Remote sensing data management method based on lattice quantization space-space coding and multi-dimensional Trie index

Abstract

The invention discloses a remote sensing data management method based on lattice quantization time-space lattice coding and multi-dimensional Trie index, which comprises the following steps: carrying out lattice quantization coding on an obtained remote sensing image to obtain lattice quantization coding information of the remote sensing image, the lattice quantization coding information comprises space lattice quantization coding information, time lattice quantization coding information and / or sensitive target lattice quantization information; organizing lattice quantization coding information by adopting the inverted file, and forming an inverted file table by taking each lattice point in the lattice quantization coding information as a key value of the inverted file; constructing a Trie tree by using the lattice quantization coding information, and organizing and storing an inverted file table by using the Trie tree; and querying a to-be-queried object in the Trie tree by utilizing a Trie parallel search algorithm, wherein the to-be-queried object comprises a to-be-queried remote sensing image or a sensitive target. According to the method, based on the obtained space-time grid codes, the multi-dimensional Trie index of the single-satellite and whole-network data is constructed, the whole-network consistent hierarchical codes are given to satellite remote sensing data, and the basis of unified data management and retrieval is provided.

Description

Technical Field

[0001] The invention belongs to the technical field of remote sensing data management, and in particular relates to a remote sensing data management method based on grid quantization time-space coding and multi-dimensional Trie indexing. Background Art

[0002] Earth observation satellites play an important role in improving the ability to independently obtain remote sensing information and grasping the development trends of the global economy, resources, environment, and society. A key core technology for serving the national economy and ensuring national security is high-precision, real-time, and intelligent remote sensing information, which is also a strategic high ground that countries around the world are competing to seize. At present, my country has hundreds of remote sensing satellites in orbit, each of which is equipped with a variety of remote sensors. Each satellite can observe the ground multiple times a day, and the space application data generated is becoming increasingly huge.

[0003] In order to facilitate the rapid query and extraction of remote sensing data, remote sensing data must be effectively managed and processed. However, at this stage, after the original remote sensing data is obtained on the satellite and transmitted to the ground in large quantities, it undergoes multi-level remote sensing data processing to form a structured form that can be retrieved, accessed and processed. The original satellite remote sensing data cannot directly establish a logical relationship with key objects in orbit and needs to be processed on the ground, making it difficult to manage and process remote sensing data efficiently.

[0004] In addition, the lack of unified data structuring description rules for different satellite data sources also leads to the dispersion of satellite data resources, making it difficult to carry out effective data fusion. Satellite constellations have not formed a capability cluster, which has caused great difficulties in the processing and integration of remote sensing data. Summary of the invention

[0005] In order to solve the above problems existing in the prior art, the present invention provides a remote sensing data management method based on grid quantization space-time coding and multi-dimensional Trie indexing, which gives the remote sensing data a consistent hierarchical coding throughout the network, and provides a basis for unified management and retrieval of remote sensing data; it realizes efficient on-orbit management of remote sensing data, and provides accurate retrieval functions for user applications, which can greatly improve the pertinence of data transmitted on the satellite-to-ground link and greatly improve the utilization efficiency of remote sensing data resources. The technical problem to be solved by the present invention is achieved through the following technical solutions:

[0006] One aspect of the present invention provides a remote sensing data management method based on grid quantization time-space coding and multi-dimensional Trie indexing, comprising:

[0007] Performing grid quantization coding on the acquired remote sensing image to obtain grid quantization coding information of the remote sensing image, wherein the grid quantization coding information includes spatial grid quantization coding information, temporal grid quantization coding information and / or sensitive target grid quantization information;

[0008] Organizing the lattice quantization coding information by using an inverted file, taking each lattice point in the lattice quantization coding information as a key value of the inverted file, and forming an inverted file table;

[0009] Constructing a Trie tree using the lattice quantization coding information, and organizing and storing the inverted file table using the Trie tree;

[0010] The Trie parallel search algorithm is used to query the object to be queried in the Trie tree, and the object to be queried includes the remote sensing image or sensitive target to be queried.

[0011] In one embodiment of the present invention, performing grid quantization coding on an acquired remote sensing image to obtain grid quantization coding information of the remote sensing image includes:

[0012] For the N remote sensing images collected, I = {I i , i=1,2,…,N}, obtain the serial number id corresponding to each remote sensing image, extract the feature vector of each remote sensing image using the time information and longitude and latitude information, and form a remote sensing image data set with the N feature vectors obtained from the N remote sensing images:

[0013] DB = {x i =f(I i ),i=1,2,…,N},

[0014] Where f(·) represents the feature extraction function, x i Represents the feature vector of the i-th remote sensing image;

[0015] Perform space-time vector quantization on the N feature vectors to obtain a grid point set of the N feature vectors:

[0016] DB vq = {LVQ(x i ), x i ∈DB}

[0017] Wherein, LVQ(·) represents the lattice vector quantization function.

[0018] In one embodiment of the present invention, constructing a Trie tree using the lattice quantization coding information, and organizing and storing the inverted file table using the Trie tree includes:

[0019] Use Trie tree to represent the set DB of all key values ​​in the inverted file vq = {l i =(l i,j ,j=1,2,…,N),i=1,2,…,M}, then the alphabet of Trie is Σ={λ L ,(λ L +1),…,0,…,(λH -1),λ H},in, l i represents the i-th key value, M represents the number of all key values, l i,j represents the feature vector of the jth remote sensing image, and N represents the number of the remote sensing images.

[0020] In one embodiment of the present invention, constructing a Trie tree using the lattice quantization coding information, and organizing and storing the inverted file table using the Trie tree specifically includes:

[0021] A collection DB based on all key values vq , construct the Trie tree using a recursive process. If the d-th dimension feature vector is currently being processed, 1≤d≤N, set a subset S, The process of establishing a Trie node for a subset S includes:

[0022] Let |S| represent the number of elements in set S. If |S|>1, separate S according to the dth dimension (λ H -λ L +1) subset: Based on According to the (d+1)th dimension, we recursively construct (λ H -λ L +1) Trie child nodes, and finally establish an intermediate node to convert the (λ H -λ L +1) Trie child nodes are connected to the corresponding positions of the intermediate nodes;

[0023] If S = Φ, create an empty node; if |S| = 1, create a terminal node. Let l be the only element of S, {id1, id2, ...} be the image sequence number set corresponding to element l, and (l d+1 ,l d+2 ,…l N ) and the sequence number {id1, id2, ...} are stored in the terminal node, where l d+1 represents the key value of the d+1th dimension, l d+2 represents the key value of the d+2th dimension, l N Represents the key value of the Nth dimension;

[0024] Using S=DB vq The process of establishing a Trie node is called with d=1, thereby recursively establishing a complete Trie tree, where d represents the dimension of the feature vector.

[0025] In one embodiment of the present invention, a Trie parallel search algorithm is used to query an object to be queried in the Trie tree, wherein the object to be queried includes a remote sensing image or a sensitive target to be queried, including:

[0026] Obtain a query vector f(Q0) of the object to be queried Q0, and quantize the query vector y of the object to be queried Q0 using a lattice vector quantization function to obtain a key value l0=LVQ(y) of the object to be queried;

[0027] Set the query threshold range T and calculate the normalized threshold value according to the query threshold range T Among them, T0 represents the maximum value of the query threshold range T, Indicates rounding operation;

[0028] Set the hyperrectangular window W to be searched Rect :

[0029] l Low =(max((l 1，1 -δ), λ L ), max((l 1，2 -δ), λ L ),…,max((l 1，n -δ), λ L ))

[0030] l High =(min((l 1，1 -δ), λ H ), min((l 1，2 +δ),λ H ))…,min((l 1，n -δ), λ H ))

[0031] Among them, l Low and l High Represent the super rectangular window W Rect The two diagonal vertices of 1,1 Indicates the first dimension key value of the first feature vector, l 1,2 represents the second dimension key value of the first feature vector, and so on. 1,n Represents the n-th dimension key value of the first feature vector;

[0032] In the inverted file table, the super rectangular window W is searched using the Trie parallel search algorithm. Rect All grid points in get the set of objects of interest

[0033] Use the range query formula to query the set of objects of interest Perform a sequential search to obtain the final query result set

[0034] In one embodiment of the present invention, in the inverted file table, the super rectangular window W is searched using a Trie parallel search algorithm. Rect All grid points in include:

[0035] Let d represent the dimension of the feature vector currently being processed, and let Root trie represents the root of the current subtree, then:

[0036] If Root trie If it is NULL, the search ends;

[0037] If Root trie As the terminal node, let l′=(l d+1 ,l d+2 ,…,l n ) is stored in Root trie The remaining part of the , for i = d + 1, d + 2, ..., n, has l Low,i ≤l i ≤l High,i , then the feature vectors corresponding to all image numbers in the terminal node are added to the set of objects of interest Among them, l Low,i represents the i-th dimension l Low , l High,i represents the i-th dimension l High ;

[0038] If Root trie If it is an intermediate node, then Root trie The first Low,d Branches to the first High,d Take the branch as the root, let d←d+1, and continue searching repeatedly.

[0039] In one embodiment of the present invention, the method further comprises:

[0040] The objects to be deleted are deleted in the Trie tree, and the objects to be deleted include remote sensing images or sensitive targets to be deleted.

[0041] In one embodiment of the present invention, deleting the object to be deleted in the Trie tree includes:

[0042] Obtaining the grid vector encoding information of the object to be deleted;

[0043] Call the corresponding index function and determine whether the grid vector encoding information of the object to be deleted has a corresponding index. If so, call the erase function to erase the original remote sensing data corresponding to the object to be deleted; if not, the erasure ends.

[0044] Another aspect of the present invention provides a storage medium storing a computer program for executing the steps of the remote sensing data management method based on grid quantization time-space coding and multi-dimensional Trie indexing as described in any one of the above embodiments.

[0045] Another aspect of the present invention provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the processor calls the computer program in the memory, it implements the steps of the remote sensing data management method based on grid quantization time-space coding and multi-dimensional Trie indexing as described in any one of the above embodiments.

[0046] Compared with the prior art, the present invention has the following beneficial effects:

[0047] The remote sensing data management method based on grid quantization space-time coding and multi-dimensional Trie indexing of the present invention gives satellite remote sensing data a hierarchical coding that is consistent across the entire network, providing a basis for unified data management and retrieval. Based on the obtained space-time grid coding, a multi-dimensional Trie index for single satellite and network-wide data is constructed to achieve efficient on-orbit management of remote sensing data, provide accurate retrieval functions for user applications, greatly improve the pertinence of data transmitted on satellite-to-ground links, greatly improve the efficiency of remote sensing data resource utilization, support very high-dimensional data indexing, make full use of the sparsity of point distribution in high-dimensional space, and establish a relatively low-complexity index structure.

[0048] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 It is a flowchart of a remote sensing data management method based on grid quantization time-space coding and multi-dimensional Trie indexing provided by an embodiment of the present invention;

[0050] Figure 2 It is a data storage flow chart based on lattice quantization coding and multi-dimensional Trie indexing provided by an embodiment of the present invention;

[0051] Figure 3 It is a data retrieval and playback flow chart based on lattice quantization coding and multi-dimensional Trie indexing provided by an embodiment of the present invention;

[0052] Figure 4 It is a data deletion flow chart based on lattice quantization coding and multi-dimensional Trie indexing provided by an embodiment of the present invention;

[0053] Figure 5 Schematic diagram of the device required for the experiment provided by the embodiment of the present invention. DETAILED DESCRIPTION

[0054] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined purpose of the invention, the remote sensing data management method based on grid quantization time-space coding and multi-dimensional Trie indexing proposed by the present invention is described in detail below in combination with the accompanying drawings and specific implementation methods.

[0055] The above and other technical contents, features and effects of the present invention are clearly presented in the following detailed description of the specific implementation modes in conjunction with the accompanying drawings. Through the description of the specific implementation modes, the technical means and effects adopted by the present invention to achieve the predetermined purpose can be more deeply and specifically understood. However, the attached drawings are only for reference and explanation purposes and are not used to limit the technical solutions of the present invention.

[0056] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants are intended to cover non-exclusive inclusion, so that an article or device including a series of elements includes not only those elements, but also other elements that are not explicitly listed. In the absence of more restrictions, the elements defined by the statement "including one..." do not exclude the existence of other identical elements in the article or device including the elements.

[0057] In order to further promote the development of near-real-time intelligent services for remote sensing data, it is necessary to improve the efficiency of remote sensing data collection and maintenance through satellite cluster networking, and combine multiple satellite observation and service capabilities to meet data acquisition needs such as wider observation, rapid response to specific mission objectives, shorter revisit cycles, and continuous dynamic monitoring. Therefore, it is necessary to form a unified remote sensing data organization form and application interface from the source of remote sensing data storage, and form a satellite capability cluster to realize data resource sharing between satellites and ground satellites.

[0058] See also Figure 1 , Figure 1 The present invention provides a flowchart of a remote sensing data management method based on grid quantization time-space coding and multi-dimensional Trie indexing. The remote sensing data management method includes:

[0059] S1: performing grid quantization coding on the acquired remote sensing image to obtain grid quantization coding information of the remote sensing image, where the grid quantization coding information includes spatial grid quantization coding information, temporal grid quantization coding information and / or sensitive target grid quantization information.

[0060] Step S1 of this embodiment specifically includes: performing spatial and temporal quantization coding on the acquired remote sensing image to obtain spatial grid quantization coding information and temporal grid quantization coding information of the remote sensing image.

[0061] Specifically, for the N remote sensing images collected I = {I i , i=1,2,…,N}, each remote sensing image corresponds to a unique serial number id, and a feature vector is extracted from each remote sensing image using time information and longitude and latitude information. The N feature vectors obtained from N remote sensing images form a remote sensing image data set:

[0062] DB = {x i =f(I i ),i=1,2,…,N},

[0063] Where f(·) represents the feature extraction function, x i Represents the feature vector of the i-th remote sensing image.

[0064] Then, the above N feature vectors are subjected to time-space vector quantization. After grid vector quantization, different feature vectors may fall into the same cell cavity and be quantized to the same grid point, thereby obtaining a grid point set of N feature vectors:

[0065] DB vq = {LVQ(x i ),x i ∈DB),

[0066] Where LVQ(·) is the grid vector quantization function. Assume that the grid point set DB vq The number of grid points is M. Since different eigenvectors may fall into the same cell, it is obvious that M≤N.

[0067] In another embodiment of the present invention, step S1 further includes: performing feature grid quantization encoding on sensitive targets in the remote sensing image.

[0068] Specifically, obtain K sensitive targets F = {F j ,j=1,2,…,K}, each sensitive target corresponds to a unique serial number id, and the corresponding feature vector is extracted from each sensitive target using the feature extraction algorithm. Then these K feature vectors will form the sensitive target data set:

[0069] DB1={x j =f(F j)j=1,2,…,K}

[0070] The feature vectors of K sensitive targets are quantized by grid vectors. After grid vector quantization, different feature vectors may fall into the same cell cavity and be quantized to the same grid point, thereby obtaining the grid point set of K sensitive targets:

[0071] DB1 vq = {l k =LVQ(x j ),x j ∈DB1}

[0072] The sensitive targets here can be ships, aircraft carriers, tanks, fighter jets and other targets.

[0073] S2: Use an inverted file to organize the lattice quantization coding information, and use each lattice point in the lattice quantization coding information as a key value of the inverted file to form an inverted file table.

[0074] In this embodiment, an inverted file is used to organize the grid set DB. vq and / or DB1 vq , the grid points are used as the key values ​​of the inverted file. Each grid point (that is, each key value) can correspond to multiple image serial numbers id, as shown in Table 1. The key value l1 corresponds to the image serial number id 1,1 ,id 1,2 ,….

[0075] Table 1 Inverted file storage grid set and image sequence number

[0076] Key-value Image number <![CDATA[l1]]> <![CDATA[id 1,1 ,id 1,2 ,…]]> <![CDATA[l2]]> <![CDATA[id 2,1 ,id 2,2 ,…]]> … … <![CDATA[l M ]]> <![CDATA[id M,1 ,id M,2 ,…]]>

[0077] S3: Use the lattice quantization coding information to construct a Trie tree, and use the Trie tree to organize and store the inverted file table.

[0078] See also Figure 2 , Figure 2 This is a data storage flow chart based on lattice quantization coding and multi-dimensional Trie indexing provided by an embodiment of the present invention. In order to quickly access the inverted file table, this embodiment uses a Trie tree (prefix tree) to represent the set DB of all key values ​​(lattice points) in the inverted file. vq = {l i =(l i,j ,j=1,2,…,N),i=1,2,…,M}, then the alphabet of the Trie tree is Σ={λ L ,(λ L +1),…,0,…,(λ H -1),λ H},in, Therefore, there is l irepresents the i-th key value, M represents the number of all key values, l i,j represents the feature vector of the jth remote sensing image, and N represents the number of remote sensing images.

[0079] In fact, based on the key-value collection DB vq Constructing a Trie tree is a recursive process. Specifically, assuming that the d-th dimension feature vector (1≤d≤N) is currently being processed, a subset S is set (S represents a subset of the storage grid DBvq) So the process of building a Trie node for subset S is as follows:

[0080] Let |S| represent the number of elements in subset S. If |S|>1, separate S according to the dth dimension (λ H -λ L +1) subset: Then based on the subset According to the (d+1)th dimension, recursively construct (λ H -λ L +1) Trie child nodes; establish an intermediate node to convert the above (λ H -λ L +1) Trie child nodes are connected to the corresponding positions of the intermediate nodes.

[0081] If S=Φ, create an empty node.

[0082] If |S|=1, then establish a terminal node. Let l∈S be the unique element of S, {id1,id2,…} be the image sequence number set corresponding to l, and (l d+1 ,l d+2 ,…l N ) and the image sequence number {id1, id2, ...} are stored in the terminal node.

[0083] Using S=DB vq The above process of establishing Trie nodes can be called with d=1 to recursively establish a complete Trie tree. At this point, the reorganization and storage of the inverted file list using Trie is completed, where d represents the dimension of the feature vector.

[0084] S4: Query the query object in the Trie tree using the Trie parallel search algorithm, where the query object includes the remote sensing image or sensitive target to be queried.

[0085] See also Figure 3 , Figure 3 Detailed description of the present invention is a data retrieval and playback flow chart based on lattice quantization coding and multi-dimensional Trie indexing. Step S4 of this embodiment specifically includes the following steps:

[0086] S4.1: Assuming that the object to be queried (the remote sensing image or sensitive target to be queried) is Q0, the query vector y=f(Q0) is calculated using the feature extraction function, and then the query vector y of the query object Q0 is quantized using the same grid vector quantization function to obtain l0=LVQ(y).

[0087] S4.2: Set the query threshold range T and calculate the normalized threshold value based on the query threshold range T Among them, T0 represents the maximum value of the query threshold range T, Indicates rounding operation.

[0088] S4.3: Set the hyper-rectangular window W to be searched Rect :

[0089] l Low =(max((l 1，1 -δ), λ L ), max((l 1，2 -δ), λ L ),…,max((l 1，n -δ), λ L ))

[0090] l High =(min((l 1，1 -δ), λ H ), min((l 1，2 +δ),λ H ))…,min((l 1，n -δ), λ H ))

[0091] Among them, l Low and l High Represent the super rectangular window W Rect The two diagonal vertices of 1,1 Indicates the first dimension key value of the first feature vector, l 1,2 represents the second dimension key value of the first feature vector, and so on, l1, n Represents the n-th dimension key value of the first feature vector.

[0092] S4.4: In the inverted file list, use the Trie Parallel Search Algorithm (TPSA) to search the super rectangular window W Rect All grid points in get the set of objects of interest

[0093] S4.5: Based on the range query formula, the set of objects of interest Perform a sequential search to obtain the final query result set The range query formula of this embodiment is:

[0094]

[0095] in, is the query result set, T is the query threshold range, which can be set empirically, and D represents the distance measure. The smaller D is, the greater the similarity is.

[0096] It should be noted that the complexity of the above range query formula mainly depends on the search speed of the postings list. The slowest way to search the postings list is sequential scanning, which requires comparing the key value in the grid point l1 with the key value in the postings list one by one. In fact, as the dimension increases, the hyperrectangular window W Rect The number of grid points increases rapidly and may be much larger than DB vq The number of grid points in M. However, |W Rect ∩DB vq | is always very finite, because the super-rectangular window W Rect The number of valid grid points in is very small, and will not exceed M at most. There are many algorithms for quickly accessing inverted files, such as hash search, binary search, and Trie tree search. The embodiment of the present invention adopts a method based on a Trie tree because both binary search and hash search must use a super rectangular window W Rect All grid points in the inverted file are searched, so even if W does not exist in the inverted file Rect The total complexity of |W Rect |Second inverted file search.

[0097] Furthermore, in order to quickly search for the valid grid point set (W Rect ∩DB vq ),get The present invention proposes a Trie Parallel Search Algorithm (TPSA). TPSA is a recursive process. d represents the dimension currently being processed, and Root trie represents the root of the current subtree, then:

[0098] If Root trie If it is NULL (empty), the search ends.

[0099] If Root trie As the terminal node, let l′=(l d+1 ,l d+2 ,…,l n ) is stored in Root trie The remaining part of the , for i = d + 1, d + 2, ..., n, has l Low,i ≤l i ≤l High,i , then the feature vectors corresponding to all image numbers in the terminal node are added to the set of objects of interest Among them, l Low,i represents the i-th dimension l Low , l High,i represents the i-th dimension l High .

[0100] If Root trie If it is an intermediate node, then Root trie The first Low,d Branches to the first High,d Take the branch as the root, let d←d+1, call this recursive process, and continue searching.

[0101] Let d = 1, The initial value of Root is NULL. trie As the root, calling the above recursive process, we get In the experiment, it was found that there are a large number of empty nodes in the Trie tree. In order to further reduce the complexity of TPSA, the original Trie tree was improved. By adding a small amount of empty node indication information (indicating the location of the empty node information), deleting all empty nodes, and obtaining a compressed Trie tree. Then, with a slight modification to TPSA, fast search can be performed on the compressed Trie, which improves the calculation speed.

[0102] In this embodiment, there are four scenarios for playing back data according to the index: single spatial grid coding (optional auxiliary time coding) mode, single time grid coding index mode, time period index mode and sensitive target (trajectory) index. The specific steps of data retrieval and playback are: 1) perform spatial grid quantization coding, temporal grid quantization coding or sensitive target feature grid quantization coding according to the information to be indexed; 2) use different indexing methods according to different indexing conditions; 3) find the query results that meet the conditions and play back the data; 4) end the playback of all data.

[0103] Furthermore, the remote sensing data management method of this embodiment also includes: S5: deleting the objects to be deleted in the Trie tree, where the objects to be deleted include the remote sensing images or sensitive targets to be deleted.

[0104] See also Figure 4 , Figure 4 This is a data deletion flow chart based on lattice quantization coding and multi-dimensional Trie indexing provided by an embodiment of the present invention. This step specifically includes:

[0105] S5.1: Obtain lattice vector coding information of the object to be deleted (at least one of spatial lattice vector coding, temporal lattice vector coding, and sensitive target lattice vector coding);

[0106] S5.2: Call the corresponding index function and determine whether the grid vector encoding information of the object to be deleted has a corresponding index. If so, call the erase function to erase the original remote sensing data corresponding to the object to be deleted. If not, the erasure ends.

[0107] The effect of the remote sensing data management method based on grid quantization time-space coding and multi-dimensional Trie indexing according to the embodiment of the present invention is further illustrated by experiments below.

[0108] See also Figure 5 , Figure 5 , in addition to other peripherals such as power supply, the main unit equipment includes camera simulation unit equipment, satellite storage equipment, data transmission simulation unit equipment, satellite service computer simulation computer PC1 and index data demonstration computer PC2.

[0109] PC1, as the main control unit of the experimental system, communicates commands and data with the camera simulation unit through the network port. It communicates with the onboard storage device through the CAN interface to control the process of storing and retrieving remote sensing data in the onboard storage device, and sends auxiliary information of remote sensing images.

[0110] The camera simulation unit device uses a development board with Xilinx Zynq7035 FPGA. With FPGA software logic, it can generate image data in a specified format. On the right side of the development board is an interface adapter board, which converts the external data interface into two Camera LinkBase interfaces, which are connected to the onboard storage device through the Camera Link cable. The camera simulation unit device communicates with PC1 through the network port, receives control instructions from PC1, and generates remote sensing image data with a size of 3960 pixels * 3960 pixels.

[0111] The main core components of the onboard storage device include a Xilinx Zynq7100 FPGA and 8 NAND flash arrays. There are three external peripheral interfaces: two Camera Link Base interfaces are used to receive remote sensing data from the front-end camera; two CAN interfaces are provided to receive image auxiliary data (such as longitude and latitude) from PC1, integrate and store them, and establish a grid storage management system for image data on the satellite. In addition, there is an LVDS output interface, which is connected to the digital transmission analog unit device through an LVDS cable to play back data to the digital transmission analog unit device.

[0112] The data transmission simulation unit device and the camera simulation unit device use the same development board with Xilinx Zynq7035FPGA, equipped with FPGA software logic, to receive the specified remote sensing data and forward it to the Gigabit Ethernet port, and send the satellite remote sensing data to PC2 through the network port. The leftmost part of the data transmission simulation unit device is the interface adapter board, which mainly converts the external data interface into an LVDS interface to interconnect with the onboard storage device.

[0113] PC2 is mainly used to receive remote sensing data and auxiliary information sent by the onboard storage device for display. Since there is no interconnection between PC1 and PC2, the data of PC2 comes entirely from the data retrieved and played back by the onboard storage device, so PC2 can also be used to judge the correctness of the data retrieval algorithm of the onboard storage device.

[0114] 1. Data storage

[0115] The steps of the data storage process demonstration process are as follows: 1) Start the camera simulation unit device, the onboard storage device and the data transmission simulation unit device, and put them in standby mode; 2) Open the demonstration system of PC1, select the satellite ID, and click the button to start recording image data; 3) Run the recording process test script program of PC1 to start recording 10 pictures of the specified satellite; 4) Repeat steps 1) to 2) to complete the data recording process of 3 satellites. The experimental test script of the storage process is designed on PC1, and its main functions include: (1) Generate different image information for each satellite in the storage (image longitude center position, image latitude center position, image generation time, sensitive target information, etc.), call the grid quantization coding algorithm database, and calculate the corresponding spatial grid quantization coding, temporal grid quantization coding and sensitive target grid quantization information for each image; (2) Send control instructions through the CAN interface to start the onboard storage device to start working, and then control the camera simulation unit device to generate image data. When the image data is sent to the onboard storage device, the auxiliary data is also sent to the storage unit through the auxiliary data instructions of the CAN interface, and the spatial grid quantization coding and temporal grid quantization coding of the image are sent to the storage unit.

[0116] 2. Spatiotemporal retrieval of massive remote sensing data

[0117] The data retrieval and playback process is: 1) Start the camera simulation unit device, the onboard storage device and the digital transmission simulation unit device, and put them in standby mode; 2) Run the playback process test script program on PC2, and put it in a state of waiting to receive data; 3) Open the demonstration system of PC2 and wait for the data to be displayed; 4) Select the satellite ID, longitude and latitude range and time information to be played back in the upper left corner window of the demonstration system of PC1, and click to start query playback. At this time, the main window will display the retrieval results circled by a red frame; 5) At the same time, according to the options on the demonstration system of PC1, send the specified playback command to the onboard storage device, start the onboard storage device to start playing back the retrieved data on demand, and send it to the digital transmission simulation unit device, and the digital transmission simulation unit device sends it to PC2; 6) The playback process test script program on PC2 receives the played back data, forms a file, and parses the image data; 7) At this time, the results of the playback retrieval can be seen on the demonstration system of PC2. Since the onboard storage device supports retrieval by location interval and by time period, the playback experiment can cycle steps 4) to 7) to test different playback modes.

[0118] (III) Sensitive target trajectory query

[0119] Public ship track data was obtained from MarineCadastre.gov, mainly in the Gulf of Mexico and its surrounding areas from January to March 2018. The data contains 47,182,135 track points, and 66,322 spatiotemporal track lines can be formed by constructing these track points. To prevent the track line from being too long, the time width of a single track line is limited to 24 hours. Table 2 shows the average time consumption comparison of sensitive target track query using the existing ST-Geohash method and the method proposed in the embodiment of the present invention. It can be seen from the experimental results that the efficiency advantage of the method proposed in the embodiment of the present invention for spatiotemporal track line query is obvious, which verifies the effectiveness of the method of the present invention.

[0120] Table 2 Comparison of average time consumption for trajectory line query

[0121]

[0122] The embodiment of the present invention is based on a remote sensing data management method based on grid quantization space-time coding and multi-dimensional Trie indexing, which gives satellite remote sensing data a hierarchical coding that is consistent across the entire network, and provides a basis for unified data management and retrieval. Based on the obtained space-time grid coding, a multi-dimensional Trie index for single satellite and network-wide data is constructed to achieve efficient on-orbit management of remote sensing data, provide accurate retrieval functions for user applications, greatly improve the pertinence of data transmitted on satellite-to-ground links, greatly improve the efficiency of remote sensing data resource utilization, support very high-dimensional data indexing, make full use of the sparsity of point distribution in high-dimensional space, and establish a relatively low-complexity index structure.

[0123] Another embodiment of the present invention provides a storage medium, wherein a computer program is stored in the storage medium, and the computer program is used to execute the steps of the remote sensing data management method based on grid quantization time space coding and multidimensional Trie indexing as described in the above embodiment. Another aspect of the present invention provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the processor calls the computer program in the memory, the steps of the remote sensing data management method based on grid quantization time space coding and multidimensional Trie indexing as described in the above embodiment are implemented. Specifically, the above integrated module implemented in the form of a software function module can be stored in a computer-readable storage medium. The above software function module is stored in a storage medium, including several instructions for enabling an electronic device (which can be a personal computer, a server, or a network device, etc.) or a processor (processor) to execute some steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a disk or an optical disk.

[0124] The above contents are further detailed descriptions of the present invention in combination with specific preferred embodiments, and it cannot be determined that the specific implementation of the present invention is limited to these descriptions. For ordinary technicians in the technical field to which the present invention belongs, several simple deductions or substitutions can be made without departing from the concept of the present invention, which should be regarded as falling within the protection scope of the present invention.

Claims

1. A remote sensing data management method based on grid quantization time-space coding and multi-dimensional Trie indexing, characterized in that: include: Performing grid quantization coding on the acquired remote sensing image to obtain grid quantization coding information of the remote sensing image, wherein the grid quantization coding information includes spatial grid quantization coding information, temporal grid quantization coding information and / or sensitive target grid quantization information; Organizing the lattice quantization coding information by using an inverted file, taking each lattice point in the lattice quantization coding information as a key value of the inverted file, and forming an inverted file table; Constructing a Trie tree using the lattice quantization coding information, and organizing and storing the inverted file table using the Trie tree; The Trie parallel search algorithm is used to query the object to be queried in the Trie tree, and the object to be queried includes the remote sensing image or sensitive target to be queried.

2. The remote sensing data management method based on lattice quantization time-space coding and multi-dimensional Trie indexing according to claim 1 is characterized in that: Performing grid quantization coding on the acquired remote sensing image to obtain grid quantization coding information of the remote sensing image includes: For the N remote sensing images collected, I = {I i , i=1,2,…,N}, obtain the serial number id corresponding to each remote sensing image, extract the feature vector of each remote sensing image using the time information and longitude and latitude information, and form a remote sensing image data set with the N feature vectors obtained from the N remote sensing images: DB={x i =f(I i ),i=1,2,…,N}, Where f(·) represents the feature extraction function, x i Represents the feature vector of the i-th remote sensing image; Perform space-time vector quantization on the N feature vectors to obtain a grid point set of the N feature vectors: DB vq ={LVQ(x i )x i ∈DB} Wherein, LVQ(·) represents the lattice vector quantization function.

3. The remote sensing data management method based on lattice quantization time-space coding and multi-dimensional Trie indexing according to claim 2 is characterized in that: Constructing a Trie tree using the lattice quantization coding information, and organizing and storing the inverted file table using the Trie tree, including: Use Trie tree to represent the set DB of all key values ​​in the inverted file vq = {l i =(l i,j ,j=1,2,…,N),i=1,2,…,M}, then the alphabet of Trie is Σ={λ L ,(λ L +1),…,0,…,(λ H -1),λ H },in, l i represents the i-th key value, M represents the number of all key values, l i,j represents the feature vector of the jth remote sensing image, and N represents the number of the remote sensing images.

4. The remote sensing data management method based on lattice quantization time-space coding and multi-dimensional Trie indexing according to claim 2 is characterized in that: Constructing a Trie tree using the lattice quantization coding information, and organizing and storing the inverted file table using the Trie tree, specifically includes: A collection DB based on all key values vq , construct the Trie tree using a recursive process. If the d-th dimension feature vector is currently being processed, 1≤d≤N, set a subset S, The process of establishing a Trie node for a subset S includes: Let |S| represent the number of elements in set S. If |S|>1, separate S according to the dth dimension (λ H -λ L +1) subset: Based on According to the (d+1)th dimension, we recursively construct (λ H -λ L +1) Trie child nodes, and finally establish an intermediate node to convert the (λ H -λ L +1) Trie child nodes are connected to the corresponding positions of the intermediate nodes; If S = Φ, create an empty node; if |S| = 1, create a terminal node. Let l be the only element of S, {id1, id2, ...} be the image sequence number set corresponding to element l, and (l d+1 ,l d+2 ,…l N ) and the sequence number {id1, id2, ...} are stored in the terminal node, where l d+1 represents the key value of the d+1th dimension, l d+2 represents the key value of the d+2th dimension, l N Represents the key value of the Nth dimension; Using S=DB vq The process of establishing a Trie node is called with d=1, thereby recursively establishing a complete Trie tree, where d represents the dimension of the feature vector.

5. The remote sensing data management method based on lattice quantization time-space coding and multi-dimensional Trie indexing according to claim 4 is characterized in that: The Trie parallel search algorithm is used to query the query object in the Trie tree, and the query object includes the remote sensing image or sensitive target to be queried, including: Obtain a query vector f(Q0) of the object to be queried Q0, and quantize the query vector y of the object to be queried Q0 using a lattice vector quantization function to obtain a key value l0=LVQ(y) of the object to be queried; Set the query threshold range T and calculate the normalized threshold value according to the query threshold range T Among them, T0 represents the maximum value of the query threshold range T, Indicates rounding operation; Set the hyperrectangular window W to be searched Rect : l Low =(max((l 1,1 -d),l L ),max((l 1,2 -d),l L ),…,max((l 1,n -d),l L )) l High =(min((l 1,1 +d),l H ),min((l 1,2 +d),l H ),…,min((l 1,n +d),l H )) Among them, l Low and l High Represent the super rectangular window W Rect The two diagonal vertices of 1,1 Indicates the first dimension key value of the first feature vector, l 1,2 represents the second dimension key value of the first feature vector, and so on. 1,n Represents the n-th dimension key value of the first feature vector; In the inverted file table, the super rectangular window W is searched using the Trie parallel search algorithm. Rect All grid points in get the set of objects of interest Use the range query formula to query the set of objects of interest Perform a sequential search to obtain the final query result set 6. The remote sensing data management method based on lattice quantization time-space coding and multi-dimensional Trie indexing according to claim 5 is characterized in that: In the inverted file table, the super rectangular window W is searched using the Trie parallel search algorithm. Rect All grid points in get the set of objects of interest include: Let d represent the dimension of the feature vector currently being processed, and let Root trie represents the root of the current subtree, then: If Root trie If it is NULL, the search ends; If Root trie As the terminal node, let l′=(l d+1 ,l d+2 ,…,l n ) is stored in Root trie The remaining part of the , for i = d + 1, d + 2, ..., n, has l Low,i ≤l i ≤l High,i , then the feature vectors corresponding to all image numbers in the terminal node are added to the set of objects of interest Among them, l Low,i represents the i-th dimension l Low , l High,i represents the i-th dimension l High ; If Root trie If it is an intermediate node, then Root trie The first Low,d Branches to the first High,d Take the branch as the root, let d←d+1, and continue searching repeatedly.

7. The remote sensing data management method based on lattice quantization time-space coding and multi-dimensional Trie indexing according to claim 6 is characterized in that: The method further comprises: The objects to be deleted are deleted in the Trie tree, and the objects to be deleted include remote sensing images or sensitive targets to be deleted.

8. The remote sensing data management method based on lattice quantization time-space coding and multi-dimensional Trie indexing according to claim 7 is characterized in that: Deleting the object to be deleted in the Trie tree includes: Obtaining the grid vector encoding information of the object to be deleted; Call the corresponding index function and determine whether the grid vector encoding information of the object to be deleted has a corresponding index. If so, call the erase function to erase the original remote sensing data corresponding to the object to be deleted; if not, the erasure ends.

9. A storage medium storing a computer program for executing the steps of the remote sensing data management method based on lattice quantization time-space coding and multi-dimensional Trie indexing as described in any one of claims 1 to 8.

10. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, and when the processor calls the computer program in the memory, the processor implements the steps of the remote sensing data management method based on grid quantization time-space coding and multi-dimensional Trie indexing as described in any one of claims 1 to 8.

Images