CN105528622B - A kind of high-resolution remote sensing image sorting algorithm spatiotemporal efficiency optimization method - Google Patents
A kind of high-resolution remote sensing image sorting algorithm spatiotemporal efficiency optimization method Download PDFInfo
- Publication number
- CN105528622B CN105528622B CN201511003105.3A CN201511003105A CN105528622B CN 105528622 B CN105528622 B CN 105528622B CN 201511003105 A CN201511003105 A CN 201511003105A CN 105528622 B CN105528622 B CN 105528622B
- Authority
- CN
- China
- Prior art keywords
- pixel
- block
- cluster centre
- remote sensing
- image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24133—Distances to prototypes
- G06F18/24137—Distances to cluster centroïds
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23211—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with adaptive number of clusters
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Probability & Statistics with Applications (AREA)
- Image Analysis (AREA)
- Television Signal Processing For Recording (AREA)
Abstract
The present invention relates to Computer Image Processing and remote sensing technology field, disclose a kind of high-resolution remote sensing image sorting algorithm spatiotemporal efficiency optimization method, using the strategy for carrying out piecemeal processing to image, introduce CUDA parallelization calculating and storage mapping file technique, the read-write operation of disk is avoided to be greatly saved the runing time of algorithm, by the algorithm of reasonably optimizing, remain the few advantage of piecemeal processing method EMS memory occupation, and image file access speed is improved, improve the time efficiency and space efficiency of remote sensing image processing.
Description
Technical field
The present invention relates to Computer Image Processing and remote sensing technology field, especially a kind of high-resolution remote sensing image classification
Algorithm spatiotemporal efficiency optimization method.
Background technique
With the fast development of satellite remote sensing technology, the data volume of remote sensing images is increasing.For example, SPOT-5 satellite mapping
As storing 5.76*108 pixel, WorldView-2 satellite image stores 2.62*108 pixel.So big data volume
Great pressure and challenge are brought to Remote Sensing Data Processing, is mainly manifested in that program runtime is too long, calculator memory is insufficient.
Current computer CPU frequency and memory size the comparison of the growth are slow, therefore solve at remote sensing images by the raising of hardware performance
The problem of managing overlong time and low memory is relatively difficult.How the time efficiency and space efficiency of remote sensing image processing are improved,
Have become one of the most pressing problem that field of remote sensing image processing faces.
Summary of the invention
The purpose of the present invention is to provide a kind of high-resolution remote sensing image sorting algorithm spatiotemporal efficiency optimization methods, retain
Piecemeal processing method EMS memory occupation few advantage, and image file access speed is improved, improve remote sensing image processing
Time efficiency and space efficiency.
To realize above-mentioned technical purpose and the technique effect, the invention discloses a kind of high-resolution remote sensing images
Sorting algorithm spatiotemporal efficiency optimization method, optimization algorithm specific steps are as follows:
A: image data is divided into several block of pixels;
B: text is mapped using the set of pixels of the RasterIO function block-by-block load image in GDAL function library, and using memory
Part technology block-by-block pixel data dumps in new disk file;
C: K pixel is randomly selected as initial cluster center;
D: the pixel data of first piecemeal is read from disk file using storage mapping file technique, and is transferred to and sets
Standby end;
E: in each pixel that equipment end calculates piecemeal at a distance from each cluster centre, divided by phase approximately principle
Class;
F: classification results are passed back host side, and result data is dumped into disk text using storage mapping file technique
Part;
G: if the data read in step D are the last one block datas, step H is gone to, is otherwise read next
Block data simultaneously goes to step E;
H: the average value of each pixel in every one kind is calculated, and using this average value as new cluster centre;
I: whether newer cluster centre is identical as old cluster centre, old with the replacement of new cluster centre after comparing
Cluster centre, comparison result difference then goes to step D, identical, goes to step J;
J: the pixel value of each cluster centre is assigned to such each pixel.
Wherein, image data is divided into several block of pixels, pixel block size be 128*128,256*256,512*512 and
One of 1024*1024.
Wherein, newer cluster centre and old cluster centre, can be used absolute distance, Euclidean distance, mahalanobis distance,
Manhatton distance, included angle cosine distance metric formula carry out similarity evaluation.
The invention has the following advantages:
1. introducing CUDA parallelization calculating and memory mapping present invention employs the strategy for carrying out piecemeal processing to image
File Technology avoids the read-write operation of disk to be greatly saved the runing time of algorithm.
2. remaining the few advantage of piecemeal processing method EMS memory occupation, and improve figure by the algorithm of reasonably optimizing
As file access speed, the time efficiency and space efficiency of remote sensing image processing are improved.
Detailed description of the invention
Fig. 1 is flow diagram of the invention.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to the accompanying drawings and embodiments, right
The present invention is further elaborated.
Embodiment 1
The invention discloses a kind of high-resolution remote sensing image sorting algorithm spatiotemporal efficiency optimization method, optimization algorithm is specific
Step are as follows:
A: image data is divided into several block of pixels, and every piece of block of pixels is 512*512;
B: text is mapped using the set of pixels of the RasterIO function block-by-block load image in GDAL function library, and using memory
Part technology block-by-block pixel data dumps in new disk file;
C: K pixel is randomly selected as initial cluster center;
D: the pixel data of first piecemeal is read from disk file using storage mapping file technique, and is transferred to and sets
Standby end;
E: in each pixel that equipment end calculates piecemeal at a distance from each cluster centre, divided by phase approximately principle
Class;
F: classification results are passed back host side, and result data is dumped into disk text using storage mapping file technique
Part;
G: if the data read in step D are the last one block datas, step H is gone to, is otherwise read next
Block data simultaneously goes to step E;
H: the average value of each pixel in every one kind is calculated, and using this average value as new cluster centre;
I: whether newer cluster centre is identical as old cluster centre, old with the replacement of new cluster centre after comparing
Cluster centre, comparison result difference then goes to step D, identical, goes to step J;
J: the pixel value of each cluster centre is assigned to such each pixel.
Wherein, newer cluster centre and old cluster centre, included angle cosine distance metric formula carry out similitude and comment
Valence.
Embodiment 2
Experiment purpose and method: in order to evaluate the application feasibility and validity, the present embodiment is with K-Means algorithm
Example, small image (600*600 pixel) the He Zhengjing image (6000*6000 pixel) after having used a width to cut are tested,
For the image of 600*600 pixel, it compared " CPU+ memory ", " CUDA+ memory " and " CUDA+MMF " three kinds of tactful singles and change
The time that seville orange flower is taken;Scape image whole for the SPOT-5 of 6000*6000 pixel, " CUDA+ memory ", " CUDA+MMF " and " CUDA+
The time that three kinds of piecemeal " tactful single iterations are spent.
Experimental result: as shown in table 1:
The execution time (unit: ms) of four kinds of implementation strategies of table 1K-Means clustering algorithm
As can be seen that the time that " CUDA+ memory " and " CUDA+MMF " strategy is spent is far smaller than " CPU+ memory " strategy
The time of cost, only 3% or so of the latter, wherein the time that " CUDA+MMF " strategy is spent is slightly above " CUDA+ memory " plan
The time slightly spent.
It can be seen that the time that " CUDA+ memory " and " CUDA+MMF " strategy is spent is essentially identical, and far smaller than
Time that " CUDA+ piecemeal " strategy is spent, also only 3% or so of the latter.It can be seen that MMF technology can actually be mentioned greatly
The time efficiency of high K-Means clustering algorithm.
Embodiment 3
Experiment purpose and method: in order to evaluate the application feasibility and validity, the present embodiment is with K-Means algorithm
Example, for 600*600,3000*3000 and 6000*6000 pixel image carry out " CUDA+ memory ", " CUDA+MMF " and
EMS memory occupation quantity is compared evaluation and test when " CUDA+ piecemeal " three kinds of strategies.
Experimental result:
As shown in table 2, for the small image of 600*600 pixel, what " CUDA+MMF " and " CUDA+ piecemeal " strategy occupied
Memory is slightly less than the memory that " CUDA+ memory " strategy occupies;For the median size image of 3000*3000 pixel, " CUDA+
The memory that MMF " strategy and " CUDA+ piecemeal " strategy occupy is significantly less than the memory that " CUDA+ memory " strategy occupies, only the latter
25% or so;For the big image of 6000*6000 pixel, the memory that " CUDA+MMF " and " CUDA+ piecemeal " strategy occupies is remote
Less than the memory that " CUDA+ memory " strategy occupies, only 10% or so of the latter.It can be seen that the size of remote sensing images is bigger,
The saving of amount of memory is more obvious using " CUDA+MMF " and " CUDA+ piecemeal " strategy of piecemeal processing method.
The amount of memory (unit: KB) that the various implementation strategies of table 2K-Means clustering algorithm occupy
Embodiment 4
Experiment purpose and method: in order to evaluate the application feasibility and validity, the present embodiment is with K-Means algorithm
Example has studied the influence that tile size executes the time to " CUDA+MMF " implementation strategy, has done 5 groups of comparative tests in total.It is each
The size of tile is respectively set as tetra- kinds of sizes of 128*128,256*256,512*512 and 1024*1024 in group experiment
Experimental result: as shown in table 3:
The execution time (unit: ms) of CUDA+MMF strategy under the various tile sizes of table 3
As can be seen that the execution time of " CUDA+MMF " strategy is on a declining curve as tile dimensions become larger, but decline
Amplitude very little, tile size is set as 512*512 pixel or 256*256 pixel under normal conditions.
The foregoing is only a preferred embodiment of the present invention, but scope of protection of the present invention is not limited thereto,
In the technical scope disclosed by the present invention, any changes or substitutions that can be easily thought of by anyone skilled in the art,
It should be covered by the protection scope of the present invention.
Claims (1)
1. high-resolution remote sensing image sorting algorithm spatiotemporal efficiency optimization method, which is characterized in that the optimization algorithm is specific
Step are as follows:
A: image data is divided into several block of pixels;
B: using the set of pixels of the Raster IO function block-by-block load image in GDAL function library, and Memory Mapping File is utilized
Technology block-by-block pixel data dumps in new disk file;
C: K pixel is randomly selected as initial cluster center;
D: the pixel data of first piecemeal is read from disk file using storage mapping file technique, and is transferred to equipment
End;
E: in each pixel that equipment end calculates piecemeal at a distance from each cluster centre, classify by phase approximately principle;
F: classification results are passed back host side, and result data is dumped into disk file using storage mapping file technique;
G: if the data read in step D are the last one block datas, step is gone to
Otherwise H reads next block data and goes to step E;
H: the average value of each pixel in every one kind is calculated, and using this average value as new cluster centre;
I: whether newer cluster centre is identical as old cluster centre, replaces old cluster with new cluster centre after comparing
Center, comparison result difference then goes to step D, identical, goes to step J;
J: the pixel value of each cluster centre is assigned to such each pixel;
The image data is divided into several block of pixels, and pixel block size is 128*128,256*256,512*512 and 1024*
One of 1024;
Absolute distance, Euclidean distance, mahalanobis distance, graceful can be used in the new cluster centre of the comparison and old cluster centre
Hatton's distance, included angle cosine distance metric formula carry out similarity evaluation.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201511003105.3A CN105528622B (en) | 2015-12-28 | 2015-12-28 | A kind of high-resolution remote sensing image sorting algorithm spatiotemporal efficiency optimization method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201511003105.3A CN105528622B (en) | 2015-12-28 | 2015-12-28 | A kind of high-resolution remote sensing image sorting algorithm spatiotemporal efficiency optimization method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN105528622A CN105528622A (en) | 2016-04-27 |
CN105528622B true CN105528622B (en) | 2019-03-05 |
Family
ID=55770836
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201511003105.3A Expired - Fee Related CN105528622B (en) | 2015-12-28 | 2015-12-28 | A kind of high-resolution remote sensing image sorting algorithm spatiotemporal efficiency optimization method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105528622B (en) |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109522852B (en) * | 2018-11-22 | 2023-01-24 | 中国科学院长春光学精密机械与物理研究所 | Artificial target detection method, device and equipment based on optical remote sensing image |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101241504A (en) * | 2008-01-23 | 2008-08-13 | 武汉大学 | Remote sense image data intelligent search method based on content |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
AU2012258421A1 (en) * | 2012-11-30 | 2014-06-19 | Canon Kabushiki Kaisha | Superpixel-based refinement of low-resolution foreground segmentation |
-
2015
- 2015-12-28 CN CN201511003105.3A patent/CN105528622B/en not_active Expired - Fee Related
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101241504A (en) * | 2008-01-23 | 2008-08-13 | 武汉大学 | Remote sense image data intelligent search method based on content |
Non-Patent Citations (2)
Title |
---|
"基于GDAL大于2G遥感图像的快速浏览";张宏伟 等;《计算机工程与应用》;20121231;论文第161页第3节、第4.1节 |
"基于超像素的面向对象遥感图像分类方法研究";吴洋;《中国优秀硕士学位论文全文数据库 信息科技辑》;20140615;论文第2.2.1节 |
Also Published As
Publication number | Publication date |
---|---|
CN105528622A (en) | 2016-04-27 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US9851917B2 (en) | Method for de-duplicating data and apparatus therefor | |
CN106486167B (en) | Improve the method and system that flash memory is removed | |
US20140351285A1 (en) | Platform and method for analyzing electric power system data | |
TW201447776A (en) | Performing object detection operations via a graphics processing unit | |
JP2017151953A (en) | Identifying lines of table | |
CN113836084A (en) | Data storage method, device and system | |
CN111309825A (en) | Data clustering storage method and device and computer readable storage medium | |
US20160026646A1 (en) | Recording medium having data recorded therein in data file format structure for visualization of large capacity cfd parallel data and method for generating said data file format structure | |
TWI709049B (en) | Random walk, cluster-based random walk method, device and equipment | |
US11822595B2 (en) | Incremental agglomerative clustering of digital images | |
CN103150260A (en) | Method and device for deleting repeating data | |
JP6674094B2 (en) | Region extraction program, region extraction method, and region extraction device | |
Kumar et al. | Efficient I/O and storage of adaptive-resolution data | |
CN103207889A (en) | Method for retrieving massive face images based on Hadoop | |
CN105589908A (en) | Association rule computing method for transaction set | |
CN104407982B (en) | A kind of SSD discs rubbish recovering method | |
WO2022007596A1 (en) | Image retrieval system, method and apparatus | |
CN105528622B (en) | A kind of high-resolution remote sensing image sorting algorithm spatiotemporal efficiency optimization method | |
CN111143146A (en) | Health state prediction method and system of storage device | |
JP2017073093A (en) | Index generation program, index generation device, index generation method, retrieval program, retrieval device and retrieval method | |
CN113971225A (en) | Image retrieval system, method and device | |
CN116129325A (en) | Urban treatment image target extraction method and device and application thereof | |
US11775515B2 (en) | Dataset optimization framework | |
US10817179B2 (en) | Electronic device and page merging method therefor | |
CN110262758B (en) | Data storage management method, system and related equipment |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20190305 Termination date: 20191228 |