Natural resource element remote sensing intelligent identification system

Abstract

The invention provides a natural resource element remote sensing intelligent identification system, and relates to the technical field of data processing, and the system comprises a sample production subsystem which is used for making and storing a multi-source remote sensing data sample set; the model training subsystem is used for a user to construct a model based on an application reasoning architecture, load the remote sensing multi-source sample set to a model optimization link, train and optimize the model, analyze a training result and publish the model; the intelligent identification subsystem is used for carrying out remote sensing earth surface classification, deformation accumulation area detection and other applications based on the published model; the management subsystem is used for providing access modes of various data types and providing a unified data source management mode for model research and development and authority management; and managing the data content. According to the method, sample construction, model training and intelligent identification are integrated, function expansion, process customization and model iteration are realized, and the calculation efficiency, the monitoring precision and the application capability of natural resource element extraction are improved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a remote sensing intelligent identification system for natural resource elements. Background Technology

[0002] With the development of remote sensing technology, the key issue has become how to quickly and accurately extract target elements from massive amounts of data. In recent years, artificial intelligence technology, especially deep learning algorithms, has developed rapidly, providing new solutions for intelligent recognition of remote sensing images. With the continuous development of deep learning technology, especially the emergence of large-scale remote sensing models, intelligent recognition technology for remote sensing elements has entered a new stage. In the future, intelligence, real-time processing, multi-source data fusion, and cloud platform services will become important development directions for remote sensing technology.

[0003] The development of intelligent remote sensing identification is rapid both domestically and internationally. Google's Google Earth Engine cloud platform, launched in 2010, has become a crucial tool in the field of remote sensing and is currently the most widely used cloud platform for remote sensing image processing and analysis. Domestically, intelligent remote sensing element identification platforms are developing rapidly. Remote sensing and geographic information cloud service platforms, represented by PIE-Engine launched by Aerospace Hongtu, have evolved from simple multi-source remote sensing data processing tools into a new generation of smart earth platforms capable of handling massive amounts of Earth observation data and conducting spatiotemporal intelligent analysis. Alibaba Cloud's Earth Science Cloud Platform, AIEarth, provides AI interpretation tools such as land cover classification, change detection, and target recognition, as well as basic processing tools such as index calculation and band synthesis.

[0004] The booming development of intelligent remote sensing element recognition platforms at home and abroad reflects the transformation of remote sensing interpretation from traditional remote sensing technology to AI-driven intelligent remote sensing, and the level of automation and intelligence has been significantly improved. However, some common problems cannot be ignored: (1) Data security and privacy issues. If data is accessed, tampered with, or abused without authorization, it will affect the safety and interests of users. (2) Cost and resource constraints. The long-term operating cost investment is not to be underestimated, and once the service is stopped, the consequences will be unimaginable. (3) Technology dependence and autonomy issues. Users may be technically constrained by the platform provider and lack autonomy and flexibility. For example, when the platform's technology is updated, it may affect the user's business continuity. In addition, the applications opened by the platform are relatively mature, and their algorithm rules are unknown or highly encapsulated. During use, most of them cannot be further optimized or deepened, let alone "customized" optimization of certain functions. Especially for users with a certain development foundation, they cannot give full play to their advantages and strengths.

[0005] While existing remote sensing intelligent identification platforms are powerful, they have limitations in data security protection, meeting specific industry needs, and long-term cost control. Therefore, it is particularly important to build a remote sensing element intelligent identification platform that meets the application needs of the natural resources field, ensures data security, achieves technological autonomy and controllability, and can provide secure and reliable support for efficient management and scientific decision-making in natural resources. Summary of the Invention

[0006] To address the aforementioned issues, this invention provides a remote sensing intelligent identification system for natural resource elements, breaking through the bottlenecks in the full-process, large-scale intelligent identification application of natural resource elements. It achieves scalable functions, customizable processes, and iterative models, thereby improving the computational efficiency, monitoring accuracy, and application capabilities of natural resource element extraction.

[0007] To achieve the above objectives, the present invention provides a remote sensing intelligent identification system for natural resource elements, comprising a sample production subsystem, a model training subsystem, an intelligent identification subsystem, and a management subsystem; the sample production subsystem includes a sample acquisition module and a sample set creation module; the model training subsystem includes a model development module and a model release module; the intelligent identification subsystem includes a remote sensing surface classification module and a deformation cluster detection module; and the management subsystem includes a data management module, a code management module, and a resource management module.

[0008] The sample production subsystem is used for:

[0009] The sample acquisition module collects multi-source, multi-type remote sensing data.

[0010] The sample set creation module creates a remote sensing multi-source sample set, including reading remote sensing images, setting the sample size and cropping step size, and storing the sample set.

[0011] The model training subsystem is used for:

[0012] The model development module is configured to allow users to build models based on application inference architecture;

[0013] The remote sensing multi-source sample set is loaded into the model optimization stage through the model publishing module to train and optimize the model, analyze the training results and publish the model.

[0014] The intelligent recognition subsystem is used for:

[0015] The remote sensing surface classification module performs remote sensing surface classification based on the published model, and the deformation cluster detection module performs deformation cluster detection.

[0016] The management subsystem is used for:

[0017] The data management module is configured to provide access methods for various data types, offer a unified data source management approach for model development and access control, and manage data content.

[0018] The code management module provides a Notebook-style code-level programming tool and modeling tools. It supports managing script files and pre-trained models as Notebook dependencies and automatically formats the edited code according to Python PEP 8 specifications after the user finishes writing the code or during the coding process.

[0019] The resource management module is configured to support the number of CPU cores, CPU memory capacity, and GPU memory for different cluster types, and to configure the single available time for users of different cluster types.

[0020] As a further improvement to the present invention, the access methods for multiple data types include:

[0021] Access methods include user-owned data file uploads and online access to existing data storage media, including online database access and NAS online access.

[0022] As a further improvement of the present invention, managing data content includes:

[0023] Data extraction and dataset construction from multiple sources, data source version management, unified management and content extraction of the accessed data.

[0024] As a further improvement of the present invention, the model training subsystem is used for:

[0025] The remote sensing multi-source sample set is loaded into the model optimization stage. The smooth operation of the model is ensured through mirror creation and resource scheduling. Then, the intelligent model is trained and optimized, the training results are analyzed, and the model is released.

[0026] As a further improvement of the present invention, the modeling tool provides code snippet library, resource usage analysis, variable tracking, code standardization and version management functions.

[0027] As a further improvement of the present invention, the provision to support configuring the single availability duration for users of different cluster types includes:

[0028] Users are divided into high-priority users and low-priority users. The single available time allocated to high-priority users is longer than that allocated to low-priority users. The minimum division granularity is 1 minute.

[0029] As a further improvement of the present invention, the resource management module supports the configuration of resource reclamation time for different cluster types, with a minimum division granularity of 1 minute.

[0030] As a further improvement of the present invention, if the user closes the application model within the resource reclamation time, the user will still occupy the resource; if the resource reclamation time is exceeded, resource reclamation will be triggered.

[0031] As a further improvement of the present invention, the resource management module supports managing multiple GPUs as a cluster, supports configuring the number of GPUs, and supports configuring the user workspace size and image version for different cluster types.

[0032] As a further improvement of the present invention, the application inference architecture includes overall design, parallel scheduling technology, plug-in workflow technology, and algorithm integration specification design.

[0033] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0034] This invention provides a remote sensing intelligent identification system for natural resource elements that integrates sample construction, model training, and intelligent identification. It breaks through the bottleneck of full-process, large-scale intelligent identification of natural resource elements, and achieves scalable functions, customizable processes, and iterative models, thereby improving the computational efficiency, monitoring accuracy, and application capabilities of natural resource element extraction.

[0035] This invention constructs an iterative training framework of "sample-training-optimization" driven by intelligent identification applications through overall system architecture design. Through business process design, it forms a sample library of multi-source data, an algorithm model library suitable for natural resource element extraction, and a knowledge base for publishing model applications, laying the foundation for remote sensing intelligent interpretation of natural resource elements. Through the design of system functional modules, it realizes multi-source data access, sample set creation, data security and privacy management, and can meet users' needs for various remote sensing intelligent identification systems. Attached Figure Description

[0036] Figure 1 This is a schematic diagram of a remote sensing intelligent identification system for natural resource elements disclosed in one embodiment of the present invention;

[0037] Figure 2 This is a diagram illustrating the overall architecture of a remote sensing intelligent identification system for natural resource elements, as disclosed in one embodiment of the present invention.

[0038] Figure 3 This is a business process diagram of a remote sensing intelligent identification system for natural resource elements disclosed in one embodiment of the present invention;

[0039] Figure 4 This is a schematic diagram illustrating a sample set creation module disclosed in one embodiment of the present invention;

[0040] Figure 5This is a schematic diagram of a page for configuring data connection in a data management module according to an embodiment of the present invention;

[0041] Figure 6 This is a schematic diagram illustrating the data management module for page search data source in one embodiment of the present invention;

[0042] Figure 7 This is a Notebook mode interface diagram of the remote sensing intelligent identification system for natural resource elements disclosed in one embodiment of the present invention;

[0043] Figure 8 This is a schematic diagram of the computational resource management module disclosed in one embodiment of the present invention. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0045] The present invention will now be described in further detail with reference to the accompanying drawings:

[0046] like Figure 1 As shown, the present invention provides a remote sensing intelligent identification system for natural resource elements. This system was developed through requirements analysis, system design, system implementation, and application. Specifically:

[0047] Demand Analysis: The system needs to have autonomous service capabilities, meet the requirements of an intelligent recognition system that can be developed independently, and at the same time ensure data security and achieve independent control of technology, so as to provide a safe and reliable system support for efficient management of natural resources and scientific decision-making; Functional Requirements: Starting from the integrated iterative training framework of "sample-training-optimization" corresponding to deep learning, functional requirements analysis is carried out for three aspects: sample production, intelligent model training, and intelligent model prediction, including: (1) Sample production requirements: Sample production for multi-source spatiotemporal data is the foundation of intelligent recognition of remote sensing images, and sample production and management tools for multi-source spatiotemporal data are needed; (2) Intelligent model training requirements: For the integrated training framework, intelligent model training is the core of platform construction. Based on the use of samples and algorithms, model training and arrangement are carried out, feature extraction is performed, and large-scale computing is achieved; (3) Intelligent model prediction requirements: Based on microservice architecture and container technology, an application inference framework that meets the requirements of configurable environment and parallel hardware is built, which can integrate and release a variety of prediction models and algorithms to support intelligent applications of "multi-element, multi-scenario, and multi-model". Non-functional requirements: (1) Multi-source data access management requirements: Remote sensing intelligent identification business involves multi-source data types, including remote sensing image data (optical data, InSAR data, DEM, etc.), geological element data, meteorological data, etc. At the same time, in intelligent identification business, there are both small-scale scientific research computing experiments and large-scale massive data processing. Therefore, data access needs to be flexible, and both sample data and large-scale data can be accessed into the system. Because natural resource element data and its processing results are sensitive, the data should have access permissions. (2) Code compilation requirements: At present, artificial intelligence technology is developing rapidly, and information extraction methods in the field of remote sensing are also constantly being updated and iterated. Remote sensing images themselves are highly complex, so the platform's built-in algorithms often cannot meet the personalized information extraction requirements. Therefore, the platform needs to support Jupyter Notebook for redesign and implementation. (3) Computational resource management requirements: In the extraction of natural resource elements, the flexible allocation of computing resources is particularly important. For scientific research computing tasks, the required amount of computing resources is relatively small, while for production computing tasks, the required amount of computing is continuous and stable, and the computing resources used are relatively large. How to ensure the stability of computing services to the greatest extent possible under limited computing resources? System Design: The system's overall architecture and business process were designed around the intelligent extraction of natural resource elements; Overall System Architecture: as follows... Figure 2 As shown, to address the need for intelligent identification of natural resource elements, a reasoning framework for remote sensing intelligent identification of natural resource elements is constructed, and the overall framework design is carried out. Based on a structured analysis model of "business-scenario-rules," an intelligent interpretation model library is designed, and an iterative training framework of "sample-training-optimization" driven by intelligent identification applications is constructed, laying the platform foundation for intelligent remote sensing interpretation of natural resource elements. Figure 3As shown, the business process design is as follows: Built on a microservice architecture and container technology, the system follows an integrated business process of "sample analysis - model iteration - application inference," constructing a natural resource element intelligent identification system that is "functionally scalable," "process customizable," and "model iterative." Based on multi-source and multi-type data, a remote sensing multi-source sample set is created; samples are loaded into the model optimization stage, and the smooth operation of the model is ensured through image creation and resource scheduling. The intelligent model is trained and optimized, and the results are analyzed and the model service is released; model applications are released for industry applications such as remote sensing surface classification and deformation cluster detection. This results in a sample library of multi-source data, an algorithm model library suitable for natural resource element extraction, and a knowledge base for releasing model applications. System implementation and application: Following the functional logic and overall business process of the system design, the system functions are developed and deployed. The functional architecture of the remote sensing natural resource element intelligent identification system is as follows: Figure 1 As shown, it includes a sample production subsystem, a model training subsystem, an intelligent recognition subsystem, and a management subsystem. The sample production subsystem includes a sample acquisition module and a sample set creation module. The model training subsystem includes a model development module and a model release module. The intelligent recognition subsystem includes a remote sensing surface classification module and a deformation cluster detection module. The management subsystem includes a data management module, a code management module, and a resource management module.

[0048] The sample production subsystem is used for:

[0049] Multi-source, multi-type remote sensing data are collected through the sample acquisition module;

[0050] The sample set creation module enables the creation of multi-source remote sensing sample sets from various remote sensing images. This includes directly reading remote sensing images, setting sample size and cropping step size, and storing the sample sets locally or on the system NAS disk. Figure 4 The image shown is a demonstration of the sample set creation module;

[0051] The model training subsystem is used for:

[0052] A model development module is set up to allow users to build models based on application inference architecture;

[0053] The remote sensing multi-source sample set is loaded into the model optimization stage through the model release module. The smooth operation of the model is ensured through mirror creation and resource scheduling. Then, the model is trained, optimized, and the training results are analyzed and released.

[0054] in,

[0055] The application inference architecture includes overall design, parallel scheduling technology, plug-in workflow technology, and algorithm integration specification design.

[0056] The intelligent recognition subsystem is used for:

[0057] The remote sensing surface classification module performs remote sensing surface classification based on the published model, and the deformation cluster detection module detects deformation clusters.

[0058] The management subsystem is used for:

[0059] The data management module provides access methods for various data types and a unified data source management approach for model development, enabling access and access control; it also manages data content.

[0060] The code management module provides a Notebook-style code-level programming tool and modeling tools. It supports managing script files and pre-trained models as Notebook dependencies and automatically formats the edited code according to Python PEP 8 specifications after the user finishes writing the code or during the coding process, making the code highly readable.

[0061] The resource management module supports configuring the number of CPU cores, CPU memory capacity, and GPU memory for different cluster types, and also supports configuring the single-use duration for different cluster types.

[0062] in,

[0063] Data Management Module:

[0064] Multiple data type access methods are supported, including user-owned data file uploads and online access to various existing data storage media, such as online database access and NAS online access. Furthermore, after logging into the system, users can click on three different connection methods—creating a new database connection, object storage connection, and NAS space—to enter the data connection configuration page. Figure 5 The image shows a diagram illustrating the clicking of a database connection. Furthermore, users can filter and search for the desired data source by its type; for example... Figure 6 As shown, users can filter and find the data sources they need by tags. Each data source can be tagged, and users only need to enter the relevant tags to quickly find all data sources containing those tags.

[0065] Data content management includes: multi-source data extraction and dataset construction, data source version management, unified data management and content extraction of the accessed data.

[0066] Code management module:

[0067] Notebook-based code-level programming tools enable in-depth exploration and model development for challenging research tasks related to remote sensing identification of natural resource elements, resulting in operationally relevant model outcomes. To meet the demands of interactivity and rapid model iteration during the research process, the platform's analysis interface needs to be primarily notebook-based. Jupyter Notebook has been redesigned and implemented to improve communication security, functional scalability, and user experience. The notebook-mode interface is shown in the image below. Figure 7 As shown.

[0068] In addition to the basic functions of the Notebook, the modeling tool provides a code snippet library, resource usage analysis, variable tracking, code standardization, and version management features.

[0069] like Figure 8 As shown, the resource management module:

[0070] The minimum granularity of CPU cores can be 0.5 cores, and the minimum granularity of memory and video memory can be 1 Mib.

[0071] It supports configuring the single available time for users of different cluster types, with the smallest division granularity being 1 minute. Users can be reminded before the time expires, so as to improve the resource utilization of the infrastructure cloud platform and avoid resources being occupied by a single user.

[0072] It supports configuring the single availability duration for users of different cluster types, including: dividing users into high-priority users and low-priority users, with high-priority users allocated a longer single availability duration than low-priority users, and the minimum division granularity being 1 minute.

[0073] The resource management module also supports configuring resource reclamation time for different cluster types, with a minimum granularity of 1 minute.

[0074] If a user closes the application model, they will still have access to the resource within the resource reclamation period to allow researchers to continue working. If the resource reclamation period is exceeded, resource reclamation will be triggered to prevent the center from consuming idle resources.

[0075] The resource management module supports managing multiple GPUs as a cluster and allows users to configure the number of GPUs themselves. It also supports configuring the user workspace size and image version for different cluster types, thereby matching the basic environment and intermediate process storage space requirements of different research scenarios.

[0076] Advantages of this invention:

[0077] This invention provides a remote sensing intelligent identification system for natural resource elements that integrates sample construction, model training, and intelligent identification. It breaks through the bottleneck of full-process, large-scale intelligent identification of natural resource elements, and achieves scalable functions, customizable processes, and iterative models, thereby improving the computational efficiency, monitoring accuracy, and application capabilities of natural resource element extraction.

[0078] This invention constructs an iterative training framework of "sample-training-optimization" driven by intelligent identification applications through overall system architecture design. Through business process design, it forms a sample library of multi-source data, an algorithm model library suitable for natural resource element extraction, and a knowledge base for publishing model applications, laying the foundation for remote sensing intelligent interpretation of natural resource elements. Through the design of system functional modules, it realizes multi-source data access, sample set creation, data security and privacy management, and can meet users' needs for various remote sensing intelligent identification systems.

[0079] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A remote sensing intelligent identification system for natural resource elements, characterized in that: It includes a sample production subsystem, a model training subsystem, an intelligent recognition subsystem, and a management subsystem; the sample production subsystem includes a sample acquisition module and a sample set creation module; the model training subsystem includes a model development module and a model release module; the intelligent recognition subsystem includes a remote sensing surface classification module and a deformation cluster detection module; and the management subsystem includes a data management module, a code management module, and a resource management module. The sample production subsystem is used for: The sample acquisition module collects multi-source, multi-type remote sensing data. The sample set creation module creates a remote sensing multi-source sample set, including reading remote sensing images, setting the sample size and cropping step size, and storing the sample set. The model training subsystem is used for: The model development module is configured to allow users to build models based on application inference architecture; The remote sensing multi-source sample set is loaded into the model optimization stage through the model publishing module to train and optimize the model, analyze the training results and publish the model. The intelligent recognition subsystem is used for: The remote sensing surface classification module performs remote sensing surface classification based on the published model, and the deformation cluster detection module performs deformation cluster detection. The management subsystem is used for: The data management module is configured to provide access methods for various data types, offer a unified data source management approach for model development and access control, and manage data content. The code management module provides a Notebook-style code-level programming tool and modeling tools. It supports managing script files and pre-trained models as Notebook dependencies and automatically formats the edited code according to Python PEP 8 specifications after the user finishes writing the code or during the coding process. The resource management module is configured to support the number of CPU cores, CPU memory capacity, and GPU memory for different cluster types, and to configure the single available time for users of different cluster types.

2. The remote sensing intelligent identification system for natural resource elements according to claim 1, characterized in that: Multiple data type access methods are available, including: Access methods include user-owned data file uploads and online access to existing data storage media, including online database access and NAS online access.

3. The remote sensing intelligent identification system for natural resource elements according to claim 1, characterized in that: Managing data content includes: Data extraction and dataset construction from multiple sources, data source version management, unified management and content extraction of the accessed data.

4. The remote sensing intelligent identification system for natural resource elements according to claim 1, characterized in that: The model training subsystem is used for: The remote sensing multi-source sample set is loaded into the model optimization stage. The smooth operation of the model is ensured through mirror creation and resource scheduling. Then, the intelligent model is trained and optimized, the training results are analyzed, and the model is released.

5. The remote sensing intelligent identification system for natural resource elements according to claim 1, characterized in that: The modeling tool provides a code snippet library, resource usage analysis, variable tracking, code standardization, and version management functions.

6. The remote sensing intelligent identification system for natural resource elements according to claim 1, characterized in that: The ability to configure the single-use availability duration for different cluster types includes: Users are divided into high-priority users and low-priority users. The single available time allocated to high-priority users is longer than that allocated to low-priority users. The minimum division granularity is 1 minute.

7. The remote sensing intelligent identification system for natural resource elements according to claim 1, characterized in that: The resource management module supports configuring resource reclamation time for different cluster types, with a minimum granularity of 1 minute.

8. The remote sensing intelligent identification system for natural resource elements according to claim 7, characterized in that: If a user closes the application model, the user will still have access to the resource within the specified resource reclamation time; if the specified resource reclamation time is exceeded, resource reclamation will be triggered.

9. The remote sensing intelligent identification system for natural resource elements according to claim 1, characterized in that: The resource management module supports managing multiple GPUs as a cluster and allows users to configure the number of GPUs themselves. It also supports configuring the user workspace size and image version for different cluster types.

10. The remote sensing intelligent identification system for natural resource elements according to claim 1, characterized in that: The application inference architecture includes overall design, parallel scheduling technology, plug-in workflow technology, and algorithm integration specification design.

Images