CN109246095A - A kind of communication data coding method suitable for deep learning - Google Patents
A kind of communication data coding method suitable for deep learning Download PDFInfo
- Publication number
- CN109246095A CN109246095A CN201810996375.6A CN201810996375A CN109246095A CN 109246095 A CN109246095 A CN 109246095A CN 201810996375 A CN201810996375 A CN 201810996375A CN 109246095 A CN109246095 A CN 109246095A
- Authority
- CN
- China
- Prior art keywords
- communication
- data
- field
- deep learning
- feature vector
- 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.)
- Granted
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L69/00—Network arrangements, protocols or services independent of the application payload and not provided for in the other groups of this subclass
- H04L69/18—Multiprotocol handlers, e.g. single devices capable of handling multiple protocols
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
- H04L41/02—Standardisation; Integration
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L69/00—Network arrangements, protocols or services independent of the application payload and not provided for in the other groups of this subclass
- H04L69/06—Notations for structuring of protocol data, e.g. abstract syntax notation one [ASN.1]
Landscapes
- Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Computer Security & Cryptography (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Data Exchanges In Wide-Area Networks (AREA)
Abstract
The present invention discloses a kind of communication data coding method suitable for deep learning, convection current grade communication data is pre-processed, IP communication pair is aggregated into, then internal Flow record is communicated to IP and is ranked up, and extracts the field for needing to encode and corresponding attribute value sequence;Each attribute value sequence is encoded based on pyramid pond method, is spliced into original feature vector;The original feature vector of formation is normalized, and is applied in deep neural network and is verified.The present invention can be such that this kind of irregular hetero-com-munication data can be applied in the scene of deep learning automatically by the Flow sequential coding of any length at the original feature vector of fixed length;In the case where not needing expertise intervention, it can be sufficiently reserved characteristic of the communication behavior in terms of time and space, to enable the data after coding sufficiently to represent the communication feature between node, to be applied in all kinds of communication behavior analysis tasks.
Description
Technical field
It is specially a kind of suitable for the logical of deep learning the present invention relates to network communication detection and depth learning technology field
Letter data coding method.
Background technique
Deep learning method is considered to have the cognitive ability and data Extracting Ability of height, is widely used in computer
The fields such as vision, image procossing, speech recognition.At the same time, deep learning is also widely used in network communication row in recent years
For context of detection.However, deep learning algorithm has strict requirements, network communication analysis field and image to the input of data
The data characteristics of analysis field have certain difference.In the research of existing network communication detection field, according to depth
These work, can be divided into the research method based on feature and the research based on initial data by the difference for practising algorithm input type
Method.
(1) based on the research method of feature
Research method based on feature is directly using existing, predefined communication feature as input, and deep learning
The role that algorithm is intended only as classifier exists.Certain methods using famous KDD-99 data set or NSL-KDD data set come
Training detection model, these data sets provide predefined 41 features, therefore this kind of research method provides data set
41 dimensional features are used to training and detect and model directly as the input of deep learning algorithm.In addition, will be based on there are also other researchs
Input of the feature of expertise design as deep learning algorithm, such as different attribute (such as data packet, byte number, session number)
Statistical data (such as sum, max, min, mean value, variance, deviation, comentropy).
(2) based on the research method of initial data
Research method based on initial data mainly uses deep learning algorithm to learn automatically from original network traffic data
Practising communication feature indicates, so as to avoid Feature Engineering.Certain methods from a large amount of unlabelled original network traffic datas from
Learn effective character representation dynamicly, in this type of method, header information and part payload in network session it is original
Data are by the input data as deep learning algorithm, for extracting network communication feature.Other methods are by network flow
The initial data of top n byte is expressed as image, is then based on the methods of CNN, AutoEncoder, DBN and carries out mark sheet dendrography
It practises.
Existing research is primarily upon the Payload data of the feature or network flow extracted in advance, and less focuses on
The communication datas such as NetFlow, NetStream, IPFIX.Research method based on feature is directly constructed using based on expertise
Communication feature as input, and underuse the Automatic signature extraction ability and advantage of deep learning.Based on initial data
Research method be typically concerned with flow load, rather than the communication data of isomery.It is this kind of for payload length inconsistence problems
Method mainly takes simple method for cutting (only remaining top n byte).However, under large-scale network traffic application scenarios
The fields such as Network anomaly detection, network application identification, user's behaviors analysis, network flow data is usually to obtain and store into
The formats such as this lower, scalability preferable NetFlow, NetStream, IPFIX indicate.However, hetero-com-munication data have not
The characteristics such as same scale, different scales not can be used directly in deep learning scene, in terms of being in particular in following two: 1) former
The communication data of beginning has different scale, and in field of image recognition, the pixel data as the input of deep learning algorithm has
Identical dimension, and a netflow record is made of the attribute of different scale, such as number of giving out a contract for a project, byte number, duration category
Property, their dimension is different or even data type is also variant;2) original structured communication data have scale difference
Property, in field of image recognition, the image data as deep learning training sample is of the same size, and different communication nodes
Between NetFlow sequence, different scales is all shown as over time and space, such as NetFlow sequence between different communication node
Length it is different, even if NetFlow sequence length is identical, total communication lasts duration is also not necessarily the same.Therefore, original structure
Changing communication data can not be applied directly as the input of deep learning algorithm.
Summary of the invention
In view of the above-mentioned problems, the purpose of the present invention is to provide a kind of passage data encoding sides suitable for deep learning
Method realizes the coding to hetero-com-munication data on the basis of retaining communication space-time characterisation as far as possible.This method can solve well
Certainly Automatic signature extraction problem of the deep learning in data on flows, while can also be identified for user's behaviors analysis, network application
Deep reference function is brought Deng other related communication behavioural analysis fields.Technical solution is as follows:
A kind of communication data coding method suitable for deep learning, comprising the following steps:
Step 1: the stream grade communication data of various formats being pre-processed, specification is melted into specific Flow format, retains
The specific fields that information extraction and coding need;
Step 2: the Flow data f with identical sources IP and destination IP being aggregated in a set, communication pair: NF is formed
(x → y)=< f1,f2,…,fn>, the internal all Flow data of a communication indicate all communications between communication node x and y
Record;
Step 3: being ranked up at the beginning of internal Flow data record will be communicated by stream, according to the stream of front and back two
The difference of time started calculates time interval, increases time interval field newly, forms new communication to NF ' (x → y);
Step 4: according to the field for including in every Flow data, by the Flow sequence resolving in NF ' (x → y) at correspondence
Attribute value sequence, the corresponding attribute value sequence AVS of each field;
Step 5: utilizing spatial pyramid pond method, each attribute value sequence AVS is encoded, it is fixed to be encoded into
Long data;
Step 6: after to each attribute value sequence AVS coding, being spliced, form original feature vector FV;
Step 7: original feature vector FV being normalized, unified original feature vector FV ' is generated, as to any
The coded data of the stream records series of length.
Further, the calculating process of data encoding is carried out in the step 5 based on spatial pyramid pond method are as follows:
Step 51: each attribute value sequence AVS is successively averagely divided into 1,2,4,8 ..., 2L-1Block, L indicate pyramid
The number of plies;
Step 52: calculating separately average, sum and the mode of every block number evidence, extract 3 values in from every block number;
Step 53: these values being spliced, each attribute value sequence AVS will generate 3* (2L- 1) a value.
Further, after the step 2 further include: filter out Flow record number less than 2L-1Communication pair.
Further, when being encoded to the Flow data for communicating internal indefinite length, while retaining between communication node
The spatial character and time response of communication behavior;The field of statistical conditions and distribution situation for corresponding attribute value carries out
Further selection, source port, destination port, byte number, packet digital section are added in field to be encoded;
The field for the trend that attribute value corresponding for fields related and all kinds of to call duration time changes over time carries out
It is further processed: internal Flow record will be communicated and be ranked up from small to large by time field, according to the stream of front and back two
The difference of time calculates time interval, while time interval and Duration field being added in field to be encoded.
Further, original feature vector FV is normalized using the standardization side Z-score in the step 7
Method calculates function are as follows:
In formula, x is original feature vector initial value, x*For original feature vector standard value;μ is mean value, and σ is standard deviation.
The beneficial effects of the present invention are: the present invention can be automatically by the Flow sequential coding of any length at the original of fixed length
Feature vector can be applied to this kind of irregular hetero-com-munication data in the scene of deep learning;Know not needing expert
In the case where knowing intervention, it can be sufficiently reserved characteristic of the communication behavior in terms of time and space, to make the data after coding
The communication feature between node can be sufficiently represented, to be applied in all kinds of communication behavior analysis tasks.
Detailed description of the invention
Fig. 1 is that communication data encodes flow chart in the present invention.
Fig. 2 is the data-encoding scheme schematic diagram based on pyramid pond thought in the present invention.
Fig. 3 is the feature extraction flow diagram based on CNN in the present invention.
Specific embodiment
The present invention is described in further details in the following with reference to the drawings and specific embodiments.The method of the present invention is being embodied
When process as shown in Figure 1, mainly comprising the steps that
A. convection current grade communication data is pre-processed, and aggregates into IP communication pair;
B. internal Flow record is communicated to IP to be ranked up, and extract the field for needing to encode and corresponding attribute value
Sequence;
C. each attribute value sequence is encoded based on pyramid pond method, is spliced into original feature vector;
D. the original feature vector of formation is normalized, and is applied in deep neural network and is verified.
The stream grade communication data of various formats in the step A includes the xFlow such as NstFlow, NetStream, IPFIX
Type, this kind of data provide the stream grade view of data communication between node, include the data packet system in one section of call duration time
Count information.Every kind of xFlow type both provides customized format and agreement, in order to carry out unification to this kind of stream grade communication data
Processing, the field that the present invention is jointly comprised for the attribute and this kind of agreement paid close attention in communication behavior analysis, extracts unification
Flow format.The field type of the Flow format includes source IP (sip), destination IP (dip), source port (sport), purpose
Port (dport), agreement (prot), byte number (bytes), packet number (pkts), duration (duration), stream time started
(time), time interval (interval) etc..In order to analyze the communication behavior between node, the present invention is with source IP and mesh
IP be key, Flow data are polymerize, the Flow with identical sources IP and destination IP will be gathered in a set, be made
It is a communication to handling.Main target of the invention be exactly to the internal Flow data (indefinite length) of communication into
Row coding, makes that it is suitable for deep learning algorithms.In step A operating process it should be noted that in order in step C
Piecemeal processing is carried out to attribute value sequence, needs filtration fraction Flow record number to be not up to desired communication pair in step, such as
In order to be divided into 8 parts to attribute value sequence, then require that communicating internal Flow record number has to be larger than or be equal to 8,
Therefore, Flow records number less than those of 8 communications to will be filtered.
In the step B, in order to be encoded to the internal Flow data (indefinite length) of communication, retain simultaneously
The space time statistical properties of communication behavior between communication node, the present invention are targetedly grasped to internal Flow sequence is communicated
Make.In terms of operation content mainly includes following two:
1) retain the spatial character of communication behavior: the spatial character of communication behavior is mainly reflected in the corresponding category of various fields
Property value statistical conditions and distribution situation, this patent carried out further selection, including source for the field of this kind of concern
These fields are added to word to be encoded by mouth (sport), destination port (dport), byte number (bytes), packet number (pkts) etc.
Section;
2) retain the time response of communication behavior: the time response of communication behavior is mainly reflected in relevant to call duration time
The trend that some characteristics and the corresponding attribute value of above-mentioned all kinds of fields change over time, therefore this patent will communicate internally
Flow record is ranked up from small to large by time field, is calculated time interval according to the difference of the time of two stream in front and back, is increased newly
Time interval field (interval), the interval attribute value of first stream is according to the interval attribute value of Article 2 stream
It is configured.Time interval (interval) and duration (duration) field are added in field to be encoded simultaneously.
After code field has been determined based on the above operation, then by Flow sequence resolving at corresponding attribute value sequence, including
It source port (sport), destination port (dport), byte number (bytes), packet number (pkts), time interval (interval) and holds
Attributes value sequences such as continuous time (duration), enabling A is set of fields to be encoded, then | A | indicate Field Count to be encoded, this
It will affect the length for finally encoding obtained original feature vector.
In the step C, according to pyramid number of plies L, each attribute value sequence will be divided into 1 respectively, 2,4,
8、……、2L-1Then block carries out the data pick-up of different scale, extraction process is to calculate the statistical property of each data block, takes out
The principle taken is as shown in Figure 2.For all kinds of statistical properties and variation tendency of reserved property value sequence, the present invention is from average spy
Property, total flow characteristic and the aspect of characteristic three is concentrated to extract the statistical information in data block, calculate separately every block number evidence
Average, sum and mode, to extract 3 values in from every block number.Then, these values are spliced, then one
Attribute value sequence AVS will generate 3* (1+2+4+ ...)=3* (2L- 1) a primitive character, a communication will be to will generate 3* | A | *
(2L- 1) a primitive character.
In the step D, we are 3* to the length that coding obtains | A | * (2L- 1) original feature vector is returned
One changes, and common method for normalizing includes the methods of min-max standardization, Z-score standardization.In view of under real network environment
The complicated multiplicity of communication data, the difference of each communication pair is larger, and the present invention uses Z-score standardized method.Such methods base
Data are normalized in mean μ and standard deviation sigma, calculate function are as follows:
Based on the original feature vector that step A~D is encoded, it can be applied directly in deep learning algorithm and be answered
With.When being applied to deep neural network (DNN), the one-dimensional original feature vector spliced can be inputted directly as Input
Into neural network;When being applied to convolutional neural networks (CNN), then need one-dimensional original feature vector being converted into two-dimentional shape
Formula, as shown in formula 2:
W × h=3 × | A | × (2L-1) (2)
Wherein w and h respectively indicates the width and height of two-dimensional feature vector, and initial characteristics vector is applied to convolutional neural networks
Schematic diagram is as shown in Figure 3.
Claims (5)
1. a kind of communication data coding method suitable for deep learning, which comprises the following steps:
Step 1: the stream grade communication data of various formats being pre-processed, specification is melted into specific Flow format, retains information
Extract and encode the specific fields needed;
Step 2: the Flow data f with identical sources IP and destination IP is aggregated in a set, formation communication pair: NF (x →
Y)=< f1,f2,…,fn>, the internal all Flow data of a communication indicate all communications note between communication node x and y
Record;
Step 3: being ranked up at the beginning of internal Flow data record will be communicated by stream, according to the beginning of two stream in front and back
The difference of time calculates time interval, increases time interval field newly, forms new communication to NF ' (x → y);
Step 4: according to the field for including in every Flow data, by the Flow sequence resolving in NF ' (x → y) at corresponding category
Property value sequence, the corresponding attribute value sequence AVS of each field;
Step 5: utilizing spatial pyramid pond method, each attribute value sequence AVS is encoded, fixed length number is encoded into
According to;
Step 6: after to each attribute value sequence AVS coding, being spliced, form original feature vector FV;
Step 7: original feature vector FV being normalized, unified original feature vector FV ' is generated, as to any length
Stream records series coded data.
2. the communication data coding method according to claim 1 suitable for deep learning, which is characterized in that the step
The calculating process of data encoding is carried out in 5 based on spatial pyramid pond method are as follows:
Step 51: each attribute value sequence AVS is successively averagely divided into 1,2,4,8 ..., 2L-1Block, L indicate pyramidal layer
Number;
Step 52: calculating separately average, sum and the mode of every block number evidence, extract 3 values in from every block number;
Step 53: these values being spliced, each attribute value sequence AVS will generate 3* (2L- 1) a value.
3. the communication data coding method according to claim 2 suitable for deep learning, which is characterized in that the step
After 2 further include: filter out Flow record number less than 2L-1Communication pair.
4. the communication data coding method according to claim 1 suitable for deep learning, which is characterized in that communication pair
When the Flow data of interior indefinite length are encoded, while it is special to retain the spatial character of communication behavior and time between communication node
Property;The field of statistical conditions and distribution situation for corresponding attribute value has carried out further selection, by source port, purpose
Port, byte number, packet digital section are added in field to be encoded;
The field for the trend that attribute value corresponding for fields related and all kinds of to call duration time changes over time is carried out into one
Step processing: will communicate internal Flow and record and be ranked up from small to large by time field, according to the time of two stream in front and back it
Difference calculates time interval, while time interval and Duration field being added in field to be encoded.
5. the communication data coding method according to claim 1 suitable for deep learning, which is characterized in that the step
Original feature vector FV is normalized using Z-score standardized method in 7, calculates function are as follows:
In formula, x is original feature vector initial value, x*For original feature vector standard value;μ is mean value, and σ is standard deviation.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810996375.6A CN109246095B (en) | 2018-08-29 | 2018-08-29 | A kind of communication data coding method suitable for deep learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810996375.6A CN109246095B (en) | 2018-08-29 | 2018-08-29 | A kind of communication data coding method suitable for deep learning |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109246095A true CN109246095A (en) | 2019-01-18 |
CN109246095B CN109246095B (en) | 2019-06-21 |
Family
ID=65068889
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810996375.6A Active CN109246095B (en) | 2018-08-29 | 2018-08-29 | A kind of communication data coding method suitable for deep learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109246095B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110191024A (en) * | 2019-05-31 | 2019-08-30 | 中国联合网络通信集团有限公司 | Network flow monitoring method and device |
Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP1280348A1 (en) * | 2000-04-21 | 2003-01-29 | Sony Corporation | Information processing apparatus and method, program, and recorded medium |
CN1728718A (en) * | 2004-07-27 | 2006-02-01 | 邓里文 | Adaptation method in sue for syncretizing Internet and plesiochronous digital hierarchy |
CN101695035A (en) * | 2009-10-21 | 2010-04-14 | 成都市华为赛门铁克科技有限公司 | Flow rate identification method and device thereof |
US20120099465A1 (en) * | 2010-04-22 | 2012-04-26 | Yuefeng Ji | Method and its devices of network tcp traffic online identification using features in the head of the data flow |
CN102523241A (en) * | 2012-01-09 | 2012-06-27 | 北京邮电大学 | Method and device for classifying network traffic on line based on decision tree high-speed parallel processing |
CN103812700A (en) * | 2014-02-18 | 2014-05-21 | 西南大学 | Message classifying method based on rule information entropy |
CN103970867A (en) * | 2014-05-09 | 2014-08-06 | 浙江科技学院 | Social police-calling service position coding method based on multi-source data matrix |
CN105260736A (en) * | 2015-09-08 | 2016-01-20 | 西安交通大学 | Fast image feature representing method based on normalized nonnegative sparse encoder |
CN106022374A (en) * | 2016-05-18 | 2016-10-12 | 中国农业银行股份有限公司 | Method and device for classifying historical process data |
CN106555788A (en) * | 2016-11-11 | 2017-04-05 | 河北工业大学 | Application of the deep learning based on Fuzzy Processing in hydraulic equipment fault diagnosis |
US20180144041A1 (en) * | 2016-11-21 | 2018-05-24 | International Business Machines Corporation | Transaction discovery in a log sequence |
CN108418792A (en) * | 2018-01-29 | 2018-08-17 | 华北电力大学 | Network escape behavior detection algorithm based on deep-cycle neural network |
-
2018
- 2018-08-29 CN CN201810996375.6A patent/CN109246095B/en active Active
Patent Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP1280348A1 (en) * | 2000-04-21 | 2003-01-29 | Sony Corporation | Information processing apparatus and method, program, and recorded medium |
CN1728718A (en) * | 2004-07-27 | 2006-02-01 | 邓里文 | Adaptation method in sue for syncretizing Internet and plesiochronous digital hierarchy |
CN101695035A (en) * | 2009-10-21 | 2010-04-14 | 成都市华为赛门铁克科技有限公司 | Flow rate identification method and device thereof |
US20120099465A1 (en) * | 2010-04-22 | 2012-04-26 | Yuefeng Ji | Method and its devices of network tcp traffic online identification using features in the head of the data flow |
CN102523241A (en) * | 2012-01-09 | 2012-06-27 | 北京邮电大学 | Method and device for classifying network traffic on line based on decision tree high-speed parallel processing |
CN103812700A (en) * | 2014-02-18 | 2014-05-21 | 西南大学 | Message classifying method based on rule information entropy |
CN103970867A (en) * | 2014-05-09 | 2014-08-06 | 浙江科技学院 | Social police-calling service position coding method based on multi-source data matrix |
CN105260736A (en) * | 2015-09-08 | 2016-01-20 | 西安交通大学 | Fast image feature representing method based on normalized nonnegative sparse encoder |
CN106022374A (en) * | 2016-05-18 | 2016-10-12 | 中国农业银行股份有限公司 | Method and device for classifying historical process data |
CN106555788A (en) * | 2016-11-11 | 2017-04-05 | 河北工业大学 | Application of the deep learning based on Fuzzy Processing in hydraulic equipment fault diagnosis |
US20180144041A1 (en) * | 2016-11-21 | 2018-05-24 | International Business Machines Corporation | Transaction discovery in a log sequence |
CN108418792A (en) * | 2018-01-29 | 2018-08-17 | 华北电力大学 | Network escape behavior detection algorithm based on deep-cycle neural network |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110191024A (en) * | 2019-05-31 | 2019-08-30 | 中国联合网络通信集团有限公司 | Network flow monitoring method and device |
Also Published As
Publication number | Publication date |
---|---|
CN109246095B (en) | 2019-06-21 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN104113789B (en) | On-line video abstraction generation method based on depth learning | |
CN105512289B (en) | Image search method based on deep learning and Hash | |
CN108768986A (en) | A kind of encryption traffic classification method and server, computer readable storage medium | |
CN109218223B (en) | Robust network traffic classification method and system based on active learning | |
CN111144470B (en) | Unknown network flow identification method and system based on deep self-encoder | |
CN110427990A (en) | A kind of art pattern classification method based on convolutional neural networks | |
CN107786388B (en) | Anomaly detection system based on large-scale network flow data | |
CN109726735A (en) | A kind of mobile applications recognition methods based on K-means cluster and random forests algorithm | |
CN108537134A (en) | A kind of video semanteme scene cut and mask method | |
CN110704649B (en) | Method and system for constructing flow image data set | |
CN110059616A (en) | Pedestrian's weight identification model optimization method based on fusion loss function | |
CN103200133A (en) | Flow identification method based on network flow gravitation cluster | |
CN111627080B (en) | Gray level image coloring method based on convolution nerve and condition generation antagonistic network | |
CN107516103A (en) | A kind of image classification method and system | |
CN110263236B (en) | Social network user multi-label classification method based on dynamic multi-view learning model | |
CN110929848A (en) | Training and tracking method based on multi-challenge perception learning model | |
CN110061931B (en) | Industrial control protocol clustering method, device and system and computer storage medium | |
CN109547251B (en) | Service system fault and performance prediction method based on monitoring data | |
CN107154923A (en) | A kind of network inbreak detection method based on the very fast learning machine of multilayer | |
CN110751191A (en) | Image classification method and system | |
WO2023207013A1 (en) | Graph embedding-based relational graph key personnel analysis method and system | |
CN109246095B (en) | A kind of communication data coding method suitable for deep learning | |
CN113627502A (en) | Network encryption traffic classification method | |
US11570069B2 (en) | Network traffic classification method and system based on improved K-means algorithm | |
CN112887291A (en) | I2P traffic identification method and system based on deep learning |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |