CN112561832A

CN112561832A - Remote sensing image data storage method and system

Info

Publication number: CN112561832A
Application number: CN202011551192.7A
Authority: CN
Inventors: 张鹏; 戚文来; 朱丰琪; 张允涛; 徐杰; 王宏昌; 马玉忠; 张广庆; 王博; 孟萌
Original assignee: Shandong Provincial Institute of Land Surveying and Mapping
Current assignee: Shandong Provincial Institute of Land Surveying and Mapping
Priority date: 2020-12-24
Filing date: 2020-12-24
Publication date: 2021-03-26
Anticipated expiration: 2040-12-24
Also published as: CN112561832B

Abstract

The invention provides a remote sensing image data storage method and a system, which are used for acquiring remote sensing image data; preprocessing and texture analysis are carried out on the three-dimensional remote sensing image data, an image pyramid is generated through wavelet transformation, top images of the pyramid are transformed and matched based on message propagation, layer-by-layer matching is carried out, and a digital surface model is generated; performing radiation correction, differential correction and embedding on scanned three-dimensional remote sensing image data pixel by using a digital elevation model, and cutting according to a set image range to generate a digital orthophoto map; according to the remote sensing image data acquisition time, storing data in a set time period in a local storage area, storing data outside the set time period in an external storage area, wherein each area comprises a plurality of storage nodes, and respectively handing original remote sensing image data and processing results generated based on the original data to different storage nodes for storage. The invention improves the storage efficiency of the image data.

Description

Remote sensing image data storage method and system

Technical Field

The invention belongs to the technical field of image data storage, and particularly relates to a remote sensing image data storage method and system.

Background

The statements in this section merely provide background information related to the present disclosure and may not necessarily constitute prior art.

With the rapid development of remote sensing technology and computer information technology, remote sensing image products are applied in more and more fields, the resolution requirements of part of application fields on remote sensing image data are higher and higher, and the data volume is increased. The remote sensing data volume is expected to increase by 20% -30% every year by researchers.

Because the image data is bulky and complex, in the prior art, the image data is generally compressed or divided into different image blocks, and then the image blocks are distributed to different processing nodes for processing. However, these methods still have problems of poor processing speed and low efficiency; moreover, the processed image blocks need to be subjected to relatively complicated processing such as displacement, rotation, affine, interpolation, distortion, offset and the like, and the problems that the image blocks are lost, or cannot be spliced, and need to be repeatedly processed easily occur. The processing speed of the remote sensing image data is seriously influenced.

Disclosure of Invention

The invention provides a remote sensing image data storage method and system for solving the problems, and the remote sensing image data storage method and system can improve the processing rate while finishing image data processing.

According to some embodiments, the invention adopts the following technical scheme:

a remote sensing image data storage method comprises the following steps:

acquiring remote sensing image data;

preprocessing and texture analysis are carried out on the three-dimensional remote sensing image data, an image pyramid is generated through wavelet transformation, top images of the pyramid are transformed and matched based on message propagation, layer-by-layer matching is carried out, and a digital surface model is generated;

performing radiation correction, differential correction and embedding on scanned three-dimensional remote sensing image data pixel by using a digital elevation model, and cutting according to a set image range to generate a digital orthophoto map;

according to the remote sensing image data acquisition time, storing data in a set time period in a local storage area, storing data outside the set time period in an external storage area, wherein each area comprises a plurality of storage nodes, and respectively handing original remote sensing image data and processing results generated based on the original data to different storage nodes for storage.

As an optional implementation mode, terrain mapping auxiliary adjustment is carried out, the resolution ratio of a three-dimensional image is reduced, connection points of each fast-view image pair are extracted, the fast-view resolution ratio is reduced, conventional connection points of an original image are extracted by combining the terrain mapping elevation constraint of a space plane radar, free network evaluation of original influences is carried out, mountain top feature points and flat area feature points are extracted according to a certain distance, the two types of feature points are projected onto the original influences respectively, and dense connection points are extracted.

As an alternative embodiment, the method further comprises a coverage map product making module, which performs concurrent map tile drawing based on the time, the data type, the area information and the cloud cover information of the stereo image data input, and generates a coverage range product.

As an alternative embodiment, different remote sensing image data are transmitted to different processing nodes, and data processing is performed among the processing nodes in parallel.

As a further limited implementation manner, before the remote sensing image data processing process is performed, the maximum parallel number and the computing capacity of each processing node are determined to determine, and according to the maximum parallel number N, N processing nodes with the highest computing capacity rank are determined to perform data processing tasks.

As a further limited implementation, the remote sensing image data processing tasks are grouped, and the tasks and the processing nodes are matched and distributed by using an optimal algorithm according to the computing power ranking of each processing node.

As an alternative embodiment, before data storage, data integrity is checked, and if the data integrity meets the requirement, the data is stored.

A remote sensing image data storage system comprising:

the data receiving module is configured to acquire remote sensing image data;

the digital surface model generation module is configured to preprocess and analyze texture of the stereo remote sensing image data, generate an image pyramid through wavelet transformation, transform a top image of the pyramid and perform matching based on message propagation, perform layer-by-layer matching and generate a digital surface model;

the digital orthophoto map generation module is configured to utilize a digital elevation model to carry out radiation correction, differential correction and mosaic on scanned and processed three-dimensional remote sensing image data pixel by pixel, and cut according to a set image range to generate a digital orthophoto map;

the data sorting module is configured to store data in a set time period in a local storage area according to the remote sensing image data acquisition time and store data outside the set time period in an external storage area;

and the data storage module is configured to respectively deliver the original remote sensing image data and the processing result generated based on the original data to different storage nodes for storage.

A computer readable storage medium having stored therein a plurality of instructions adapted to be loaded by a processor of a terminal device and to perform the steps of a method of storing remote sensing image data.

A terminal device comprising a processor and a computer readable storage medium, the processor being configured to implement instructions; the computer readable storage medium is used for storing a plurality of instructions, and the instructions are suitable for being loaded by a processor and executing the steps of the remote sensing image data storage method.

Compared with the prior art, the invention has the beneficial effects that:

the invention

In order to make the aforementioned and other objects, features and advantages of the present invention comprehensible, preferred embodiments accompanied with figures are described in detail below.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, are included to provide a further understanding of the invention, and are incorporated in and constitute a part of this specification, illustrate exemplary embodiments of the invention and together with the description serve to explain the invention and not to limit the invention.

FIG. 1 is a schematic view of data storage management;

FIG. 2 is a schematic view of a data management process;

FIG. 3 is a schematic diagram illustrating a data flow of data external to the system;

FIG. 4 is a schematic diagram illustrating a data flow of data in the system;

FIG. 5 is a diagram of a parallel scheduling framework;

FIG. 6 is a diagram of a data storage system physical architecture;

FIG. 7 is a diagram of a network relationship structure;

FIG. 8 is a schematic diagram of a data management process based on modeling techniques;

FIG. 9 is a schematic diagram of a flow of parallel data warehousing based on autonomous tasks;

FIG. 10 is a schematic diagram of an implementation of fast retrieval in an image data storage system;

FIG. 11(a) is a schematic diagram illustrating a comparison of matching effective elevation points with an SGM algorithm in accordance with the present system;

FIG. 11(b) is a schematic diagram comparing the topographic aspects of DSMs extracted by the present system with DSMs extracted by other commercial software;

FIG. 12 is a specific process for global optimization matching policy automatic generation of DSMs;

FIG. 13 is a SRTM assisted adjustment procedure;

FIG. 14 is a DSM rendering effect diagram;

FIG. 15 is a schematic view of an image acquired according to the present embodiment;

FIG. 16 shows the matching result of the control points of the image data according to the prior art;

fig. 17 shows the result of matching control points for image data in the present system.

The specific implementation mode is as follows:

the invention is further described with reference to the following figures and examples.

It is to be understood that the following detailed description is exemplary and is intended to provide further explanation of the invention as claimed. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of exemplary embodiments according to the invention. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, and it should be understood that when the terms "comprises" and/or "comprising" are used in this specification, they specify the presence of stated features, steps, operations, devices, components, and/or combinations thereof, unless the context clearly indicates otherwise.

A remote sensing image data storage system comprising:

the data receiving module is configured to acquire remote sensing image data;

The system is used for effectively managing mass multi-source heterogeneous remote sensing image data, improving the processing and management efficiency of the remote sensing image data, improving the sharing and service level of the remote sensing image data and providing data guarantee for important projects such as basic geographic information resource construction, geographic province monitoring, emergency surveying and mapping and the like.

The following detailed description is to perform storage and management of data. The data put in storage includes:

(1) original data of aerial remote sensing image

The system mainly comprises ADS series, UC series, SWDC-4, SWDC-5, CS10000, DMCIII and the like, and the system also needs to support an extended function to meet the data type of a follow-up novel aerial photography instrument.

Frame type original aerial photo

The frame data is divided into five types according to the aerial photography instrument: SWDC-4, SWDC-5, UC series, CS10000 and DMCIII. The data format is that a suffix corresponding to a plurality of navigation films is txt metadata file, the metadata file records the corresponding spatial position information of each navigation film according to the line, which mainly comprises: GPS Time(s), observing (deg), Northing (deg), Ell Ht (meters), omega (deg), phi (deg), Kap (deg), Lat (deg), Lon (deg), and other attribute information, wherein the spatial position information is a point for positioning the center position of the navigation sheet. The flight data is in tiff format.

ADS push-broom raw data

The method comprises the steps of L0 level, L1 level, L2 level and the like, wherein L0 level data are original data and need to be decompressed through special software to generate L1 level data, the data quantity of L1 level data and L2 level data is large, only L0 level data need to be backed up and managed, the provided shp is a route joint list of a certain frame, and spatial position information is multi-line.

(2) Raw data of space remote sensing image

The data organization structure includes: tif image files, xml metadata files, rpc files, and other auxiliary files, the geonarge folder holds the spatial extent of the image files.

(3) The achievement data comprises: DOM, DEM, DLG, and point cloud data.

The DOM data includes: the satellite image scene completion result and the DOM result are 1: 500-1: 25000.

The achievement data organization structure comprises: tif video file, tfw file, xml file, ovr file, xls metadata file, and a shp space range combination table file.

DEM result data, the data organizational structure includes: and organizing data according to a 1:10000 map sheet number specified in the national basic scale topographic map sheet and number (GB/T13989 and 2012) as a folder name, and storing the metadata in an excel file form.

DLG (DLG) result data, the data organization structure comprises: the data are organized according to the number of a map sheet specified by the national standard of 'national basic scale topographic map framing and numbering' formulated in 2012 as the name of a folder, the metadata and the data entity are stored separately, wherein the metadata are stored in the form of an excel file, and the organization of the metadata is also organized according to the number of the map sheet as the name of the folder.

As shown in fig. 1, database storage is combined with file storage.

And various types of databases and file storage systems are comprehensively utilized to efficiently and safely store various types of data. The method adopts a mode of combining database storage and file storage, wherein the database storage adopts a relational database (a spatial database), and the file storage adopts shared file storage.

1) In view of data volume, the remote sensing image data and the product result data volume are large, and the entity data are stored by adopting files.

2) In view of data structure, the metadata relates to spatial data and attribute data, and is uniformly stored by adopting a relational database.

3) In view of application requirements, the spatial layer data is mainly used for browsing, extracting and other services, a spatial database is used for storage, and image files and other files requiring high IO operation are stored.

4) From the perspective of data security migration and cost, for data with large access amount, recently accessed data and latest results stored in a filing disk array in an intranet, a historical data entity is cleaned locally as required, and only the meta information of the data is reserved.

The method takes aviation remote sensing data with different sources and different formats, various space satellite remote sensing image data, produced product result data and the like which are collected by various production departments as main data sources, provides data service capabilities for production, migration and distribution on the basis of a Shandong province remote sensing image comprehensive database, comprises data archiving service, data query service and data extraction service, realizes unified data access service, supports data access of files and database formats, shields the difference of physical storage of various data, and enables the physical storage of the data to be transparent, thereby facilitating the use of each user.

And performing rapid warehousing centralized management on mass remote sensing image data and product data acquired and generated every day, and achieving rapid warehousing of the mass data by adopting parallel filing based on tasks. As shown in fig. 9, the data parallel warehousing technical process based on the autonomous task obtains task information, writes and reads, obtains information of a data file to be archived, writes and reads, archives a task information metadata file, file list information, and a fast attempt file, writes and reads, and distributes data file information task distribution information, which is used for distributing data to be copied to a corresponding server in a load balancing manner, writes and reads, and obtains copy task result information.

The important goal of the image data storage system is to realize the comprehensive management of the aerospace remote sensing image data and the product results, so that the data circulation of various data in the system is necessary to be clear. The overall data management process is shown in fig. 2, and specifically includes:

(1) for aerospace remote sensing data and product result data, a data integration module needs to perform standardized data integration work, and then data classification is performed. The data integration module comprises the work flows of data integrity check, usability check, data type identification and the like, wherein the integrity check is to check the image data file, check whether a related reference file or an attribute file is lacked, and remove the missing data of the file to ensure the integrity of the data; the usability check mainly checks whether the data can be decompressed or not, whether the file is damaged or not, and eliminates the problem data found by the check to ensure the usability of the data.

(2) And extracting the metadata and the auxiliary information according to the belonged data category. According to different data categories, different database tables are designed to ensure that the metadata and the auxiliary information data of each type of data are completely and effectively extracted and input.

(3) And carrying out data format and integrity check on metadata information, data composition, auxiliary information and the like of the data, filing and warehousing the data which is in accordance with the check, and marking the data with missing information so as to check the data in the later period.

(4) And the data is subjected to filing directory data creation, metadata registration and data volume injection according to classification setting, so that the data filing work is completed.

(5) And performing online storage on the data according to a storage strategy. And the new warehousing data are stored online, and historical data or other data with low use frequency required by a user are stored offline.

(6) And each service system inquires data through the inquiry retrieval interface and extracts the required data. Different business users have different authorities, and after a common user submits a data requirement, a super manager user verifies an order and can extract data through a rear user.

(7) And the data synchronization work of the internal data and the data publishing server is realized through data synchronization.

The external data flow of the system mainly describes how data enters the image data storage system and how data is exported to a user through the system. The external data circulation of the system is mainly divided into: data archiving, data querying, data extraction, and data synchronization, the specific system external data flow is as shown in fig. 3. The data circulation in the system mainly describes how data are circulated among the functional modules after entering the image data storage system. The image data storage system classifies, inspects, files and stores the data in a warehouse according to the data filing task, is responsible for maintaining and managing data resources and systems, provides inquiry and retrieval services of various filed data, provides statistics and analysis of related data information, provides reference for a system manager maintenance system, and completes the function of providing data synchronization for the internet and an e-government intranet image data storage system. The specific system internal data flow is shown in fig. 4.

Of course, in order to achieve better management of data, specifications of operations such as selection of various data storage spaces, naming and the like can be set. These are all flexibly set by the skilled person according to circumstances and will not be described in detail here.

With the prediction that the total amount of management data reaches the PB level, for services such as data archiving and statistics, the traditional single-machine processing mode cannot meet the requirements of timely data archiving and efficient statistics, and the parallel architecture design is required for system construction. The system construction is realized based on an autonomous parallel computing framework, and by adopting the framework, the parallel processing capacity and performance of the system are mainly limited by the number of parallel task processing nodes, the network throughput and the performance of disk array IO.

The parallel scheduling framework supports parallel resource scheduling, parallel task allocation and parallel computation execution, and a data archiving subsystem, a data retrieval subsystem and a statistical analysis subsystem are developed based on the framework to realize multi-machine and multi-process parallel execution of archiving, retrieval and statistical services. During execution, the data archiving, retrieving and counting tasks are decomposed into a plurality of subtasks, the subtasks are distributed to corresponding execution nodes for execution, and after the execution of each node is completed, the execution results are collected and returned. By parallel distribution and execution of serial processes, the data archiving efficiency is improved, the data query speed is increased, and the data statistical time is shortened.

As shown in fig. 5, in the present system, parallel scheduling is employed. And the registration and management of the parallel computing node resources are supported. The maximum parallel number and the computing capacity of the nodes can be reasonably configured so as to fully utilize computing resources. And realizing a task parallel scheduling algorithm. And grouping the task items by adopting an optimal algorithm, and assigning the task items to the nodes for execution according to the resource occupation condition and the performance of each computing node. And task scheduling service is provided, a high availability mechanism is realized, and the single-point problem is avoided. And based on the idea of plug-ins and workflows, scheduling and distributing the parallel tasks.

As shown in fig. 6, the system architecture supported by the server may include, in this embodiment: calculating a layer: the system comprises a parallel computing server, a task allocation scheduling server and an application server, and is used for background parallel scheduling and computing of services such as parallel filing, parallel data retrieval, parallel data cleaning, parallel statistics and the like.

A storage layer: the system comprises a database server, a database server and a database server, wherein the database server is used for storing contents such as space data, attribute data, quick view data, document data and the like so as to meet high transaction request response and concurrent access requests under the condition of large data volume; the storage device is used for storing the aerospace remote sensing images and the product result data, the mobile hard disk is used for data migration and backup, the online storage is shared storage, and the offline storage is the mobile hard disk.

An application layer: the system can be a workstation, a desktop, a notebook or a remote desktop, and is used for business operations of data management, data browsing and querying of a client and the like.

Network layer: it is necessary to connect the computation layer, the storage layer and the application layer in series to provide an efficient data transmission channel therebetween.

Meanwhile, in order to support the user to access and refer to the relevant data, as shown in fig. 7, the system provides intranet and extranet services. As shown in fig. 7, there are 3 sets of networks, which are respectively deployed in 3 sets of networks, namely, a business intranet, an e-government intranet and the Internet, where the business intranet is a private network in a unit established by using the Internet technology, and forms a unified and convenient information exchange platform by using TCP/IP protocol as a basis and Web as a core application; the electronic government affair intranet is a confidential party government office business network, is physically isolated from the Internet, has the smallest coverage range on the premise of meeting working requirements, and is interconnected with the national electronic government affair intranet. The information is not isolated from an external network through a traditional firewall, but is exchanged only in a data ferry mode through a gateway (general protocols such as HTTP, FTP, SMTP and the like of the gateway are all closed or do not provide support of the protocols) so as to realize the connection of public service and internal service flow; the internet also becomes an external network, is connected with all external open information and resources, has large information storage capacity, rapid and efficient information transmission and wide user-oriented range, and is not limited by regions.

The data storage system adopts the modularized design, the whole set of system covers the functions of all internal networks, the Internet and government affairs networks, different networks are distinguished and different functional modules are loaded through configuration files, and finally the adaptation of different networks is realized:

a service intranet system: the user object is an internal user, modules such as data archiving and warehousing, space retrieval, data query and statistics and the like can be automatically loaded according to the authority of a login user, the daily work requirements of the internal user of a unit are met, and all functions of the system are covered. All original data images and product result images need to be managed by the system, all data warehousing needs to be achieved through an intranet system, data and data cataloging used by the internet system and the electronic government intranet system need to be achieved through an intranet system data synchronization function, and the intranet system has a synchronous data export function.

An internet system: the user object is a public user and provides a query and download function of a small amount of public edition data, the data source of the Internet system is realized by performing a data synchronization function with the intranet system, and the Internet system has a synchronous data import function.

E-government affair intranet system: the user object is a government user and provides management and query functions of partial original data and partial secret-related data, a data source of the electronic government intranet system is realized by performing a data synchronization function with the intranet system, and the electronic government intranet system has a synchronous data importing function.

Between business intranet system, internet system and the electronic government affair intranet system, because the security level is different, 3 sets of networks physical isolation each other, the data synchronization function can't realize cross-system automation, needs the manual work to copy synchronous data file to internet system and electronic government affair intranet system.

Certainly, in the communication process, some data confidentiality and security safeguards, such as encryption, setting of permissions, account management, data consistency check, and the like, are also required, and these may all adopt the prior art, and are not described herein again.

Because data is multi-source, multi-standard and multi-format, and needs to be stored in a unified and standardized manner, the system adopts a data management mode based on a modeling technology, as shown in fig. 8, the data modeling process can be divided into three levels, namely, cataloguing planning, data modeling and physical storage. The resource cataloging system is divided into three levels, namely a cataloging node, an abstract node and a data node. The cataloging nodes and the data nodes are only in one stage, and the abstract nodes are in multiple stages. The cataloging node is a root node of the cataloging of the database. The abstract nodes are logical classification nodes of data, and the data nodes are data nodes of each data product. For each data node, the image data storage system is managed in a modeling mode, and each data product is not directly defined.

The data modeling is a process for performing modeling definition on a type of data, wherein the data comprises two parts of description information and an entity, the description information comprises data metadata and spatialization information, the entity comprises a video entity and a video block, and the data modeling is an instantiation process of the data metadata, the spatialization information, the data type and the storage position, wherein the spatialization information and the storage position are extensions of the data and are not necessary conditions.

The image data storage system manages data in a manner of combining catalogues and metadata, manages spatial data in a spatial data set manner, constructs resource libraries conforming to respective characteristics on the basis of a set of unified database management basic platform, and registers the resource catalogues of the two resource libraries in the resource catalog storage system to support the publishing and sharing of resource information.

The system comprises an engineering building subsystem, a control point, a check point acquisition subsystem, a DSM editing subsystem, a control point, a connection point extraction and area network adjustment subsystem, a DSM production subsystem and a DOM production subsystem.

The engineering establishment subsystem is used for automatically retrieving the SC-level satellite images according to a set data directory, and matching and grouping the retrieved images according to indexes such as acquisition time, position, overlapping degree and the like, namely stereo image pairs, stereo image pairs and multi-spectrum images, and panchromatic images and same-view angle multi-spectrum images; checking the integrity of the images and RPC, displaying all the images as a block diagram, and storing the images as a shp file.

And the control point and check point acquisition subsystem is used for calling an SC-level image according to the ground coordinates manually input, calling a control base map under the condition that the control base map is set, acquiring the same-name point of the overlapped image by matching after pricking a point on one image, and calling the image at the selected position.

And the DSM editing subsystem is used for simultaneously opening the DSM to be edited and a replacement DSM (such as SRTM), checking the vulnerability area of the DSM to be edited to realize vulnerability repair, simultaneously opening the DSM to be edited and the DOM, checking the water body area on the DOM, realizing the neat cutting of the water body boundary on the DSM, and realizing the deletion of a building or the leveling of the water surface of the selected area.

And the control point and connection point extraction and adjustment subsystem is used for automatically matching control points (stereo and non-stereo) with a reference base map and a reference DEM in a cluster environment, acquiring SRTM control points for processing stereo images in the cluster environment (stereo), automatically acquiring connection points in the cluster environment (stereo and non-stereo) and adjusting regional nets in the cluster environment (stereo and non-stereo).

And the DSM production subsystem is used for automatic production of vision-divided DSMs in a cluster environment, automatic splicing of the DSMs in the cluster environment, amplitude division according to the degree, division of 1 pixel by 1 pixel and 2 pixel by DSM matching distance, adjustable resolution of the DSM and support of two-line and three-line array stereo image matching.

The data storage system supports a data query function and mainly comprises three modules, namely a data query condition, a query result and a coverage rate result, wherein the query condition supports a space range query mode such as manual drawing, administrative region query, map sheet query, text import and shp import and an attribute query mode such as time, resolution, cloud cover, a sensor and a map sheet number. In order to conveniently inquire different requirements of users, the system supports two display modes of a vector base map and an image base map, and the image base map and the vector base map are additionally arranged, so that the users can conveniently position and inquire.

And the DOM production subsystem is used for correcting 8-bit and 16-bit orthoimages in a cluster environment, fusing 8-bit and 16-bit panchromatic and multispectral orthoimages in the cluster environment and homogenizing the whole 8-bit RGB orthoimages in the cluster environment.

In the data processing process, the more critical technical scheme mainly comprises the following steps:

(1) pixel-by-pixel dense matching plus global optimization

The global optimization matching algorithm of the stereo images aims at global energy minimization, and overall matching is achieved through global optimization based on an energy equation. The energy equation generally includes a data term and a smoothing term. The data item refers to the similarity of corresponding points of a certain parallax corresponding to the stereo image, and can be represented by one or a combination of more of gray scale distance, mutual information, correlation coefficient, CENSUS distance and the like; the smoothing item refers to a penalty parameter applied when the parallax of two adjacent pixels changes, and the smoothing item and the corresponding global optimization method are key steps different from a local matching algorithm. The essence of the global optimization matching algorithm is that the matching degree and the parallax smoothness of the points are balanced in the whole matching area, and a result superior to that of the local matching algorithm can be obtained under most conditions. However, the global optimization matching algorithm often has the disadvantages of large calculation amount, large memory consumption and the like, and global optimization may form an aggregation or extension effect of parallax, and the direction of aggregation or extension is not necessarily performed along an ideal direction, so that an erroneous matching result may be caused instead.

Common global optimization matching algorithms include Belief Propagation (BP), Graph Cut (GC), Total Variation (TV) and Generalized Total Variation (TGV). The traditional global optimization algorithm implies a kind of 'front surface parallel'

The (Frontal-Parallel) effect, i.e. more "like" or "good" matching planes with the same parallax, is that in the area of parallax variation, especially in the area of weak texture with parallax variation, stepped parallax caused by the "Frontal Parallel" effect occurs instead of smooth parallax planes. The improved algorithm appearing in recent years initializes the parallax of each point into a random plane by an over-parameterization method, and obtains a good matching effect on an inclined plane by the combined use of local optimization and global optimization. But the amount of calculation is significantly increased compared to the conventional practice.

The global optimization matching method aims at global energy minimization, and generally adopts an energy equation of the following form:

where e (D) represents the global energy level, D represents the "disparity map" of the entire video, q is the neighboring pixel of pixel p, and Np represents the set of neighboring pixels of pixel p. Dp and Dq represent the disparity of the two pixels. C represents a data item of the pixel p when the disparity is Dp, representing the degree of similarity of the left and right patch pixels, and S represents a smoothing item, representing a penalty not applied at the same time by Dp and Dq.

In this embodiment, a message propagation method is used as a basic strategy for global optimization, a pyramid matching method is adopted, and the disparity obtained by matching the pyramid image of the previous layer is used as an initial disparity value of the current layer and basic data for estimating the local disparity gradient. Assume that the disparity map D0 of the previous layer has been obtained. And obtaining the initial parallax value D of the current layer through linear interpolation.

As can be seen from fig. 11(a) and (b), the comparison between the system matching effective elevation point and the SGM algorithm, and the comparison between the DSM extracted by the present system and the DSM extracted by other methods in the aspect of terrain are obvious, and the optimization method of the present storage system has better effect.

The control points and the connection points are automatically matched:

besides supporting the traditional control point acquisition mode, the high-density control points can be automatically matched from the reference base map, and the high-density connection points can be automatically matched. As shown in fig. 15-17, enough control points and connection points can be matched even where the cloud cover is very large.

Fig. 16 shows that 18994 control points are matched on the google base map, and the same area in the present application can be matched with 31519 connection points. Is obviously superior to the prior art.

In addition, in the full-automatic ortho-image production process, the conventional adjustment calculation is generally called that the stereo image condition is a strong intersection condition, and intersection of a plurality of images imaged without lateral swing is a weak intersection condition. In order to acquire images of a specific area in actual work of a satellite, observation angles are random, so that intersection angles are different in size. The system can jointly adjust the images of different sensors and any intersection angle. In the aspect of precision, taking nationwide land utilization survey data of a certain province as an example, the edge connecting precision is generally superior to 0.5 pixel and maximally does not exceed 2 pixels. In the experiment, a resource three-Dimensional (DOM) is taken as a reference base map, the registration precision of the DOM and the base map is generally better than 1 pixel, and the maximum registration precision of the DOM and the base map is not more than 3 pixels.

The uncontrolled accuracy is improved by adding generalized control such as SRTM, GLAS and the like to the model concatenation/free net adjustment, and DSM is automatically generated by adopting a global optimization matching strategy based on message propagation, as shown in fig. 12, including:

1) stereo image and RPC import: and distinguishing the front-view image and the back-view image and the corresponding RPC according to the file name, wherein the RPC is subjected to adjustment correction.

2) Image preprocessing and texture analysis: one key step in image preprocessing is to determine the minimum valid value of the image and uniformly set the DN values below this value to the minimum valid value, so that smooth transition of these low value regions can be achieved, otherwise some false texture effects match will occur. The satellite image is in the area with low gray level such as the mountain backlight area with large fluctuation, the area with large area and no texture or false texture is easy to appear, the areas are detected by calculation based on local window variance, the matching cost of the areas is uniformly set to be a fixed value, and smooth transition can be realized through the following matching cost propagation step.

3) Generating an image pyramid: and generating an image pyramid through wavelet transformation, and storing the image pyramid in a diagonal region of the data volume. The key here is whether the pyramid layer number and the set top search range correspond to the height difference in the region.

4) CENSUS-based top-level matching: and carrying out CENSUS transformation on the top layer image, carrying out matching based on message propagation, detecting a suspicious region, if the suspicious region caused by large-area non-texture loss occurs and the elevation of the region is close to the maximum or minimum elevation, considering that the search range is set to be too small, enlarging the search range, matching the error region again, if the error region is not represented as a large-area region any more, enlarging the search range if the region is reduced and the error region with a larger area is bundled until the suspicious region caused by the search range does not exist (the matching results of the previous and subsequent times are not obviously different). The result of the matching is the elevation of each image point of the front view image.

5) Layer-by-layer matching, suspicious region detection and refinement matching based on CENSUS and Mutual Information (Mutual Information, hereinafter abbreviated as MI): and (3) performing CENSUS transformation on each layer of image below the top layer by adopting block matching and a certain overlapping area between two adjacent images, calculating a front-view and rear-view correction image by utilizing the elevation acquired by each front-view image point of the previous layer, and calculating an MI lookup table. The initial elevation of each point of the current layer is obtained from the matched elevation of the previous layer, the search range of each point is determined by the elevation in the neighborhood of the point (the size of the neighborhood depends on the local height difference), and a certain margin is left. And performing message propagation optimization matching based on MI and CENSUS, and determining suspicious regions according to various methods such as the setting of smooth parameters and the difference between the multipath independent matching result and the aggregated matching result. Classifying the mismatching areas according to texture features of the front-view image and the front-view and back-view corrected images in the mismatching areas, the surrounding elevations of the mismatching areas, the matching measurement of the mismatching areas and the like, carrying out refined matching on different types of suspicious areas, and calculating images without using the suspicious areas by using the MI lookup table during the refined matching.

6) Generating a DSM: and after the bottom layer matching is finished, calculating ground points according to the main image points, the corresponding elevations of the image points and the RPC, and generating the DSM by a distance inverse weighting method according to the set DSM grid interval.

7) DSM generation and splicing: and generating the DSM by a distance inverse weighting method according to the set DSM grid spacing, and splicing the single model DSM into the whole DSM. And recording the single model serial number corresponding to each grid point in the splicing process for the DOM extraction link.

When the suspicious regions are matched, the following method can be selected:

1) generating an epipolar stereoscopic image pair of the stereoscopic images, wherein the epipolar stereoscopic image pair comprises a left epipolar image and a right epipolar image;

and generating a nuclear line stereo image pair of the stereo image by adopting a projection trajectory method. Generating a epipolar stereoscopic image pair of stereoscopic images by a projection trajectory method is well known in the art and will not be described herein.

2) Reading the epipolar line stereo image pair, and respectively performing wavelet transformation on a left epipolar line image and a right epipolar line image of the epipolar line stereo image pair to establish respective image pyramids of the left epipolar line image and the right epipolar line image, wherein the image pyramids comprise multilayer pyramid images, the originally read left epipolar line image and right epipolar line image are first-layer pyramid images, and the second-layer pyramid images and the pyramid images above the second layer obtained after wavelet transformation both comprise an image area and three texture areas;

the image pyramid refers to a reduced image obtained by down-sampling an image, that is, the image pyramid is a plurality of wavelet transform images (and texture images) with the length and width decreasing by one half in sequence.

2) The process of wavelet transform in (1) specifically includes: taking the ith pyramid image as a prokaryotic line image, and performing wavelet transformation on the ith pyramid image to obtain an i +1 th epipolar line image, wherein i is an integer greater than 1, the i +1 th epipolar line image comprises images of four regions, namely an upper left region, an upper right region, a lower left region and a lower right region, the image of the upper left region is an image region and comprises a reduced image with the length and width being one half of the length and width of the original epipolar line image, and the three regions, namely the upper right region, the lower left region and the lower right region, are texture regions and comprise texture information of the prokaryotic line image. The three texture regions after wavelet transform have the same length and width as the image region.

And after a second-layer pyramid image of the left and right epipolar line images is obtained, taking the second-layer pyramid image as a new original image, and performing wavelet transformation on the new original image to obtain a third-layer pyramid image (comprising an image area and three texture areas). And the same can be done to the fourth and fifth pyramid images. According to a preferred embodiment of the present invention, the number of layers of the pyramid image is 5.

The gray values of an image region and three texture regions after wavelet transformation are the result of weighted accumulation summation according to filter coefficients in a 4 x 4 window. Specifically, the gray scale G0(j, k) of the k-th row and j-column, the gray scale G1(W/2+ j, k) of the k-th row and W/2+ j-column, the gray scale G2(j, H/2+ k) of the left-lower texture region row H/2+ k, the gray scale G3(W/2+ j, H/2+ k) of the right-lower texture region row H/2+ k and W/2+ j-column of the video region after wavelet transform are calculated according to the following formula.

3) Determining suspicious regions in the left epipolar line image and the right epipolar line image layer by layer for the left epipolar line image and the right epipolar line image of each layer of image pyramid below the top layer;

and if the pyramid layer number is n, the top layer is the (n-1) th layer, the (n-2) th layer, the (n-3) th layer and the (0) th layer are sequentially arranged below the pyramid layer, and the (0) th layer also becomes the bottom layer, namely the initial left epipolar line image layer and the initial right epipolar line image layer.

Setting the parallax search range of each image point of the left epipolar line image and the right epipolar line image as h, and setting the search starting parallax as-h/2. Generally, the parallax of the left and right epipolar line images at the top layer of the pyramid is obtained by an SGM matching method. The top layer does not perform suspicious region detection.

4) Performing global optimization on the left epipolar image-right epipolar image matching cost and the right epipolar image-left epipolar image matching cost to obtain multiple pairs of disparity maps based on different path combinations and corresponding smooth models;

only one disparity of the plurality of candidate disparities within the corresponding search range for each pixel P is correct, which requires determining the disparity for each pixel by global optimization at step 3.4.

After the left-right matching cost and the right-left matching cost are calculated, the parallax of the image point can be obtained through two methods (a local algorithm and a global optimization algorithm), the global optimization algorithm is stronger in robustness than the local algorithm, and the cloud area and the non-cloud area are more obviously compared in matching stability, so that the left epipolar line image-right epipolar line image matching cost and the right epipolar line image-left epipolar line image matching cost are subjected to global optimization.

And performing global optimization on the left epipolar image-right epipolar image matching cost and the right epipolar image-left epipolar image matching cost by adopting an optimization strategy based on message propagation. The 16 directional center points of the message propagation represent the pixels that receive the message, and the surrounding 16 pixels represent the 16 pixels that propagate to the nearest upward. For each image pair matching direction (the two directions of the left epipolar image-right epipolar image direction and the right epipolar image-left epipolar image direction), four message propagation path combinations are adopted, which are respectively:

path combination 1: 0. 12, 1, 13;

path combination 2: 4. 14, 5, 15;

path combination 3: 2. 9, 3 and 8;

path combination 4: 6. 11, 7 and 10.

A two-parameter model is adopted for the path combination 1 and the path combination 2, but the parameters P1 and P2 are different in size, and a Huber model is adopted for the path combination 3 and the path combination 4, but the specific parameter a is different in size.

After global optimization, the matching cost L after each image point optimization is obtained, and the form of L is similar to that of the global data item C and is an array with the length equal to the image point searching range Dmax. The minimum value in the array, Dmin (array subscript 0, 1, …, Dmax-1), is added to the minimum disparity Ds of the pixel, i.e., the disparity Dend of the pixel. The pixel column coordinates col and the parallax Dend are added to obtain the pixel column coordinates col' corresponding to the right epipolar line image, and the pixel row coordinates row of the left and right epipolar line images are the same. And performing front intersection by the left and right quasi-epipolar image point coordinates and the left and right quasi-epipolar images through RPC respectively to obtain corresponding ground coordinates (longitude L, latitude B and elevation H).

5) And based on 4) multiple pairs of disparity maps obtained based on different path combinations and corresponding smooth models, preliminarily determining suspicious regions in the images.

The SRTM-assisted adjustment procedure employed by the present system is shown in fig. 13, and includes:

1) fast view based join point extraction

The resolution of the acquired three-dimensional image (resource No. three) is reduced by means of nxn pixel gray level averaging, the resolution of the resource No. three 01 star front-view image is reduced by 8 times, the resolution of the front-view image and the rear-view image is reduced by 5 times, and the resolution of the 02 star three-view image is reduced by 8 times. The corresponding RPC parameters LINE _ OFF, LINE _ SCALE, SAMP _ OFF and SAMP _ SCALE conversion method comprises the following steps:

LINE_SCALE＝LINE_SCALE/n，SAMP_SCALE＝SAMP_SCALE/n

LINE_OFF＝(LINE_OFF-n/2+0.5)/n，SAMP_OFF＝(SAMP_OFF-n/2+0.5)/n

firstly, the homonymous points of each fast-view image pair (2 or 3 views) are extracted through SIFT matching, and homonymous image points are also extracted between the front-view images of any two image pairs. And determining point pairs of both image pairs by a coordinate comparison method, and combining the point pairs into a multi-view homonymy point.

Performing 'rendezvous ahead of the same-name point-obtaining the image point of each image, the ground point pair-RPC correction of each image' circularly, correcting only the molecular constant item of the RPC for the first few iterations (such as 5 times), and correcting the molecular constant item and the first order item corresponding to the normalization factor and dimension for the later iterations.

And calculating the maximum offset of the correlation between the object elevation of the same name point and the corresponding elevation of the SRTM by using an equal step retrieval method. The coordinates of the object with the same name point are added with the offset, and the RPC of all images is corrected.

2) Fast-view RPC conversion to original resolution RPC combined with SRTM elevation constraint extraction of original image conventional connection point

And replacing the molecular constant item of the rational polynomial parameter of the original image RPC and the primary item corresponding to longitude and latitude with the corresponding parameter of the corrected fast-view RPC. The relative accuracy of the RPC parameters is mainly limited by two factors of the resolution of quick vision and the accuracy of quick-try connection points, the accuracy of the predicted initial position of the same-name point of the original image is limited by the two factors, and the deviation of the local elevation of the SRTM and the corresponding elevation of the same-name point is obviously related to the terrain. For the same-track stereo imaging, the image of the terrain is mainly reflected in the row direction of the image. Therefore, the search range in the row direction is set to a larger value, the column direction search range is set to 5, the row direction search range is set to 9, and the matching window is set to 15 × 15. Firstly, Harris characteristic points of a main image are extracted, matching points with pixel-level precision are obtained through a correlation coefficient method, then sub-pixel positions are obtained through least square matching, and least square matching only considers deformation of the image in the row direction. And preliminarily eliminating homonymous points which do not meet the conditions through correlation coefficient threshold and reverse matching.

After the extraction of the conventional connection points is finished, free net adjustment is carried out by adopting a method similar to the step 1, except that after each iteration, RPC correction of each scene of image is finished, and the maximum points of which the scene of image residual difference is greater than a certain threshold a are recorded. After sorting all the large residual points, the largest part of the points (e.g. 20%) is deleted. The threshold a will gradually decrease as the number of iterations increases.

All RPCs were reduced (quick look at RPC transform results) and re-iterated to be stable (iterated a certain number of times).

3) SRTM feature point extraction

The mountain top feature points are extracted by adopting a local difference detection operator, flat feature points are detected according to a certain step distance, the SRTM row number corresponding to the current coordinate is assumed to be (r, c), points with the difference between the elevation and the elevation of the (r, c) point being larger than a certain threshold are searched towards 8 neighborhood directions by taking the (r, c) as the center, and if no point with the difference of the elevation being larger than the certain threshold is found in 8 directions in a certain search range, the points are the flat feature points.

4) Dense connection point acquisition for SRTM feature points

The two types of feature points are projected on an original image to obtain a main image and a slave image with a certain size, the SGM object space method is used for carrying out pixel-by-pixel dense matching to obtain homonymy image points, and the homonymy image points are thinned according to 45 m point (about half of SRTM resolution).

Unlike mountain top connection points, flat dense connection points first filter out corresponding points such as buildings that may exist by morphological filtering, because the SRTM does not contain this information.

5) Area network adjustment with additional SRTM constraint

And performing conventional connection point front intersection and each scene image RPC correction alternately, and calculating the ground point cluster corresponding to the dense connection point by using the current RPC. And solving the maximum correlation point with the vertex of the SRTM mountain to obtain the horizontal offset and the vertical offset. The flat area dense connection points are vertically offset only from the corresponding SRTM. The conventional connection points correspond to the ground points with horizontal and vertical offsets added according to an inverse distance weighting method.

Meanwhile, the system adopts a method of digital surface model post-processing and information extraction, and automatically acquires the urban building area through DSM. Firstly, non-ground points are identified through a method of space hierarchical decomposition and information statistics. After non-ground points are removed, a peripheral elevation is fitted to obtain a DTM (digital terrestrial model), the DTM height is subtracted from the DSM height to obtain building area information, and noise is removed through morphological transformation.

1) Setting initial parameters, including initial window parameter cr0 being 50, initial effective point threshold value minmx0 being 800, elevation level division parameter hoff being 2.0 m, and iteration number xhnum being 3.

2) Identifying non-ground points through multiple iterations

The basic idea of identifying non-ground points is that in a space cube, ground points account for the majority, non-ground points account for the minority, possible non-ground points are separated by recording effective points in a hierarchical space, and non-ground point information is obtained through comprehensive analysis of information such as gradient and neighborhood support degree.

3) Height fitting of non-ground points

And detecting effective points from each invalid point to the 8-neighborhood direction, recording corresponding detection distances, and fitting the elevation of the non-ground point by a distance inverse weighting method. The DSM rendering effect is shown in fig. 14.

Of course, the parameters in the above embodiments are only examples, and in other embodiments, the parameters may be adjusted according to specific situations.

In addition, in view of the characteristic that the data management comprises a large number of complex queries such as spatial data query, combined data query, associated data query and the like, the image storage system realizes the rapid retrieval of data by adopting a mode of combining the spatial data query and the database.

The image storage system stores the spatial information and the metadata information which need to be inquired in a database in a partition table mode. For data containing spatial information, the image storage system is managed in a spatial data set mode, and a grid index is established through a spatial data engine. And managing the metadata information by adopting a database partition table, and establishing a database index. When data retrieval operation is performed, the image storage system performs retrieval operation by adopting different indexes based on different query conditions.

When data query is carried out, the image storage system comprehensively utilizes a multithreading query mechanism, a paging query mechanism, a real-time dynamic query mechanism and an index management mechanism, so that the query efficiency is improved, and the user experience is improved. As shown in fig. 10.

Meanwhile, the data browsing function is also supported, namely the image result of the query is browsed, and the system supports image block diagram browsing and quick view browsing. By checking the check box recorded in the query result, the map area is automatically located to the image graphic area.

And providing data archiving statistics, carrying out classification statistics on all archived data according to conditions, selecting all data types by default, and displaying in a query result column diagram. The current inventory data can be classified and counted on line and off line respectively.

When the space of the disk is insufficient, a data migration function is provided, the data which is not used for the longest time is migrated out of the disk, is transferred to the plug-in disk, the corresponding disk space is vacated, the record is written into a database, and then if the data is needed, the specific storage position of the image is inquired in the system, and the corresponding data is found out.

The storage system meets the requirement of covering the products to rapidly map out in a mode of regularly and automatically manufacturing the products at a background. The image storage system can automatically cover products in a background mode on the basis of time, data types, area information and cloud amount information defined by a user, and the system can store and archive the products after the products in a coverage area are stored. When the user uses the system, only data query and download operation is needed.

The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Although the embodiments of the present invention have been described with reference to the accompanying drawings, it is not intended to limit the scope of the present invention, and it should be understood by those skilled in the art that various modifications and variations can be made without inventive efforts by those skilled in the art based on the technical solution of the present invention.

Claims

1. A remote sensing image data storage method is characterized in that: the method comprises the following steps:

acquiring remote sensing image data;

2. The remote sensing image data storage method of claim 1, wherein: the method comprises the steps of carrying out terrain mapping auxiliary adjustment, reducing the resolution ratio of stereoscopic images, extracting connection points of each fast-view image pair, restoring the fast-view resolution ratio, extracting conventional connection points of original images by combining terrain mapping elevation constraint of space shuttle radars, carrying out free network evaluation and check of original influences, extracting mountain top feature points and flat area feature points according to a certain distance, projecting the two types of feature points onto the original influences respectively, and extracting dense connection points.

3. The remote sensing image data storage method of claim 1, wherein: the system further comprises a coverage map product making module, and the coverage map product making module is used for drawing concurrent map tiles based on the time, the data type, the area information and the cloud cover information input by the stereo image data to generate a coverage range product.

4. The remote sensing image data storage method of claim 1, wherein: and transmitting different remote sensing image data to different processing nodes, and performing data processing among the processing nodes in parallel.

5. The remote sensing image data storage method of claim 4, wherein: before the remote sensing image data processing process is carried out, the maximum parallel number and the computing capacity of each processing node are determined, and N processing nodes with the highest computing capacity rank are determined to execute data processing tasks according to the maximum parallel number N.

6. The remote sensing image data storage method of claim 4, wherein: and grouping the remote sensing image data processing tasks, ranking according to the computing power of each processing node, and matching and distributing the tasks and the processing nodes by using an optimal algorithm.

7. The remote sensing image data storage method of claim 1, wherein: before data storage, the integrity of the data is verified, and if the integrity of the data meets the requirement, the data is stored.

8. A remote sensing image data storage system is characterized in that: the method comprises the following steps:

the data receiving module is configured to acquire remote sensing image data;

9. A computer-readable storage medium characterized by: stored with instructions adapted to be loaded by a processor of a terminal device and to carry out the steps of a method of storing remote sensing image data according to any one of claims 1 to 7.

10. A terminal device is characterized in that: the system comprises a processor and a computer readable storage medium, wherein the processor is used for realizing instructions; a computer readable storage medium for storing a plurality of instructions adapted to be loaded by a processor and to perform the steps of a method of remote sensing image data storage according to any of claims 1-7.