal-to-noise ratio, image clarity, etc.), filter the first remote sensing image to remove parts that do not meet the quality standards (such as low-quality images, cloud interference areas, etc.) to obtain the second remote sensing image data that meets the quality requirements; then, according to the predetermined remote sensing data format, perform format conversion on the second remote sensing image data. Different remote sensing data sources may use different file formats, that is, convert the second remote sensing image data into a predetermined standard data format to unify the data structure and ensure that the data is compatible with subsequent processing to obtain the third remote sensing image data (meeting the quality requirements and format requirements); finally, perform preprocessing on the third remote sensing image data according to the remote sensing data processing factors, including applying remote sensing data processing factors (such as radiometric correction, geometric correction, image enhancement, etc.), adjusting the radiation information in the image so that ground features can be accurately reflected in the image, eliminating the influence of the atmosphere and sensors, performing coordinate transformation on the image to ensure that the image is aligned with the true ground coordinate system, eliminating geometric distortion, and highlighting important ground features or improving the image's recognizability by enhancing the visual performance of the image. Through quality filtering, format conversion, and preprocessing, etc., convert remote sensing data from different data sources into standardized and high-quality remote sensing image data, ensuring the accuracy, consistency, and reliability of the remote sensing image data, and enabling effective identification and analysis of geographical elements.

[0031] Step S200, obtain a geographical element identification instruction, where the geographical element identification instruction includes multi-level element identification targets, and the multi-level element identification targets include full-scale elements, prompt word elements, and point of interest elements.

[0032] Preferably, a geographical feature recognition instruction is obtained for recognizing and extracting geographical features according to different target levels. Each level of recognition target has specific functions and contents. Specifically, the geographical feature recognition instruction guides how geographical features in a remote sensing image are recognized and classified. The geographical feature recognition instruction includes multi-level feature recognition targets, that is, according to the requirements of the application scenario, different types and levels of geographical features can be specified for recognition. Among them, the multi-level feature recognition targets indicate that geographical feature recognition is not a single target, but a task divided into multiple levels or levels, with different recognition requirements and details. The multi-level feature recognition targets include full-scale elements, prompt word elements, and point of interest elements. The full-scale elements refer to all possible geographical feature categories, covering all kinds of ground object categories and features that may appear in the entire remote sensing image. All geographical features in the image are recognized and classified at this level, regardless of their quantity, type, or importance. For example, various ground objects such as urban buildings, farmland, forests, and water bodies are full-scale elements. The full-scale elements require the system to recognize all ground objects and natural features that appear in the image, and are usually used for geographical feature analysis of the entire region, such as large-scale land use and environmental monitoring.

[0033] Preferably, the prompt word elements refer to guiding the recognition of geographical features based on pre-set prompt words or keywords during the recognition process to help focus on specific ground object categories or regions. For example, the prompt words may be "bridge", "road", "river", etc. There are often multiple prompt word elements, and each prompt word points to a specific category or region. If there are specific bridges or roads in a certain remote sensing image, these specific structures are recognized based on prompt words such as "bridge" or "road", without caring about other irrelevant regions in the image. The addition of prompt words enhances the accuracy of image analysis and avoids interference from irrelevant parts during the recognition process; the point of interest elements refer to multiple regions or ground objects that are particularly concerned about in a specific application scenario. The point of interest can be a specific geographical location, a certain specific ground object (such as a historical site, a special building, a disaster-affected area, an important urban facility, etc.), or a sign of a certain special event (such as a fire or flood affected area). Recognizing the point of interest elements can help improve the monitoring and management of specific regions and meet the requirements of the application level. Through such hierarchical recognition, remote sensing image analysis can be more accurate and efficient, meeting the needs of different fields.

[0034] Step S300, perform multi-level image segmentation constraint registration on the remote sensing image data according to the multi-level feature recognition target, and establish an image segmentation constraint tree.

[0035] Preferably, according to the multi-level element recognition target, the acquired remote sensing image data is subjected to multi-level image segmentation, that is, the remote sensing image data is gradually refined and segmented according to the recognition requirements of different geographical elements (different scales, different levels). Specifically, first, a rough overall segmentation of all elements is performed to mark different major types of ground objects in the image (such as forests, farmland, water bodies, etc.), and then, based on specific targets such as prompt words and points of interest, the image is further finely segmented, focusing on specific areas or specific types of ground objects, such as identifying bridges, roads, buildings, etc. Through multi-level segmentation, it is ensured that geographical elements can be recognized from the global to the local, from the rough to the fine. Among them, image segmentation is a commonly used technique in computer vision and remote sensing image processing, and its purpose is to divide an image into several regions with certain similar characteristics (such as color, texture, shape, etc.), and each region represents a specific geographical feature or object.

[0036] Preferably, registration is to align remote sensing images from different sources, different times, or different sensors to the same coordinate system. The purpose of image registration is to eliminate geometric deformations caused by different shooting angles, shooting times, sensor differences, etc., and ensure the consistency between images. Specifically, when performing segmentation of remote sensing images, precise alignment of the images is carried out so that the segmented regions can accurately correspond to the actual geographical elements on the ground. By constraining the registration, the image accuracy and consistency between different segmentation levels can be ensured, especially in the case of multi-source data fusion (such as satellite images, aerial images, etc.), to avoid recognition errors caused by data source differences. Some ground objects may have relative position relationships between different levels, or there are spatial dependencies and constraints between the recognition targets at different levels (such as all elements, prompt word elements, etc.). Through these constrained registrations, it can be ensured that at different levels of image segmentation, the corresponding ground object features can be correctly aligned and associated. By organizing the segmentation results and constraint relationships at different levels in a tree structure, an image segmentation constraint tree is established to represent the image segmentation processes at different levels and their mutual constraint relationships, that is, the image segmentation constraint tree includes nodes at each segmentation level (i.e., different segmentation results of the image), and the constraint relationships between them. Among them, the hierarchical structure of the tree reflects the segmentation process from the rough segmentation of all elements to the detailed segmentation of prompt word elements or point of interest elements, and the edges or connections in the tree represent the constraint relationships between different segmentation levels. For example, some ground objects should maintain consistent spatial positions or shapes at different segmentation levels (for example, roads and buildings in urban areas should maintain coherent segmentation results at the same position).

[0037] Further, step S300 further includes step S310 of parsing the image segmentation constraint parameters for the remote sensing image data according to the full-scale elements to obtain the first image segmentation constraint condition; step S320 of parsing the image segmentation constraint parameters for the remote sensing image data according to the prompt word elements to obtain the second image segmentation constraint condition; step S330 of parsing the image segmentation constraint parameters for the remote sensing image data according to the point of interest elements to obtain the third image segmentation constraint condition; step S340 of using the full-scale elements as the first node for geographical element recognition, the prompt word elements as the second node for geographical element recognition, and the point of interest elements as the third node for geographical element recognition; step S350 of constructing an element recognition node tree according to the first node for geographical element recognition, the second node for geographical element recognition, and the third node for geographical element recognition; and step S360 of performing image segmentation constraint configuration on the element recognition node tree according to the first image segmentation constraint condition, the second image segmentation constraint condition, and the third image segmentation constraint condition to generate the image segmentation constraint tree.

[0038] Preferably, each geographical element (such as full-scale elements, prompt word elements, and point of interest elements) corresponds to a set of image segmentation constraint parameters, which define how to segment the image, so as to ensure that the segmentation results of different elements can accurately and finely reflect the geographical information in the image. Parsing the image segmentation constraint parameters for the remote sensing image data according to all geographical elements includes parsing the first image segmentation constraint condition through the characteristics of the full-scale elements (such as the spectral characteristics and shape characteristics of ground objects), that is, how to segment according to the full-scale ground object categories in the image (such as forests, farmlands, urban buildings, etc.) to ensure that each ground object can be accurately identified and distinguished in the image; generating the second image segmentation constraint condition through the constraint analysis of the ground object recognition guided by specific "prompt words" (such as "bridge" and "road") to ensure that the ground objects related to these prompt words in the image can be accurately segmented; and generating the third image segmentation constraint condition through the constraint analysis of the point of interest elements to ensure that the segmentation results can highlight and accurately identify these special regions or ground objects.

[0039] Preferably, the identified geographical elements are assigned to different nodes and organized into a tree structure according to these nodes. Each node represents a hierarchical geographical element recognition task, and there may be certain dependencies or hierarchical relationships between the nodes. Specifically, the full - scale elements are used as the first node for geographical element recognition, representing the most basic full - scale geographical element recognition of the remote - sensing image; the prompt - word elements are used as the second node for geographical element recognition, representing the recognition task for specific land - cover categories; the point - of - interest elements are used as the third node for geographical element recognition, representing the geographical features of particular interest in the image, such as certain specific buildings, urban areas or natural landscapes. The multi - level geographical element recognition process in the image is organized and managed through the nodes to construct an element recognition node tree. The root node of the tree represents the full - scale element recognition, while the other child nodes represent more refined tasks (such as prompt - word element recognition, point - of - interest element recognition). The hierarchical relationship of the tree structure indicates how the different - level recognition tasks depend on each other. Then, according to the first image - segmentation constraint condition, the second image - segmentation constraint condition, and the third image - segmentation constraint condition, image - segmentation constraint configuration is performed on the element recognition node tree, that is, these constraint conditions are applied to the element recognition node tree, so as to adjust and configure the segmentation tasks of each node, ensure that the image - segmentation process of each node can be carried out efficiently and accurately, and finally generate an image - segmentation constraint tree, which not only defines how to recognize different - level geographical elements, but also stipulates how to affect the details of segmentation through the constraint conditions, ensuring that the boundaries and categories of land - covers are accurately recognized in different - level segmentations, and further ensuring that the geographical element recognition of the remote - sensing image can be carried out more efficiently and accurately.

[0040] Further, step S310 further includes step S311 of retrieving twin data according to the remote - sensing image data to obtain Q remote - sensing image twin samples, where Q is a positive integer greater than 1; step S312 of collecting the historical parameters of the full - scale element image - segmentation constraints corresponding to the Q remote - sensing image twin samples to obtain Q full - scale element segmentation constraint groups, where each full - scale element segmentation constraint group includes M full - scale element image - segmentation constraint samples corresponding to each remote - sensing image twin sample, and M is a positive integer greater than 1; step S313 of performing a global confidence evaluation on the Q full - scale element segmentation constraint groups based on a global confidence evaluation channel to obtain Q global confidences of the constraint groups; step S314 of performing optimization and selection on the Q full - scale element segmentation constraint groups according to a predetermined global confidence based on the Q global confidences of the constraint groups to establish a full - scale element segmentation constraint space; step S315 of calculating a central value according to the full - scale element segmentation constraint space to generate the first image - segmentation constraint condition, where the first image - segmentation constraint condition includes M image - segmentation constraint central values.

[0041] Step S313 further includes that the global confidence evaluation channel includes a global confidence evaluation formula, and the global confidence evaluation formula is:

[0042] ;

[0043] where, represents the q-th constraint group global confidence corresponding to the q-th full-scale element segmentation constraint group, q is a positive integer, 1 ≤ q ≤ Q, represents the q-th constraint group support corresponding to the q-th full-scale element segmentation constraint group, represents the support of the m-th full-scale element image segmentation constraint sample in the q-th full-scale element segmentation constraint group.

[0044] Preferably, twin data retrieval refers to retrieving Q twin samples (Q is a positive integer greater than 1) similar to the current remote sensing image data from the historical dataset. These are historical image samples similar to the current remote sensing image data, with similar ground object features, geographical locations, or environmental backgrounds. Each remote sensing image twin sample has a corresponding historical segmentation constraint. The full-scale element segmentation constraint historical parameters extracted from the Q remote sensing image twin samples constitute Q full-scale element segmentation constraint groups. Each segmentation constraint group contains M full-scale element image segmentation constraint samples (M is a positive integer greater than 1), representing various possible constraint conditions when a full-scale element is segmented in the historical image. For example, different parameter setting strategies should be adopted according to different extraction methods, extraction object fineness, and actual hardware, etc. The parameters include the minimum mask area, filtering threshold, and stability threshold. Among them, the minimum mask area parameter needs to be greater than 0, and the filtering threshold and stability threshold range from 0 to 1.

[0045] Preferably, the global confidence evaluation channel is used to evaluate the global confidence of the Q full-scale element segmentation constraint groups, that is, to evaluate the reliability and effectiveness of each constraint group through certain criteria (such as segmentation accuracy, accuracy, matching degree of historical data, etc.). A higher global confidence means that the constraint group may be more effective and reliable in practical applications, and then the global confidence of the Q constraint groups is obtained, reflecting the effect of each group of constraints. Among them, the global confidence evaluation channel includes a global confidence evaluation formula, and the global confidence evaluation formula is:

[0046] ;

[0047] where, represents the q-th constraint group global confidence corresponding to the q-th full-scale element segmentation constraint group, q is a positive integer, 1 ≤ q ≤ Q, represents the q-th constraint group support corresponding to the q-th full-scale element segmentation constraint group, Characterize the support degree of the m-th full-element image segmentation constraint sample in the q-th full-element segmentation constraint group.

[0048] Preferably, according to the global confidence of each full-element segmentation constraint group, optimization and screening are carried out. The constraint group with high confidence is considered a more reliable segmentation standard, and the constraint group with lower confidence may be excluded or adjusted. The full-element segmentation constraint group corresponding to the global confidence of the constraint group greater than or equal to the predetermined global confidence is added to the full-element segmentation constraint space to ensure the accuracy and reliability of subsequent image segmentation. In the full-element segmentation constraint space, the central value of the constraint conditions is calculated. The central value usually refers to the average value or median of the constraint conditions, representing the best constraint standard. The calculation of the central value can balance the influence of different constraints and ensure that the finally selected constraint conditions are the most reasonable in practical applications. Finally, the first constraint condition for image segmentation is obtained, which includes M central values of image segmentation constraints and is the comprehensive evaluation result of all historical data constraint conditions. Each central value of image segmentation constraint represents a segmentation rule, providing an accurate constraint basis for the image segmentation task.

[0049] Further, step S320 further includes step S321, where the prompt word elements include multiple geographical element prompt words; step S322, the synonym expansion of the multiple geographical element prompt words is performed by the prompt word expansion large model to obtain multiple prompt word clusters; step S323, the multiple prompt word clusters and the remote sensing image data are input into the image target detection model to obtain multiple sets of target element annotation frames corresponding to the multiple prompt word clusters; step S324, coupling optimization is performed according to the multiple sets of target element annotation frames to generate the second constraint condition for image segmentation.

[0050] Preferably, the prompt word refers to the keyword used to guide image analysis. The prompt word elements include multiple geographical element prompt words, representing different types of ground object categories. Each prompt word corresponds to a specific geographical element, such as "building", "road", "river", "construction", "cultivated land", "green space", "vegetation", etc. Then, the synonym expansion of the multiple geographical element prompt words is performed by the prompt word expansion large model, that is, based on natural language processing (NLP) technology, the synonym expansion is performed through the existing prompt words. For example, based on a large-scale language model (such as BERT, etc.), each geographical element prompt word is expanded into a synonym cluster. For example, the prompt word "river" is expanded into "watercourse", "stream", "water system", etc., and the prompt word "road" is expanded into "street", "lane", "highway", etc., thereby obtaining multiple prompt word clusters. Each cluster contains a specific geographical element category and all its synonyms.

[0051] Preferably, multiple prompt clusters and remote sensing image data are input into an image object detection model. The image object detection model (such as YOLO, Faster R-CNN, etc.) automatically detects the target ground objects in the image (i.e., specific geographical elements in the remote sensing image) based on the input prompt clusters and remote sensing image data. Specifically, the object detection model uses the synonym information in the prompt clusters to search for matching regions in the image and generate corresponding annotation boxes, and can find the regions belonging to a certain ground object category in the image, thereby obtaining multiple sets of target element annotation boxes corresponding to multiple prompt clusters. These annotation box sets contain the position information of all the target elements identified in the image (i.e., their bounding boxes in the image), so as to identify the exact positions and categories of each ground object. There may be overlapping, redundant, or misdetected cases in the annotation boxes of image object detection. Finally, coupling optimization is performed based on multiple sets of target element annotation boxes, that is, association, integration, and optimization are carried out among multiple annotation boxes to ensure that they more accurately cover the correct ground object regions, such as merging redundant annotation boxes, correcting incorrect annotation boxes, or optimizing the size and position of the boxes, and finally generating the second constraint condition for image segmentation to ensure that the bounding boxes of the target ground objects (such as roads, buildings, rivers, etc.) in the image segmentation process are more accurate, thereby improving the accuracy of image segmentation.

[0052] Further, step S330 further includes step S331, where the interest point elements include multiple interest point geographical locations; step S332, constructing a coordinate system based on the remote sensing image data to obtain a remote sensing image coordinate system; step S333, performing coordinate transformation on the multiple interest point geographical locations based on the remote sensing image coordinate system to obtain element interest point spatial data; step S334, performing CSV conversion according to the element interest point spatial data to generate the third constraint condition for image segmentation.

[0053] Preferably, the interest point elements refer to specific geographical locations or regions that are of particular concern to users or tasks. For example, certain specific buildings, facilities, natural landscapes, disaster areas, etc. Each interest point element corresponds to one or more specific geographical location coordinates, indicating the regions that need to be focused on in the remote sensing image. A coordinate system is constructed based on the remote sensing image data, for example, constructed based on a geographic coordinate system (such as WGS84, UTM, etc.) to determine the exact geographical location corresponding to each pixel in the image on the earth's surface. The obtained remote sensing image coordinate system is the geographical coordinate system of each pixel point in the image.

[0054] Preferably, the geographical locations of multiple points of interest are coordinate-transformed based on the remote sensing image coordinate system and converted into the coordinate system used by the remote sensing image to ensure that these points of interest correspond to the actual ground object positions in the remote sensing image, obtaining the spatial data of the feature points of interest. That is, the spatial position of each point of interest will be matched with the remote sensing image data and converted into spatial data consistent with the remote sensing image coordinate system, including the accurate position and spatial information of the point of interest in the image. Finally, CSV conversion is performed according to the spatial data of the feature points of interest. That is, CSV conversion is performed according to the spatial data of the feature points of interest, and each point of interest will be represented as a table row, including its position (coordinates) in the image and related attributes (such as type, label, etc.). Among them, CSV (Comma-Separated Values) is a common data storage format that can convert geospatial data into a table form. Each row represents a point of interest and contains the relevant information of the point of interest (such as latitude and longitude coordinates, ground object type, other attributes, etc.). Finally, the third constraint condition for image segmentation is obtained, guiding the image segmentation algorithm to accurately identify and process these point-of-interest regions in the image, ensuring that the segmentation result accurately captures the region where the point of interest is located and avoiding mis-segmentation or missed segmentation.

[0055] Step S400, based on the image segmentation constraint tree, the remote sensing image data is segmented according to the SAM image segmentation large model to obtain a multi-level remote sensing image segmentation result.

[0056] Preferably, based on the image segmentation constraint tree, the remote sensing image data is segmented according to the SAM image segmentation large model. Among them, SAM is a powerful deep learning-based image segmentation large model, usually using a large amount of pre-training data and a powerful model architecture (such as convolutional neural network CNN, Transformers, etc.) for fine image segmentation, capable of achieving efficient and high-precision segmentation on different types of images, especially suitable for complex remote sensing image analysis. Specifically, using the SAM model and the image segmentation constraint tree, the remote sensing image data is segmented, that is, the SAM model refines the segmentation levels and strategies according to the constraint conditions in the image segmentation constraint tree, extracts the precise boundaries of different geographical features (such as forests, urban buildings, water bodies, etc.) in the remote sensing image, and assigns corresponding class labels to each ground object.

[0057] Preferably, for full-scale feature extraction, first input the image segmentation threshold parameters, such as the minimum mask area, into the SAM large image segmentation model, and then use the image feature segmentation and inference function of the SAM model to perform segmentation prediction processing on all geographical features within the range of the remote sensing image; for prompt extraction, first input the target feature annotation box detected by the target detection model according to the prompt into the SAM large image segmentation model, use the target feature annotation box as the constraint condition of the SAM model, and then use the image feature segmentation and inference function of the SAM model to perform segmentation prediction processing on the detected geographical target features in the remote sensing image; for point of interest extraction, input the CSV file with X and Y geographical coordinates of the interest annotation points into the SAM large image segmentation model, use the coordinate points in the CSV file as the constraint condition of the SAM model, and then use the image feature segmentation and inference function of the SAM model to perform segmentation prediction processing on the geographical features annotated by the points of interest in the remote sensing image.

[0058] Preferably, since the image segmentation constraint tree is multi-level and can support the layer-by-layer refinement of remote sensing images, finally obtaining multi-level remote sensing image segmentation results, each level focusing on different types of ground objects (such as full-scale ground objects, specific prompt ground objects, point-of-interest ground objects, etc.), ensuring the accuracy and fineness of the final image segmentation. For complex remote sensing image analysis tasks, it can meet the recognition needs at different levels and improve the efficiency and accuracy of remote sensing image processing.

[0059] Step S500: Optimize the segmentation loss of the multi-level remote sensing image segmentation result according to the image segmentation loss detection model to obtain the optimized result of the multi-level remote sensing image segmentation.

[0060] Preferably, optimize the segmentation loss of the multi-level remote sensing image segmentation result through the image segmentation loss detection model to improve the quality of image segmentation, making the segmentation result more accurate and fine. Specifically, the image segmentation loss detection model is a model used to evaluate the quality of remote sensing image segmentation results. Usually, it judges the quality of the image segmentation result according to certain preset criteria. The loss detection model evaluates the accuracy and rationality of the segmentation result by detecting different errors in the image segmentation process, such as insufficient segmentation accuracy, over-segmentation, and under-segmentation. Optimize and adjust the segmentation loss (such as problems like low segmentation accuracy, over-segmentation, or under-segmentation) detected by the image segmentation loss detection model, and make fine adjustments in multiple levels of segmentation to ensure that the segmentation at each level can correctly and accurately reflect the ground object features, thereby obtaining the optimized result of the multi-level remote sensing image segmentation, which is more accurate and effective at multiple levels, including ensuring more accurate segmentation of each ground object category, reducing classification errors, as well as reducing the over-segmentation rate and under-segmentation rate, reducing redundant or missed segmentation areas, improving the overall segmentation quality, and ensuring effectiveness and accuracy in remote sensing image data.

[0061] Further, step S500 further includes step S510 of detecting the segmentation loss of the multi-level remote sensing image segmentation result according to the image segmentation loss detection model to obtain each image segmentation loss detection result, where the image segmentation loss detection model includes multi-dimensional indicators for image segmentation loss detection, and the multi-dimensional indicators for image segmentation loss detection include segmentation accuracy, over-segmentation rate, and under-segmentation rate; step S520 of determining whether each image segmentation loss detection result meets the multi-dimensional constraints for segmentation loss detection; step S530 of generating a segmentation loss optimization instruction if each image segmentation loss detection result does not meet the multi-dimensional constraints for segmentation loss detection; and step S540 of correcting the segmentation loss of the multi-level remote sensing image segmentation result according to the multi-dimensional constraints for segmentation loss detection based on the segmentation loss optimization instruction to obtain an optimized result of the multi-level remote sensing image segmentation.

[0062] Preferably, the segmentation loss of the multi-level remote sensing image segmentation result is detected according to the image segmentation loss detection model. The image segmentation loss detection model includes multi-dimensional indicators for image segmentation loss detection, and the multi-dimensional indicators for image segmentation loss detection include segmentation accuracy, over-segmentation rate, and under-segmentation rate. Among them, the segmentation accuracy reflects the matching degree between the segmentation result and the actual ground objects. Over-segmentation means that a ground object in the image is wrongly segmented into multiple small parts, which will lead to additional calculations and unnecessary regional divisions. Under-segmentation means that the ground objects that should have been separated are wrongly merged into one area, which will lead to information loss and inaccurate ground object classification. Each image segmentation loss detection result is obtained to describe whether there are problems in terms of accuracy, over-segmentation, and under-segmentation for each segmentation result. It is determined whether each image segmentation loss detection result meets the multi-dimensional constraints for segmentation loss detection. Among them, the multi-dimensional constraints for segmentation loss detection are preset standards for ensuring the quality of the segmentation result, including segmentation accuracy constraints (the segmentation accuracy should reach a certain threshold), over-segmentation rate constraints, and under-segmentation rate constraints (the over-segmentation rate and under-segmentation rate should be lower than a certain threshold).

[0063] Preferably, if the image segmentation loss detection results do not meet the multi-dimensional constraints of the segmentation loss detection, a segmentation loss optimization instruction is generated to adjust for segmentation accuracy, over-segmentation, or under-segmentation. For example, the parameters of the segmentation model are adjusted to improve the segmentation accuracy, the model is optimized to avoid under-segmentation, etc. Finally, based on the segmentation loss optimization instruction, the segmentation loss correction of the multi-level remote sensing image segmentation result is performed according to the multi-dimensional constraints of the segmentation loss detection, that is, the original segmentation result is adjusted, which may include modifying the parameters or structure in the algorithm to make the segmentation result more accurate in terms of accuracy and segmentation boundaries; reducing the over-segmentation rate and avoiding unnecessary splitting of ground objects in the image into multiple regions; at the same time, avoiding under-segmentation to ensure that different ground objects are correctly segmented, and finally obtaining the optimized result of the multi-level remote sensing image segmentation, which has higher segmentation accuracy, more reasonable regional division, and lower over-segmentation and under-segmentation problems, ensuring the accurate and reasonable identification and segmentation of ground objects in the image.

[0064] Step S600, perform vectorized feature recognition based on the optimized result of the multi-level remote sensing image segmentation to obtain a multi-level geographical feature recognition result.

[0065] Preferably, performing vectorization according to the optimized result of the multi-level remote sensing image segmentation means converting the raster data (i.e., the image composed of pixels) of the optimized result of the multi-level remote sensing image segmentation into vector data. Among them, each pixel in the raster data usually corresponds to a small area in the image, while vector data uses geometric figures such as points, lines, and surfaces to represent the boundaries of ground objects. Specifically, converting the segmentation result (usually regions or blocks) into more expressive vector boundaries, such as polygons, line segments, etc., helps to improve the accuracy of ground object representation. For example, through vectorization, the boundaries of urban areas, river basins, etc. can be converted into data with clear boundaries and geometric forms; then, feature recognition is performed on the vectorized result, that is, specific geographical features (such as buildings, roads, water bodies, etc.) in the image are recognized, and finally a multi-level geographical feature recognition result is obtained. Each level corresponds to different levels of geographical feature recognition tasks. For example, each level corresponds to different levels of geographical feature recognition tasks including all significant ground object categories in the image (such as water bodies, cities, agricultural areas, etc.); the prompt word feature recognition result further refines the recognition of specific ground object categories; the point of interest feature recognition result performs fine recognition based on specific regions or key ground objects, thereby improving the accuracy and reliability of remote sensing image analysis.

[0066] In the above text, reference is made to Figure 1 The method for automatically recognizing geographical features of remote sensing images based on large model analysis according to an embodiment of the present invention is described in detail. Next, reference will be made to Figure 2 Describe the system for automatically recognizing geographical features of remote sensing images based on large model analysis according to an embodiment of the present invention.

[0067] The automatic recognition system for remote sensing image geographical elements based on large model analysis according to an embodiment of the present invention is used to solve the technical problems of insufficient recognition accuracy, low efficiency, and inaccurate image segmentation of multi-level geographical elements in the prior art, realize the efficient and accurate recognition of geographical elements, and achieve the technical effect of improving the recognition efficiency and accuracy of geographical elements. The automatic recognition system for remote sensing image geographical elements based on large model analysis includes: a remote sensing image data acquisition module 10, a geographical element recognition instruction acquisition module 20, an image segmentation constraint tree establishment module 30, a remote sensing image segmentation result acquisition module 40, a remote sensing image segmentation optimization result acquisition module 50, and a geographical element recognition result acquisition module 60.

[0068] The remote sensing image data acquisition module 10 is used to obtain remote sensing image data based on remote sensing data processing rules and according to multiple remote sensing data sources; the geographical element recognition instruction acquisition module 20 is used to obtain geographical element recognition instructions, wherein the geographical element recognition instructions include multi-level element recognition targets, and the multi-level element recognition targets include full amount elements, prompt word elements, and point of interest elements; the image segmentation constraint tree establishment module 30 is used to perform multi-level image segmentation constraint registration on the remote sensing image data according to the multi-level element recognition targets and establish an image segmentation constraint tree; the remote sensing image segmentation result acquisition module 40 is used to perform segmentation processing on the remote sensing image data based on the image segmentation constraint tree and according to the SAM image segmentation large model to obtain multi-level remote sensing image segmentation results; the remote sensing image segmentation optimization result acquisition module 50 is used to optimize the segmentation loss of the multi-level remote sensing image segmentation results according to an image segmentation loss detection model to obtain multi-level remote sensing image segmentation optimization results; the geographical element recognition result acquisition module 60 is used to perform vectorized element recognition according to the multi-level remote sensing image segmentation optimization results to obtain multi-level geographical element recognition results.

[0069] Next, the specific configuration of the image segmentation constraint tree building module 30 will be described in detail. The image segmentation constraint tree building module 30 further includes: parsing the image segmentation constraint parameters of the remote sensing image data according to the full amount of elements to obtain the first image segmentation constraint condition; parsing the image segmentation constraint parameters of the remote sensing image data according to the prompt word elements to obtain the second image segmentation constraint condition; parsing the image segmentation constraint parameters of the remote sensing image data according to the point of interest elements to obtain the third image segmentation constraint condition; using the full amount of elements as the first node for geographical element recognition, using the prompt word elements as the second node for geographical element recognition, and using the point of interest elements as the third node for geographical element recognition; constructing an element recognition node tree according to the first node for geographical element recognition, the second node for geographical element recognition, and the third node for geographical element recognition; performing image segmentation constraint configuration on the element recognition node tree according to the first image segmentation constraint condition, the second image segmentation constraint condition, and the third image segmentation constraint condition to generate the image segmentation constraint tree.

[0070] Next, the specific configuration of the image segmentation constraint tree building module 30 will be further described in detail. The image segmentation constraint tree building module 30 further includes: retrieving twin data from the remote sensing image data to obtain Q remote sensing image twin samples, where Q is a positive integer greater than 1; collecting the historical parameters of the full amount of element image segmentation constraints corresponding to the Q remote sensing image twin samples to obtain Q full amount of element segmentation constraint groups, where each full amount of element segmentation constraint group includes M full amount of element image segmentation constraint samples corresponding to each remote sensing image twin sample, and M is a positive integer greater than 1; performing global confidence evaluation on the Q full amount of element segmentation constraint groups based on the global confidence evaluation channel to obtain Q global confidences of the constraint groups; based on the Q global confidences of the constraint groups, performing optimization selection on the Q full amount of element segmentation constraint groups according to a predetermined global confidence to establish a full amount of element segmentation constraint space; calculating the central value according to the full amount of element segmentation constraint space to generate the first image segmentation constraint condition, where the first image segmentation constraint condition includes M image segmentation constraint central values.

[0071] Next, the specific configuration of the image segmentation constraint tree building module 30 will be further described in detail. The image segmentation constraint tree building module 30 further includes: the global confidence evaluation channel includes a global confidence evaluation formula, and the global confidence evaluation formula is:

[0072] ;

[0073] where represents the q-th global confidence of the constraint group corresponding to the q-th full amount of element segmentation constraint group, q is a positive integer, 1 ≤ q ≤ Q, Characterize the support degree of the q-th constraint group corresponding to the q-th full-element segmentation constraint group, Characterize the support degree of the m-th full-element image segmentation constraint sample in the q-th full-element segmentation constraint group.

[0074] Next, the specific configuration of the image segmentation constraint tree building module 30 will be further described in detail. The image segmentation constraint tree building module 30 further includes: The prompt word elements include multiple geographic element prompt words. The prompt word expansion large model performs synonym expansion on the multiple geographic element prompt words to obtain multiple prompt word clusters; The multiple prompt word clusters and the remote sensing image data are input into the image target detection model to obtain multiple target element annotation box sets corresponding to the multiple prompt word clusters; According to the multiple target element annotation box sets, coupling optimization is performed to generate the second image segmentation constraint condition.

[0075] Next, the specific configuration of the image segmentation constraint tree building module 30 will be further described in detail. The image segmentation constraint tree building module 30 further includes: The point of interest elements include multiple point of interest geographical locations; According to the remote sensing image data, a coordinate system is constructed to obtain a remote sensing image coordinate system; Based on the remote sensing image coordinate system, coordinate conversion is performed on the multiple point of interest geographical locations to obtain element point of interest spatial data; According to the element point of interest spatial data, CSV conversion is performed to generate the third image segmentation constraint condition.

[0076] Next, the specific configuration of the remote sensing image segmentation optimization result obtaining module 50 will be described in detail. The remote sensing image segmentation optimization result obtaining module 50 further includes: According to the image segmentation loss detection model, segmentation loss detection is performed on the multi-level remote sensing image segmentation result to obtain each image segmentation loss detection result, where the image segmentation loss detection model includes image segmentation loss detection multi-dimensional indicators, and the image segmentation loss detection multi-dimensional indicators include segmentation accuracy, over-segmentation rate, and under-segmentation rate; Determine whether each image segmentation loss detection result meets the multi-dimensional constraints of the segmentation loss detection; If each image segmentation loss detection result does not meet the multi-dimensional constraints of the segmentation loss detection, generate a segmentation loss optimization instruction; Based on the segmentation loss optimization instruction, according to the multi-dimensional constraints of the segmentation loss detection, the multi-level remote sensing image segmentation result is corrected for segmentation loss to obtain a multi-level remote sensing image segmentation optimization result.

[0077] Next, the specific configuration of the remote sensing image data acquisition module 10 will be described in detail. The remote sensing image data acquisition module 10 further includes: The remote sensing data processing rules include remote sensing data quality constraints, a predetermined remote sensing data format, and remote sensing data processing factors. The remote sensing data processing factors include radiometric correction, geometric correction, and image enhancement; According to the multiple remote sensing data sources, load the first remote sensing image data; According to the remote sensing data quality constraints, filter the first remote sensing image data to obtain the second remote sensing image data; According to the predetermined remote sensing data format, perform format conversion on the second remote sensing image data to obtain the third remote sensing image data; Preprocess the third remote sensing image data according to the remote sensing data processing factors to generate the remote sensing image data.

[0078] The remote sensing image geographic feature automatic recognition system based on large model analysis provided by the embodiments of the present invention can execute the remote sensing image geographic feature automatic recognition method provided by any embodiment of the present invention, and has functional modules and beneficial effects corresponding to the execution of the method.

[0079] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, however, any number of different modules can be used and run on the user terminal and / or server. The various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; In addition, the specific names of the functional units are only for the convenience of mutual distinction and do not limit the protection scope of the present invention.

[0080] The above specific embodiments do not constitute a limitation to the protection scope of the present application. Those skilled in the art should understand that various modifications, combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principle of the present application shall be included within the protection scope of the present application.

Claims

1. A method for automatically identifying geographic elements in remote sensing images based on large model analysis, characterized in that: The method comprises: Based on remote sensing data processing rules, remote sensing image data is obtained according to multiple remote sensing data sources; Obtaining a geographic element recognition instruction, wherein the geographic element recognition instruction includes a multi-level element recognition target, and the multi-level element recognition target includes a full element, a prompt word element, and an interest point element; Perform multi-level image segmentation constraint registration on the remote sensing image data according to the multi-level element recognition target, and establish an image segmentation constraint tree; Based on the image segmentation constraint tree, the remote sensing image data is segmented according to the SAM image segmentation model to obtain a multi-level remote sensing image segmentation result; Performing segmentation loss optimization on the multi-level remote sensing image segmentation result according to the image segmentation loss detection model to obtain a multi-level remote sensing image segmentation optimization result; Performing vectorized element recognition based on the multi-level remote sensing image segmentation optimization result to obtain a multi-level geographic element recognition result; Performing multi-level image segmentation constraint registration on the remote sensing image data according to the multi-level element recognition target, and establishing an image segmentation constraint tree, including: Performing image segmentation constraint parameter analysis on the remote sensing image data according to the total amount of elements to obtain a first constraint condition for image segmentation; Perform image segmentation constraint parameter analysis on the remote sensing image data according to the prompt word elements to obtain a second image segmentation constraint condition; Performing image segmentation constraint parameter analysis on the remote sensing image data according to the point of interest elements to obtain a third image segmentation constraint condition; A first node is identified by taking the total amount element as a geographical element, a second node is identified by taking the prompt word element as a geographical element, and a third node is identified by taking the point of interest element as a geographical element; Constructing a feature identification node tree according to the geographic feature identification first node, the geographic feature identification second node and the geographic feature identification third node; Performing image segmentation constraint registration on the element recognition node tree according to the first image segmentation constraint condition, the second image segmentation constraint condition and the third image segmentation constraint condition to generate the image segmentation constraint tree; The remote sensing image data is subjected to image segmentation constraint parameter analysis according to the total amount of elements to obtain a first image segmentation constraint condition, including: Perform twin data retrieval based on the remote sensing image data to obtain Q remote sensing image twin samples, where Q is a positive integer greater than 1; Collect the full-factor image segmentation constraint historical parameters corresponding to the Q remote sensing image twin samples to obtain Q full-factor segmentation constraint groups, wherein each full-factor segmentation constraint group includes M full-factor image segmentation constraint samples corresponding to each remote sensing image twin sample, and M is a positive integer greater than 1; Based on the global confidence evaluation channel, a global confidence evaluation is performed on the Q full-factor segmentation constraint groups to obtain global confidences of the Q constraint groups; Based on the global confidences of the Q constraint groups, the Q full-factor segmentation constraint groups are optimally selected according to a predetermined global confidence to establish a full-factor segmentation constraint space; The concentrated value is calculated according to the total element segmentation constraint space to generate the first constraint condition for image segmentation, wherein the first constraint condition for image segmentation includes M concentrated values ​​of image segmentation constraints; the concentrated value refers to the average value and median of the constraint condition.

2. The method for automatic identification of geographical elements in remote sensing images based on large model analysis according to claim 1, characterized in that: The global confidence evaluation channel includes a global confidence evaluation formula, and the global confidence evaluation formula is: ； Among them, KSZ q Characterizes the global confidence of the qth constraint group corresponding to the qth full-factor segmentation constraint group, q is a positive integer, 1≤q≤Q, f q represents the support of the qth constraint group corresponding to the qth full-element segmentation constraint group, f qm Characterizes the support of the mth full-element image segmentation constraint sample in the qth full-element segmentation constraint group.

3. The method for automatic identification of geographical elements in remote sensing images based on large model analysis according to claim 1, characterized in that: The remote sensing image data is analyzed for image segmentation constraint parameters according to the prompt word elements to obtain a second image segmentation constraint condition, including: The prompt word element includes a plurality of geographic element prompt words; Synonymically expanding the plurality of geographic element prompt words according to the prompt word expansion model to obtain a plurality of prompt word clusters; Inputting the plurality of prompt word clusters and the remote sensing image data into an image target detection model to obtain a plurality of target element annotation frame sets corresponding to the plurality of prompt word clusters; Coupling optimization is performed according to the multiple target element annotation frame sets to generate the second constraint condition for image segmentation.

4. The method for automatic identification of geographical elements in remote sensing images based on large model analysis according to claim 1, characterized in that: The remote sensing image data is subjected to image segmentation constraint parameter parsing according to the interest point elements to obtain a third image segmentation constraint condition, including: The point of interest elements include a plurality of geographical locations of points of interest; Constructing a coordinate system according to the remote sensing image data to obtain a remote sensing image coordinate system; Based on the remote sensing image coordinate system, coordinate transformation is performed on the geographical locations of the plurality of interest points to obtain spatial data of element interest points; The third constraint condition for image segmentation is generated by performing CSV conversion according to the spatial data of the element interest points.

5. The method for automatic identification of geographical elements in remote sensing images based on large model analysis according to claim 1, characterized in that: The multi-level remote sensing image segmentation result is optimized according to the image segmentation loss detection model to obtain the multi-level remote sensing image segmentation optimization result, including: Performing segmentation loss detection on the multi-level remote sensing image segmentation results according to the image segmentation loss detection model to obtain segmentation loss detection results of each image, wherein the image segmentation loss detection model includes a multi-dimensional index for image segmentation loss detection, and the multi-dimensional index for image segmentation loss detection includes segmentation accuracy, over-segmentation rate, and under-segmentation rate; Determine whether the segmentation loss detection results of each image meet the multi-dimensional constraints of segmentation loss detection; If the segmentation loss detection results of each image do not satisfy the segmentation loss detection multi-dimensional constraint, generating a segmentation loss optimization instruction; Based on the segmentation loss optimization instruction, segmentation loss correction is performed on the multi-level remote sensing image segmentation result according to the segmentation loss detection multi-dimensional constraint to obtain a multi-level remote sensing image segmentation optimization result.

6. The method for automatic identification of geographical elements in remote sensing images based on large model analysis according to claim 1, characterized in that: Based on remote sensing data processing rules, remote sensing image data is obtained from multiple remote sensing data sources, including: The remote sensing data processing rules include remote sensing data quality constraints, a predetermined remote sensing data format and remote sensing data processing factors, and the remote sensing data processing factors include radiation correction, geometric correction and image enhancement; Loading first remote sensing image data according to the multiple remote sensing data sources; filtering the first remote sensing image data according to the remote sensing data quality constraint to obtain second remote sensing image data; According to the predetermined remote sensing data format, format conversion is performed on the second remote sensing image data to obtain third remote sensing image data; The third remote sensing image data is preprocessed according to the remote sensing data processing factor to generate the remote sensing image data.

7. The automatic identification system of geographical elements in remote sensing images based on large model analysis is characterized by: The system is used to implement the method for automatically identifying geographical elements in remote sensing images based on large model analysis according to any one of claims 1 to 6, and the system comprises: A remote sensing image data acquisition module is used to obtain remote sensing image data based on remote sensing data processing rules and multiple remote sensing data sources; A geographic element recognition instruction acquisition module, used to obtain a geographic element recognition instruction, wherein the geographic element recognition instruction includes a multi-level element recognition target, and the multi-level element recognition target includes a full element, a prompt word element, and an interest point element; An image segmentation constraint tree establishment module is used to perform multi-level image segmentation constraint registration on the remote sensing image data according to the multi-level element recognition target to establish an image segmentation constraint tree; A remote sensing image segmentation result acquisition module is used to perform segmentation processing on the remote sensing image data based on the image segmentation constraint tree and the SAM image segmentation large model to obtain a multi-level remote sensing image segmentation result; A remote sensing image segmentation optimization result acquisition module is used to optimize the segmentation loss of the multi-level remote sensing image segmentation result according to the image segmentation loss detection model to obtain the multi-level remote sensing image segmentation optimization result; The geographic element recognition result acquisition module is used to perform vectorized element recognition according to the multi-level remote sensing image segmentation optimization result to obtain the multi-level geographic element recognition result.

Images