CN104036286B - A kind of multi-spectral Images Classification processing method based on Hadoop - Google Patents
A kind of multi-spectral Images Classification processing method based on Hadoop Download PDFInfo
- Publication number
- CN104036286B CN104036286B CN201410201066.7A CN201410201066A CN104036286B CN 104036286 B CN104036286 B CN 104036286B CN 201410201066 A CN201410201066 A CN 201410201066A CN 104036286 B CN104036286 B CN 104036286B
- Authority
- CN
- China
- Prior art keywords
- task
- classification
- node
- jobs node
- jobs
- 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.)
- Active
Links
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The present invention relates to Classification in Remote Sensing Image and the crossing domain of computer distribution type computing technique, more particularly to a kind of multi-spectral Images Classification processing method based on Hadoop, its difference is that its method is comprised the following steps：Step a）, the structure of disaggregated model is carried out in single work child node first；Step b）And then disaggregated model is sent to each jobs node using the heartbeat message communication mechanism of Distributed Architecture, jobs node starts concurrently to carry out image blocks classification task after receiving disaggregated model file；Step c), by each piecemeal classification results reduction it is finally a complete classification results figure.The present invention greatly improved the classification speed when classifying to mass remote sensing image data.
Description
Technical field
It is based on the present invention relates to Classification in Remote Sensing Image and the crossing domain of computer distribution type computing technique, more particularly to one kind
The multi-spectral Images Classification processing method of Hadoop.
Background technology
With developing rapidly for remote sensing technology, remote sensing image be widely used in agricultural, forestry, geology, mineral products, the hydrology,
The aspects such as ocean, environmental monitoring, and classification of remote-sensing images also enters as the element task flow in remote sensing fields constantly
Step.In recent years, remote sensing image data amount is in explosive increase, forms GB grades, TB grades, PB grades of development trend.This gives remote sensing shadow
As sorting work tape has carried out huge pressure, how fast and accurately remote sensing image is categorized into order to remote sensing image should
With the problem for being badly in need of solving.
In a large amount of practical applications, classification work generally or using the artificial supervised classification method for extracting classification model or
The automatic semisupervised classification method for choosing sample, after being required for according to priori establishment disaggregated model, then to image classification
Treatment.For example, being based on SVM（SVMs, Support Vector Machine）Forest classification method is without manually gathering
Classification samples, realize machine and sample automatically the process of classification.Traditional make use of digital elevation based on SVM forest classification methods
Model (Digital Elevation Model), 250 meters of resolution images of Modis sensor productions, to Landsat satellites
30 meters of resolution images that TM, ETM+ sensor carried in series are produced carry out subsidiary classification, and SVM forest classification methods pass through
The complex calculation of various data selects accurate classification samples, and accurate classification samples enter SVM classifier successively, to SVM
Grader is trained, and until SVM classifier reaches enough niceties of grading, finally uses the SVM classifier pair obtained by training
Image classification.The characteristic of the automatic sample train classification of machine reduces the time of sample selection, saves manpower and materials.
But the construction of traditional svm classifier method be based on stand-alone environment, be limited to CPU computing capability and internal memory it is big
It is small, can only serial process remote sensing image file, it is still desirable to consume long time.And easily because machine in processing procedure
Device failure, causes program crashing, or even basic remote sensing image data to be also damaged, and the classification effectiveness of stand-alone environment has been difficult to completely
The demand of sufficient mass remote sensing image classification process task.
The content of the invention
In order to overcome above-mentioned weak point, the present invention to propose a kind of multi-spectral Images Classification treatment based on Hadoop
Method, the method greatly improved the classification speed when classifying to mass remote sensing image data.
The present invention solves the technical scheme that is used of above-mentioned technical problem：A kind of multispectral remote sensing shadow based on Hadoop
As classification processing method, its difference is that its method is comprised the following steps：
Step a）, the structure of disaggregated model is carried out in single work child node first；
Step b）And then disaggregated model is sent to each operation section using the heartbeat message communication mechanism of Distributed Architecture
Point, jobs node starts concurrently to carry out image blocks classification task after receiving disaggregated model file；
Step c), by each piecemeal classification results reduction it is finally a complete classification results figure.
The beneficial effects of the invention are as follows：The inventive method is improved on the basis of traditional remote sensing image sorting technique, profit
With distributed treatment and storage capacity, realize that classification of remote-sensing images task is one-to-many, multi-to-multi tupe, the present invention is proposed
Method ensure nicety of grading on the premise of, not only greatly improve the processing speed of classification of remote-sensing images, can more overcome biography
The issuable classification disruption of mechanical disorder in system method, present invention compatibility remote sensing fields majority supervised classification or semi-supervised
Sorting technique, realizes the high efficiency of Remote Image Classification in the form of a kind of " total-point-total ".
The present invention improves the speed of mass remote sensing image classification using the data distribution formula concurrent processing storage capacity of PC clusters
Degree, the treatment of conventional individual environment classification is easily because the reasons such as mechanical disorder may cause tasks interrupt, even if PC clusters in the present invention
Classification task still can efficiently be given middle Partial Jobs node failure paralysis, task management center other operation sections in order
Point treatment.A small amount of time is spent to global image collecting sample train classification models in view of classification of remote-sensing images requirements of process,
Be arranged in disaggregated model training process on single jobs node by the present invention, and jobs node and task pipe are set up using heartbeat mechanism
Information exchange between reason center, JobClient, then disaggregated model is distributed to other jobs nodes, realize many jobs nodes
Parallel sorting, under the premise for ensureing nicety of grading, classification speed is substantially accelerated.
Brief description of the drawings
Fig. 1 is the multi-spectral Images Classification processing method schematic diagram based on Hadoop；
Fig. 2 is single node svm classifier model extraction schematic flow sheet；
Fig. 3 is distributed image classification handling process schematic diagram；
Fig. 4 is the multi-spectral Images Classification processing method DFD based on Hadoop；
Fig. 5 is the multi-spectral Images Classification treatment PC aggregated structure figures based on Hadoop.
Specific embodiment
In order to realize above technical scheme, the present invention needs to solve problem in detail below：How to be instructed in Distributed Architecture
How the unified disaggregated model of white silk, jobs node transmits homogeneous classification model, the task partition problem of remote sensing image, training classification
Device and carried out in computer cluster using how concurrent grader is to image classification the two series processes, how in framework
Middle transmission tasks information etc..In the inventive method, work child node is also jobs node, performs the jobs node quilt of reduction task
Referred to as reduction task node.
Referring to Fig. 1-Fig. 5, present invention method illustrates how reality by by taking the forest classification method based on SVM as an example
Existing distribution image classification.Generally speaking, the embodiment of the present invention is based on the multi-spectral Images Classification processing method of Hadoop
Comprise the following steps：
Step a）, the structure of disaggregated model is carried out in single work child node first；
Step b）And then disaggregated model is sent to each operation section using the heartbeat message communication mechanism of Distributed Architecture
Point, jobs node starts concurrently to carry out image blocks classification task after receiving disaggregated model file；
Step c), by each piecemeal classification results reduction it is finally a complete classification results figure.
Preferably, the step a）Including step in detail below：
Step a1）, by image to be sorted and classification incoming any one jobs node of related data（TaskTracker）,
Jobs node is every at regular intervals will be to task management center（JobTracker）Heartbeat message is sent, in task management
The information the such as whether heart feedback working condition of jobs node, data fresh information, jobs node normal；
Step a2）, task management center persistently receive the heartbeat message that jobs node is transmitted, judge the number in jobs node
According to whether having renewal；If certain jobs node data has renewal, task management center will replicate a disaggregated model and extract journey
Sequence bag gives the jobs node, and indicates the jobs node to extract disaggregated model to new image data；At the same time in engineering
The heart（JobClient）Message is sent, engineering center is informed（JobClient）New data is divided；
Step a3）, jobs node receive disaggregated model extraction procedure bag, locally decompression is simultaneously to start a Java Virtual Machine
Disaggregated model is performed to new image to extract；Engineering center（JobClient）Jobs node data new information is received, to new shadow
As carrying out logic partitioning, and the storage of logic partitioning file；
Step a4）, jobs node complete disaggregated model extract, task management center is informed by heartbeat message.
Preferably, the step b）Including step in detail below：
B1), task management center receives the heartbeat message that jobs node sends, and then judges engineering center
（JobClient）Whether the division work of the image has been completed；When disaggregated model extraction work and image division work are all complete
Cheng Liao, task management center starts that new image is divided into multiple classification block tasks according to logical partitioning（Task）, and it is responsible for tune
Spend each subtask to be run on jobs node, task management center is dispatched by sending heartbeat return value to jobs node
Distribution task, heartbeat return value contains task-performance instructions, task Jar files, disaggregated model file and data to be sorted institute
In positional information；
B2), jobs node receives the heartbeat return value of task management center return, judges whether oneself is allocated and appoints
Business, if being assigned to task, the task Jar files localization for just passing heartbeat return value back；Meanwhile, according in heartbeat return value
Data positional information to be sorted, the jobs node voluntarily from other jobs nodes storage data in capture all references data
Local working folder is copied directly into, a new Java Virtual Machine is then created and is run each image block sort
Task.
Preferably, the step c）Including step in detail below：
Step c1）, task management center receive the jobs node message that finishes of classification, in certain idle job node wound
Build reduction（Reduce）Task, and inform all jobs node positional informations for having completed classification task of reduction task node；
Step c2）, reduction task node receive complete classification task jobs node positional information, by the information from
The jobs node for completing classification task transfers classification results, and is complete grouped data by the classification results reduction of the image；
Step c3）, reduction task node complete reduction task, store grouped data, and by heartbeat communicate will storage letter
Breath submits to task management center.
Specifically, Fig. 1 is the multispectral image classification processing method schematic diagram based on Hadoop, image to be sorted is set up and is divided
The task of class model is completed on single jobs node, single width image global classification model is generated, according to disaggregated model to image
Being classified for task realizes that image is divided into the image blocks that logic size is 64M, and distribution is adjusted in Distributed Calculation cluster
Each jobs node parallel sorting is spent, the time required to can effectively reducing classification.
Specifically, Fig. 2 is the multi-spectral Images Classification processing method PC aggregated structure figures based on Hadoop, whole frame
Structure is by data switching center, task management center（JobTracker）, engineering center（JobClient）With N number of jobs node group
Into.
Data switching center connects task management center, engineering center and All Jobs node, wantonly two node in framework（Appoint
Business administrative center and engineering center fall within architectural node）Can be transferred through data switching center to be interconnected, data switching center
It is made up of a number of network switch.
Task management center（JobTracker）It is responsible for each subtask of scheduling to operate on jobs node, and supervises in real time
Control jobs node working condition, if it find that it is just reruned in the subtask for having failure, typically disposes task management center
On a separate machine.
Engineering center（JobClient）Logical partitioning is carried out to image, image is divided into acquiescence the image of some 64M
Reference program and configuration parameter can be packaged into Java files and stored by block, each task in engineering center, and handle
Task management center is submitted in path, and engineering center is typically located at user terminal.
Jobs node（TaskTracker）It is the fundamental unit of computing and storage in framework, remotely-sensed data can be by any
One jobs node enters（Also can be entered by engineering center）, jobs node is responsible for whole scape remotely-sensed data disaggregated model training, with
And image blocks classification task.Jobs node can have N number of in framework（N≥2）, for mass remote sensing data classification, operation
Number of nodes is proportional with classification speed.
Between task management center and jobs node, all " heartbeat mechanism " is passed through between engineering center and task management center
It is in communication with each other, heartbeat mechanism is a simple circulation, periodically sends heartbeat to recipient.Heartbeat message is sent, heartbeat is told
Whether recipient, jobs node survives, and functions simultaneously as message channel between the two.Heartbeat recipient receives newest heartbeat
Message, corresponding action command is made by heartbeat return value.
Specifically, Fig. 3 is by taking the forest classification method based on SVM as an example, after illustrating remotely-sensed data input jobs node, make
The flow of industry Node extraction image classification model.Remote sensing image can be entered by any jobs node, and jobs node passes through heartbeat
Submit the message of data renewal to task management center.Task management center is received after heartbeat message, records jobs node
Data fresh information, is communicated by heartbeat return value with jobs node, is indicated jobs node to carry out disaggregated model to new data and is carried
Take.
Jobs node receives the instruction for extracting disaggregated model, will localize the jar file of operation from shared-file system
Jar file is unziped to local working directory by the file system where copy job node, jobs node, starts a Java void
Plan machine moving model extraction procedure.Jobs node completes to submit to model storage address by heartbeat after disaggregated model is extracted appoints
Business administrative center.
Specifically, Fig. 4 illustrates to realize the method that remote sensing image is efficiently classified in Distributed Architecture.Task management center
Receive after the heartbeat message that jobs node is sent, the information of jobs node data renewal is recorded, except to jobs node
Heartbeat return value is sent, also heartbeat message can be sent to engineering center, heartbeat message have recorded the remotely-sensed data of new typing each
The storage information of jobs node；Engineering center receives the heartbeat that task management center is sent, and obtains remotely-sensed data to be sorted and deposits
Data are carried out acquiescence division by storage address, size of data K, are divided into Z image blocks, except last block size is S, Suo Youying
As block size is 64M.Relation between image block size and complete image size of data meets below equation：K = 64×Z
+ S.After engineering center completes to divide, by the storage of image division information, and task management center is informed by heartbeat return value
The division work of image to be sorted is completed.After task management center confirms that image division is completed, dividing for the image is inquired about
Whether class model extracts finishes.If image is divided and disaggregated model has one without completion, task management center will continue to supervise
Listen heartbeat；If image is divided and corresponding disaggregated model has all been completed, task management center starts the classification task of the image.
Using JobTracker in Hadoop（Task management center）Job scheduler, transfer to operation to adjust classification task
Degree device is scheduled, and it is initialized.A manipulating object for representing classification task is created, it is encapsulated and record letter
Breath, for tracking task conditions and process.Job scheduler obtains image division information, then for each image blocks creates one
Individual subtask（Map）, it is that view picture image creates a reduction task（Reduce）.
Jobs node periodically sends heartbeat to task management center, and task management center can be normal idle job node
One task of distribution, is communicated by heartbeat return value with jobs node.
Jobs node is allocated after task, the text where the jar file for localizing operation is copied into the jobs node
In part system.Meanwhile, by all files required for sort program（Including the image classification model）Copy to local disk.It
Afterwards, the content in jar file is unziped into local working directory, creates a new Java Virtual Machine to run each classification
Task.
Jobs node is completed after classification subtask, and classification results are stored, and sorting result information is submitted to by heartbeat, is appointed
Business administrative center is recorded.
Task management center can also distribute the reduction of the image while distribution classification subtask to a jobs node
Task.Reduction task node obtains the corresponding sorting result information of task management center image by heartbeat mechanism, and according to reality
The sorting result information of Shi Gengxin pulls classification results data from other on jobs node of those execution classification subtasks.
Reduction task node often pulls a classification results data, just classification results is stitched together, until view picture
The splicing of classification results figure is finished, the complete classification image of output, is stored in local directory, and submit to task complete to task management center
Into message.
Fig. 5 illustrates change and the place of remotely-sensed data stream in the multi-spectral Images Classification processing method based on Hadoop
Reason, remotely-sensed data is entered by any jobs node, and jobs node goes out corresponding disaggregated model to Extraction of Image.Engineering center is to shadow
As being divided, image division information is obtained.PC clusters utilize corresponding disaggregated model, parallel computation, by each image block sort,
Last reduction is complete classification results.
Example of the invention is realized under PC cluster environment, the experiment proved that, in treatment magnanimity multi-spectrum remote sensing image point
During class, the present invention improves operating efficiency on the premise of guarantee is suitable with the precision of conventional sorting methods acquired results, improves
Conventional sorting methods automaticity, can be used in mass remote sensing data Fast Classification, such as Landsat, Modis, Alos,
Cerbus, resource series satellite etc..
Compared to traditional remote sensing image sorting technique, the present invention substitutes unit processing environment with PC clusters processing environment, i.e.,
The Partial Jobs nodes break down in PC clusters is set to be stopped, task can be also assigned on other normal jobs nodes
Treatment, reduces the issuable risk of mechanical disorder.This brand-new mass remote sensing data tupe is with treatment image
The increase of data volume, can be further with advantage of the traditional remote sensing image processing method in efficiency it is obvious, and with suitable
Flexibility, can be embedded in the processing procedure and method of all " after first model classify " forms, such as image supervised classification, based on nerve
Network class method etc..
Remote Image Classification of the invention can select based on SVMs sorting technique, based on decision tree
Sorting technique, the sorting technique based on artificial neural network etc., comparatively these automatically select sample monitoring sorting technique, can
The characteristics of to realize computer intelligence sampling train classification models, a large amount of manpower things of artificial sampling process consumption can be saved
Power, is more suitable for mass remote sensing data classification, and bigger advantage can have been given play to reference to the present invention.
Claims (3)
1. a kind of multi-spectral Images Classification processing method based on Hadoop, it is characterised in that its method includes following step
Suddenly：
Step a）, the structure of disaggregated model is carried out in single work child node first；
Step b）And then disaggregated model is sent to each jobs node using the heartbeat message communication mechanism of Distributed Architecture,
Jobs node starts concurrently to carry out image blocks classification task after receiving disaggregated model file；
Step c), by each piecemeal classification results reduction it is finally a complete classification results figure；
Wherein, the step a）Including step in detail below：
Step a1）, by image to be sorted and classification incoming any one jobs node of related data, jobs node is every certain
Time will send heartbeat message to task management center, and working condition, the data of jobs node are fed back to task management center
Fresh information, jobs node whether normal information；
Step a2）, task management center persistently receive the heartbeat message that jobs node is transmitted, judge the data in jobs node
Whether renewal is had；If certain jobs node data has renewal, task management center will replicate a disaggregated model and extract
Program bag gives the jobs node, and indicates the jobs node to extract disaggregated model to new image data；At the same time to engineering
Center sends message, informs that engineering center divides to new data；
Step a3）, jobs node receive disaggregated model extraction procedure bag, locally decompression is simultaneously right to start a Java virtual machine
New image performs disaggregated model and extracts；Engineering center receives jobs node data new information, and logic point is carried out to new image
Block, and the storage of logic partitioning file；
Step a4）, jobs node complete disaggregated model extract, task management center is informed by heartbeat message；
Wherein, data switching center's connection task management center, engineering center and All Jobs node, appoint in data switching center
Two nodes can be transferred through data switching center and be interconnected, and data switching center is made up of a number of network switch；
Task management center is responsible for each subtask of scheduling and is operated on jobs node, and monitor in real time jobs node work shape
Condition, if it find that it is just reruned in the subtask for having failure, task management center deployment on a separate machine；
Engineering center carries out logical partitioning to image, and image is divided into acquiescence the image blocks of some 64M, each task
Reference program and configuration parameter can be packaged into Java files in engineering center to store, and task is submitted in path
Administrative center, engineering center is arranged in user terminal；
Jobs node is the fundamental unit of computing and storage in framework, and remotely-sensed data can be entered by any one jobs node,
Jobs node is responsible for whole scape remotely-sensed data disaggregated model training, and image blocks classification task, and jobs node can have N in framework
It is individual, wherein, N >=2, jobs node quantity is proportional with classification speed；
Between task management center and jobs node, heartbeat mechanism phase intercommunication is all passed through between engineering center and task management center
Letter, heartbeat mechanism is a simple circulation, periodically sends heartbeat to recipient, sends heartbeat message, tells heartbeat recipient,
Whether jobs node survives, and functions simultaneously as message channel between the two, and heartbeat recipient receives newest heartbeat message, leads to
Mistake heart is jumped return value and makes corresponding action command.
2. the multi-spectral Images Classification processing method based on Hadoop as described in claim 1, it is characterised in that institute
State step b）Including step in detail below：
B1), task management center receives the heartbeat message that jobs node sends, and then judges whether engineering center has completed this
The division work of image；When disaggregated model extraction work and image division work are completed, task management center starts basis
New image is divided into multiple classification block tasks by logical partitioning, and responsible each subtask of dispatching is run on jobs node,
Task management center sends heartbeat return value come dispatching distribution task by jobs node, and heartbeat return value includes tasks carrying
Instruction, task Jar files, disaggregated model file and data position to be sorted information；
B2), jobs node receives the heartbeat return value of task management center return, judges whether oneself is allocated task, if
Task is assigned to, the task Jar files localization for just passing heartbeat return value back；Meanwhile, treating in heartbeat return value
Grouped data positional information, the jobs node voluntarily from other jobs nodes storage data in capture all references data by its
Local working folder is copied directly into, a new Java virtual machine is then created and is run each image block sort times
Business.
3. the multi-spectral Images Classification processing method based on Hadoop as described in claim 1, it is characterised in that institute
State step c）Including step in detail below：
Step c1）, task management center receive the jobs node message that finishes of classification, created in certain idle job node and returned
About task, and inform all jobs node positional informations for having completed classification task of reduction task node；
Step c2）, reduction task node receive complete classification task jobs node positional information, by the information from completion
The jobs node of classification task transfers classification results, and is complete grouped data by the classification results reduction of the image；
Step c3）, reduction task node complete reduction task, store grouped data, and by heartbeat communicate storage information is carried
Give task management center.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410201066.7A CN104036286B (en) | 2014-05-14 | 2014-05-14 | A kind of multi-spectral Images Classification processing method based on Hadoop |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410201066.7A CN104036286B (en) | 2014-05-14 | 2014-05-14 | A kind of multi-spectral Images Classification processing method based on Hadoop |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104036286A CN104036286A (en) | 2014-09-10 |
| CN104036286B true CN104036286B (en) | 2017-06-30 |
Family
ID=51467051
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410201066.7A Active CN104036286B (en) | 2014-05-14 | 2014-05-14 | A kind of multi-spectral Images Classification processing method based on Hadoop |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104036286B (en) |
Families Citing this family (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105096334B (en) * | 2015-09-16 | 2018-02-09 | 中国科学院地理科学与资源研究所 | A kind of mining area monitoring method and system |
| CN105491108A (en) * | 2015-11-19 | 2016-04-13 | 浪潮集团有限公司 | System and method for processing remote sensing images |
| US10860461B2 (en) * | 2017-01-24 | 2020-12-08 | Transform Sr Brands Llc | Performance utilities for mobile applications |
| CN106991656B (en) * | 2017-03-17 | 2019-09-06 | 杭州电子科技大学 | A kind of mass remote sensing image distribution geometric correction system and method |
| CN107358260B (en) * | 2017-07-13 | 2020-09-29 | 西安电子科技大学 | Multispectral image classification method based on surface wave CNN |
| CN111092943B (en) * | 2019-12-13 | 2022-09-20 | 中国科学院深圳先进技术研究院 | Multi-cluster remote sensing method and system of tree structure and electronic equipment |
| CN112070062A (en) * | 2020-09-23 | 2020-12-11 | 南京工业职业技术大学 | Hadoop-based crop waterlogging image classification detection and implementation method |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102790706A (en) * | 2012-07-27 | 2012-11-21 | 福建富士通信息软件有限公司 | Safety analyzing method and device of mass events |
-
2014
- 2014-05-14 CN CN201410201066.7A patent/CN104036286B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102790706A (en) * | 2012-07-27 | 2012-11-21 | 福建富士通信息软件有限公司 | Safety analyzing method and device of mass events |
Non-Patent Citations (2)
| Title |
|---|
| 《Large-scale image classification : Fast feature extraction and SVM training》;Yuanqing Lin 等;《Computer Vision and Pattern Recognition（CVPR），2011 IEEE Conference on》;20110625;第1689-1696页 * |
| 《基于Hadoop的海量图像数据管理》;李波;《中国优秀硕士学位论文全文数据库》;20111015;第15-48页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104036286A (en) | 2014-09-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104036286B (en) | A kind of multi-spectral Images Classification processing method based on Hadoop | |
| CN109948428A (en) | The GPU cluster deep learning edge calculations system of facing sensing information processing | |
| CN104965757B (en) | Method, virtual machine (vm) migration managing device and the system of live migration of virtual machine | |
| CN101957863B (en) | Data parallel processing method, device and system | |
| CN104036451B (en) | Model method for parallel processing and device based on multi-graphics processor | |
| Varghese et al. | A survey on edge performance benchmarking | |
| CN107229695A (en) | Multi-platform aviation electronics big data system and method | |
| CN109918184A (en) | Picture processing system, method and relevant apparatus and equipment | |
| CN110347636B (en) | Data execution body and data processing method thereof | |
| DE102019103310A1 (en) | ESTIMATE FOR AN OPTIMAL OPERATING POINT FOR HARDWARE WORKING WITH A RESTRICTION ON THE SHARED PERFORMANCE / HEAT | |
| CN110308984A (en) | It is a kind of for handle geographically distributed data across cluster computing system | |
| CN103116525A (en) | Map reduce computing method under internet environment | |
| CN108628675A (en) | A kind of data processing method, device, equipment and computer readable storage medium | |
| CN112379869A (en) | Standardized development training platform | |
| CN108197656A (en) | A kind of attribute reduction method based on CUDA | |
| CN108762735A (en) | Management method and device, storage medium, the terminal of workflow engine | |
| CN104063230B (en) | The parallel reduction method of rough set based on MapReduce, apparatus and system | |
| CN107797852A (en) | The processing unit and processing method of data iteration | |
| CN110891083B (en) | Agent method for supporting multi-job parallel execution in Gaia | |
| Shahoud et al. | A meta learning approach for automating model selection in big data environments using microservice and container virtualization technologies | |
| CN113010296A (en) | Task analysis and resource allocation method and system based on formalized model | |
| CN103927406B (en) | System and aircraft for driving the aircraft with performance function server | |
| Martínez-Castaño et al. | Building python-based topologies for massive processing of social media data in real time | |
| Kashansky et al. | Monte Carlo Approach to the Computational Capacities Analysis of the Computing Continuum | |
| Krishnan | A Web-Based Software Platform for Data Processing Work Ows and Its Applications in Aerial Data Analysis |
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 |