CN107633058B - Deep learning-based data dynamic filtering system and method - Google Patents
Deep learning-based data dynamic filtering system and method Download PDFInfo
- Publication number
- CN107633058B CN107633058B CN201710853173.1A CN201710853173A CN107633058B CN 107633058 B CN107633058 B CN 107633058B CN 201710853173 A CN201710853173 A CN 201710853173A CN 107633058 B CN107633058 B CN 107633058B
- Authority
- CN
- China
- Prior art keywords
- data
- module
- deep learning
- complex
- storing
- 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
Images
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a data dynamic filtering system and method based on deep learning, and relates to the field of network data analysis. The system comprises a data source collecting module (10), a deep learning module (20), a deep recognition module (30), a data access filtering module (40) and a data processing and storing module (50); the data source collection module (10), the deep learning module (20) and the deep recognition module (30) are sequentially and circularly interacted; the data access filtering module (40), the data processing and storing module (50) and the depth recognition module (30) are in cyclic interaction for a time. Compared with the traditional data analysis, the invention expands the field of network data analysis, can reverse the basic characteristics of complex data and carries out targeted comprehensive analysis on the related data.
Description
Technical Field
The invention relates to the field of network data analysis, in particular to a data dynamic filtering system and method based on deep learning.
Background
Since AlphaGo defeats the Lishi stone on Weiqi in 2016, artificial intelligence raises a wave of heat tide, intelligent driving is realized, and the hundredth brain follows up, and the core technology of the method lies in deep learning. Compared with the traditional artificial intelligence, the deep learning greatly improves the accuracy and the intelligence in the fields of image processing, natural language processing, audio processing and the like, and the recognition and the feature extraction from mass data through the deep learning can be realized at present when the performance of a computer is continuously improved and a GPU is continuously developed, and the current popular software platform is tenor flow of Google company.
In the field of network data analysis, front-end data is accessed to an OCTEON network processing chip produced by Cavium corporation for processing, the throughput performance of the front-end data is advanced compared with the function of a coprocessor, and the performance is high in the analysis and filtration of mass network data.
Disclosure of Invention
The invention aims to provide a data dynamic filtering system and a data dynamic filtering method based on deep learning, aiming at the field of network data analysis, and aiming at filtering related source data by reversely deducing basic characteristics of complex network data, namely image, audio and signal data.
The technical scheme for realizing the aim of the invention is as follows:
firstly, a data source collection module is started, data in file systems such as linux, HDFS and/or GCS are manually collected, downloaded and/or read, and data identified in a depth identification module are collected; after data are collected, the data are sent to a deep learning module tensorflow for training, and the deep learning module stores a trained calculation graph and a parameter result and sends the calculation graph and the parameter result to a deep recognition module for matching; at the moment, starting a data access filtering module and sending data to a data processing and storing module; the data processing and storing module can perform traditional data analysis and store data; the depth recognition module reads the complex data for recognition at the moment, sends the recognized complex data to the data source collection module for training, extracts the basic features of the recognized complex data and sends the extracted basic features to the data access filtering system for filtering so as to perform special analysis.
Specifically, the method comprises the following steps:
data dynamic filtering system (short system) based on deep learning
The system comprises a data source collecting module, a deep learning module, a deep recognition module, a data access filtering module and a data processing and storing module;
the interaction relationship is as follows:
the data source collection module, the deep learning module and the depth recognition module are sequentially and circularly interacted, the data source collection module collects data and provides the data to the deep learning module for learning, the deep learning module provides a calculation map and parameters for the deep recognition module to recognize, and the depth recognition module collects the recognized complex data to the data source collection module;
the data access filtering module, the data processing and storing module and the depth recognition module are sequentially and circularly interacted, the data access filtering module filters network data to the data processing and storing module according to the basic characteristics, the data processing and storing module 50 provides the complex data to the depth recognition module for recognition, and the depth recognition module 30 provides the basic characteristics of the complex data to the data access filtering module after recognition.
Second, data dynamic filtering method (short method) based on deep learning
The method comprises the following steps:
firstly, a data source collection module is started, data in three file systems of POSIX, HDFS and GCS are collected and downloaded or read manually, and data identified in a depth identification module are collected and used for learning for the deep learning module;
after data are collected, the data are sent to a deep learning module for calculation of a convolutional neural network or a cyclic neural network, a trained calculation graph and a parameter result are stored, and the data are sent to a deep recognition module for matching complex data;
starting a data access filtering module, filtering the accessed network data according to basic characteristics, and sending the filtered and matched network data to a data processing and storing module;
the data processing and storing module can analyze the traditional data and store the complex data to be identified;
the depth recognition module reads the complex data for recognition at the moment: sending the identified complex data to a data source collection module for training, wherein the training is carried out in a circulating way through the first step; and extracting the basic features of the identified complex data and sending the extracted basic features to a data access filtering system for filtering so as to carry out targeted comprehensive analysis, wherein the step III is carried out circularly.
The invention has the following advantages and positive effects:
firstly, a tensierflow framework is used, and source codes of the tensierflow framework are provided by Google company, so that the tensierflow framework is stable, high in updating speed, capable of supporting GPU (graphics processing Unit) calculation and easy to maintain and transplant;
deep learning expands the field of network data analysis and improves the accuracy;
the basic characteristics of the complex data which can be automatically identified are extracted and sent to a data access filtering system for filtering so as to carry out special analysis, thereby enhancing the pertinence and comprehensiveness of data analysis;
an OCTEON network processor produced by Cavium corporation is used for carrying out coprocessor filtering, and the performance is high;
dynamic filtering to respond the change of the condition in real time.
In a word, compared with the traditional data analysis, the method expands the field of network data analysis, can reverse the basic characteristics of complex data, and carries out targeted comprehensive analysis on related data.
Drawings
FIG. 1 is a block diagram showing the construction of the system;
FIG. 2 is a flow chart of the operation of the data source collection module 10;
FIG. 3 is a flowchart of the operation of deep learning module 20;
FIG. 4 is a flowchart of the operation of depth identification module 30;
fig. 5 is a flowchart of the operation of the data access filtering module 40;
fig. 6 is a flowchart of the operation of the data processing and storing module 50.
In the figure:
10-a data source collection module;
20-deep learning module;
30-depth recognition module;
40-data access filtering module;
and 50, a data processing storage module.
English-translation:
1. alphago: weiqi intelligent robot designed by Google corporation;
2. HDFS (Hadoop distributed File System): (ii) a A distributed file system running on general purpose hardware;
3. GCS: distributed file system of Google corporation;
4. tensoflow: the deep learning software platform of Google corporation;
5. cavium: worldwide leading multi-core MIPS and ARM processor providers;
6. OCTEON: network processing chips provided by the company Cavium;
7. POSIX: a portable operating system interface of UNIX;
8. batch: deeply learning batch processing objects, and checking a data set;
9. URL: uniform resource locator, brief representation of resource location and access method on the internet;
10. HOST: host names in the internet;
11. GPU: a display card chip;
12. relu: a neuron activation function algorithm;
13. CUDA: the operation platform provided by NVIDIA great company realizes GPU cooperative processing;
14. HFA: a fuzzy matching coprocessor of the OCTEON chip;
15. protobuf: a data exchange format defined by Google corporation.
Detailed Description
The following detailed description is made with reference to the accompanying drawings and examples.
A, system
1. General of
As shown in fig. 1, the system includes a data source collection module 10, a deep learning module 20, a deep recognition module 30, a data access filtering module 40 and a data processing and storing module 50;
the interaction relationship is as follows:
the data source collection module 10, the deep learning module 20 and the depth recognition module 30 are sequentially and circularly interacted, the data source collection module 10 collects data and provides the data to the deep learning module 20 for learning, the deep learning module 20 provides a calculation map and parameters for the depth recognition module 30 to recognize, and the depth recognition module 30 provides recognized complex data to the data source collection module 10 for collection;
the data access filtering module 40, the data processing storage module 50 and the depth recognition module 30 are sequentially and circularly interacted, the data access filtering module 40 filters network data to the data processing storage module 50 according to the basic characteristics, the data processing storage module 50 provides the complex data to the depth recognition module 30 for recognition, and the depth recognition module 30 provides the basic characteristics of the complex data to the data access filtering module 40 after recognition.
The working mechanism is as follows:
the data source collection module 10 interacts with the deep learning module 20, transmits a data set to the deep learning module 20, interacts with the deep learning module 20 and transmits a calculation map and parameters to the deep recognition module 30;
the depth recognition module 30 interacts with the data source collection module 10, and transmits the recognized complex data to the data source collection module 10 to form a part of a data set;
the data access filtering module 40 interacts with the data processing and storing module 50, transmits the network data to the data processing and storing module 50 for traditional analysis and storage, and the data processing and storing module 50 interacts with the depth recognition module 30 and transmits the complex data to the depth recognition module 30 for recognition;
the depth recognition module 30 and the data access filtering module 40 transmit the basic features of the complex data to the data access filtering module 40 for dynamic filtering.
2. Functional module
1) Data source collection module 10
Manually collecting, downloading, network acquiring and/or reading data in three file systems of POSIX, HDFS and/or GCS, and collecting complex data identified in the depth identification module 30;
specifically, as shown in fig. 2, the workflow of the data source collection module 10 is as follows:
a. start-200
Starting a data source collection module, wherein the data source collection module needs to have a basic data preprocessing function, perform cutting, scaling and rotating functions on an image and a spectrogram, also needs to have a data set classification function, and processes noise data;
b. downloading classic dataset using network acquisition data-201
The network acquisition of data and the downloading of classical data sets are important methods for acquiring data sources;
c. manual collection of a particular data set-202
For special application scenes, a manual method is needed for acquiring a data source;
d. collecting complex data from depth recognition module-203
The complex data sent by the depth recognition module is generally a high-quality data source with characteristics and should be collected;
e. reading data-204 in POSIX, HDFS and/or GCS file systems
A part of data is stored in a POSIX, HDFS and/or GCS file system, and particularly, large data is often stored in the HDFS and GCS file systems and needs to be collected;
f. preprocessing and marshalling into three types of data sets-205 for training, validation and testing
The data sets need to be sorted into three categories, training, validation and testing, for use by deep learning module 20.
2) Deep learning module 20
Adopting a tensoflow frame, activating through a forward propagation algorithm, optimizing through a backward propagation algorithm, carrying out deep feature extraction on different complex data by adopting a convolutional neural network or a cyclic neural network algorithm, and finally training a calculation graph and parameters with high accuracy;
specifically, as shown in fig. 3, the work flow of the deep learning module 20 is as follows:
A. start-300
Initializing the tenserflow, and loading the model under the condition of transfer learning;
B. define the Forward propagation Algorithm-301
A forward propagation algorithm, namely activating input data in deep learning, generally adopting a Relu algorithm, carrying out deep feature extraction on complex data by adopting a convolutional neural network or a cyclic neural network algorithm, and adding a hidden layer, a convolutional layer and a pooling layer for calculation;
C. defining a back propagation algorithm-302
Optimizing the deep learning model by using a back propagation algorithm, converging the model by calculating a loss function, namely cross entropy or mean square error, and selecting an optimization function according to the tuning condition of parameters;
D. defining multithreading, queues, and GPU device-303
The training speed is improved by using multiple threads, a queue and GPU equipment, namely the multiple threads adopt coordinator and start _ queue _ runners functions, the queue adopts string _ input _ producer functions, and the GPU installs CUDA for use;
E. begin training and verify-304
Training by using data provided by the data source collection module 10, optimizing from an input layer to an output layer, namely learning rate, random sampling and regularization, verifying generated batch of partial data, and obtaining an optimized result through gradient reduction;
F. save training results-305
After the training result is stored, the training result is sent to the depth recognition module 30 to recognize the complex data.
3) Depth recognition module 30
The computation graph and the parameters provided by the deep learning module 20 are adopted to preprocess and identify the complex data, namely image, audio, video and signal data, provide data for the data source collection module 10 and provide the basic characteristics of the complex data for the data access filtering module 40;
specifically, as shown in fig. 4, the work flow of the depth recognition module 30 is as follows:
i, Start-400
Initializing a module, wherein different identification requirements correspond to different processes, and different calculation graphs and parameters need to be provided;
II, load and initialize model-401
Loading a calculation graph and parameters, reading data in a protobuf format, wherein the size of the data on an android system is generally not more than 64M, the size of the data on a hard disk is generally not more than 512M, and establishing a session for the calculation number and the parameters;
III, reading complex data and preprocessing-402
Converting audio and signals into corresponding spectrogram streams, and then preprocessing the images, wherein the preprocessing comprises size, turnover, brightness, contrast, hue, saturation and marking to form a tensor;
IV, running complex data on the model-403
Running complex data on the model to form an output, wherein the output is also a tensor, and in the classification problem, the tensor contains the probability of each category;
v, result-404 of TOP5
For the classification problem, giving the category name of the first five names with the maximum probability, and presenting the category name for analysis and recording;
VI, processing the identified complex data and basic characteristics-405
The complex data is provided to the data source collection module 10, and the basic features of the complex data are provided to the data access filtering module 40.
4) Data access filtering module 40
Accessing network data, dynamically filtering according to basic characteristics of complex data, and filtering related data;
specifically, as shown in fig. 5, the data access filtering module 40 works as follows:
i, start-500
The module begins initialization. Initializing HFA hardware of Cavium company, allocating corresponding memory and loading basic characteristics, and creating a TCP or UDP data stream for each data packet;
ii, extracting basic characteristics from the network data packet-501
Extracting basic characteristics, namely ip, quintuple, URL or HOST of data and special fields from a network data packet;
iii, obtaining basic characteristics from the depth recognition module and matching with the data packet 502
The depth recognition module 30 will send the basic features of the complex data to the module, and at this time, the HFA coprocessor performs quintuple matching or fuzzy matching of character strings;
iv, filter by stream and send data backward-503
And the data packets matching the basic characteristics are filtered according to the associated stream data and sent to the data processing and storing module 50.
5) Data processing memory module 50
Conventional data analysis stores complex data for processing by depth module 30.
Specifically, as shown in fig. 6, the work flow of the data processing and storing module 50 is as follows:
alpha, Start-600
Initializing a module, starting the traditional deep message detection, and starting a database;
beta, processing network data-601
Associating the basic characteristics of the network data packet with the complex data, namely the ip, quintuple, URL or HOST of the data, and corresponding the special field with the complex data;
gamma, storing relevant basic features and complex data information-602
Relevant basic features and complex data information are stored, at present, a popular HDFS file system is adopted for storing big data, and files are read by the depth recognition module 30 for recognition.
3. Working mechanism of dynamic filtering system
The data source collection module is connected with the deep learning module and transmits a data set to the deep learning module, and the deep learning module is connected with the deep recognition module and transmits the calculation map and the parameters to the deep recognition module;
the depth recognition module and the data source collection module are used for transmitting the recognized complex data to the data source collection module to form a part of a data set;
the data access filtering module is connected with the data processing and storing module, the network data is transmitted to the data processing and storing module for traditional analysis and storage, and the data processing and storing module is connected with the depth recognition module and transmits the complex data to the depth recognition module for recognition;
the depth recognition module and the data access filtering module transmit the basic characteristics of the complex data to the data access filtering module for dynamic filtering.
For example, if one picture is found in network data, the picture is found to be three through the depth recognition module, and the account corresponding to the picture is of a user with a "hacker", then the network data with the account being the "hacker" is filtered out for analysis in the data access filtering module, and the results of searching the target through the picture and filtering and analyzing the network data related to the target are achieved.
Claims (2)
1. A data dynamic filtering system based on deep learning is characterized in that:
the system comprises a data source collection module, a deep learning module, a deep recognition module, a data access filtering module and a data processing and storing module;
the interaction relationship is as follows:
the data source collection module, the deep learning module and the depth recognition module are sequentially and circularly interacted, the data source collection module collects data and provides the data to the deep learning module for learning, the deep learning module provides a calculation map and parameters for the deep recognition module to recognize, and the depth recognition module collects the recognized complex data to the data source collection module; complex data is a data source with basic features; the basic features include ip, quintuple, URL, or HOST of the data;
the data access filtering module, the data processing and storing module and the depth recognition module are sequentially and circularly interacted, the data access filtering module filters network data to the data processing and storing module according to the basic characteristics, the data processing and storing module provides the complex data to the depth recognition module for recognition, and the depth recognition module provides the basic characteristics of the complex data to the data access filtering module after recognition;
the work flow of the data source collection module is as follows:
a. starting;
starting a data source collection module, wherein the data source collection module needs to have a basic data preprocessing function, perform cutting, scaling and rotating functions on an image and a spectrogram, also needs to have a data set classification function, and processes noise data;
b. using a network to acquire data or download a classic data set;
c. manually collecting a special data set;
for a specific application scene, a manual method is needed to collect a data source;
d. receiving complex data sent by a depth recognition module;
the complex data sent by the depth recognition module is a data source with basic characteristics and is required to be collected;
e. reading data in a POSIX, HDFS and/or GCS file system;
a part of data is stored in POSIX, HDFS and/or GCS file systems, wherein the HDFS and GCS file systems store big data and need to collect the data in the file systems;
f. preprocessing and arranging into three types of data sets of training, verifying and testing;
the data set needs to be arranged into three types of training, verifying and testing so as to be conveniently used by a deep learning module;
the work flow of the deep learning module is as follows:
A. starting;
initializing the tensoflow, and loading a deep learning model under the condition of transfer learning;
B. defining a forward propagation algorithm;
a forward propagation algorithm, namely activating input data in deep learning, adopting a Relu algorithm for activation, adopting a convolutional neural network or a cyclic neural network algorithm for deep feature extraction of complex data, and adding a hidden layer, a convolutional layer and a pooling layer for calculation;
C. defining a back propagation algorithm;
optimizing the deep learning model by adopting a back propagation algorithm, converging the model by calculating a cross entropy loss function or a mean square error loss function, and selecting the optimized loss function according to the optimization condition of the parameters;
D. defining multithreading, queues and GPU equipment;
the model training speed is improved by using multiple threads, a queue and GPU equipment, wherein the multiple threads adopt coordinator and start _ queue _ runners functions, the queue adopts string _ input _ producer functions, and the GPU installs a CUDA for use;
E. starting training and verifying;
training by using data provided by a data source collection module, optimizing from an input layer to an output layer, verifying generated batch of partial data, and obtaining an optimized result through gradient reduction;
F. storing the training result;
after the training result is stored, the training result is sent to a depth recognition module to recognize the complex data;
the working flow of the depth identification module is as follows:
i, starting;
initializing a module, wherein different identification requirements correspond to different processes, and different calculation graphs and parameters are provided;
II, loading and initializing the model;
loading a calculation graph and parameters, reading data in a protobuf format, wherein the size of the data on an android system is not more than 64M, the size of the data on a hard disk is not more than 512M, and establishing a session for the calculation graph and the parameters;
III, reading and preprocessing the complex data;
converting audio and signals into corresponding spectrogram streams, and then preprocessing an image, wherein preprocessed image information comprises size, brightness, contrast, hue and saturation to form a tensor;
IV, running complex data on the deep learning model;
running complex data on a deep learning model to form an output, wherein the output is also a tensor, and in the classification problem, the output tensor comprises the probability of each class;
v, the result of TOP5 is given;
for the classification problem, giving the category name of the first five names with the maximum probability, and presenting the category name for analysis and recording;
VI, processing the identified complex data and basic characteristics;
providing complex data for a data source collection module and providing basic characteristics of the complex data for a data access filtering module;
the work flow of the data access filtering module is as follows:
i, start;
the module starts initialization; initializing HFA hardware of Cavium company, distributing corresponding memory and loading basic characteristics, and creating a TCP or UDP data stream for each data packet;
ii, extracting basic features from the network data packet;
extracting basic features from the network data packet;
iii, acquiring basic features from the depth recognition module and matching the basic features with the data packet;
the depth recognition module sends the basic features of the complex data to the data access filtering module, and at the moment, the HFA coprocessor carries out quintuple matching or fuzzy matching of character strings;
iv, filtering according to the flow and sending data backwards;
the data packets matched with the basic characteristics are filtered according to the associated stream data and are sent to a data processing and storing module;
the work flow of the data processing and storing module is as follows:
α, start;
initializing a module, detecting and starting a deep message, and starting a database;
beta, processing the network data;
associating the basic characteristics of the network data packet with the complex data, specifically, corresponding ip, quintuple, URL or HOST of the data with the complex data;
gamma, storing relevant basic features and complex data information;
and storing the related basic features and the complex data information, storing by using an HDFS file system, and reading the file by using a depth identification module for identification.
2. The dynamic data filtering method based on the system of claim 1, characterized by comprising the following steps:
firstly, a data source collection module is started, data in three file systems of POSIX, HDFS and GCS are collected and downloaded or read manually, and data identified in a depth identification module are collected and used for learning for the deep learning module;
after data are collected, the data are sent to a deep learning module for calculation of a convolutional neural network or a cyclic neural network, a trained calculation graph and a parameter result are stored, and the data are sent to a deep recognition module for matching complex data;
starting a data access filtering module, filtering the accessed network data according to basic characteristics, and sending the filtered and matched network data to a data processing and storing module;
fourthly, the data processing and storing module analyzes the data and stores the complex data to be identified;
the depth recognition module reads the complex data for recognition: sending the identified complex data to a data source collection module for training, wherein the training is carried out in a circulating way through the first step; and extracting the basic features of the identified complex data and sending the extracted basic features to a data access filtering system for filtering so as to carry out targeted comprehensive analysis, wherein the step III is carried out circularly.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710853173.1A CN107633058B (en) | 2017-09-20 | 2017-09-20 | Deep learning-based data dynamic filtering system and method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710853173.1A CN107633058B (en) | 2017-09-20 | 2017-09-20 | Deep learning-based data dynamic filtering system and method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107633058A CN107633058A (en) | 2018-01-26 |
| CN107633058B true CN107633058B (en) | 2021-06-01 |
Family
ID=61103400
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710853173.1A Active CN107633058B (en) | 2017-09-20 | 2017-09-20 | Deep learning-based data dynamic filtering system and method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107633058B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108647206B (en) * | 2018-05-04 | 2021-11-12 | 重庆邮电大学 | Chinese junk mail identification method based on chaos particle swarm optimization CNN network |
| CN109635180A (en) * | 2018-12-13 | 2019-04-16 | 武汉虹旭信息技术有限责任公司 | Keyword classification processing system and its method based on internet mass information |
| CN109448707A (en) * | 2018-12-18 | 2019-03-08 | 北京嘉楠捷思信息技术有限公司 | Voice recognition method and device, equipment and medium |
| CN112862109B (en) * | 2021-02-09 | 2024-05-24 | 上海商汤智能科技有限公司 | Deep learning model execution method and device, electronic equipment and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102546624A (en) * | 2011-12-26 | 2012-07-04 | 西北工业大学 | Method and system for detecting and defending multichannel network intrusion |
| CN102821002A (en) * | 2011-06-09 | 2012-12-12 | ***通信集团河南有限公司信阳分公司 | Method and system for network flow anomaly detection |
| CN103179105A (en) * | 2012-10-25 | 2013-06-26 | 四川省电力公司信息通信公司 | Intelligent Trojan horse detecting device based on behavior features in network flows and method thereof |
| CN106096060A (en) * | 2016-07-06 | 2016-11-09 | 吴本刚 | Ocean network security risk system of defense |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10360901B2 (en) * | 2013-12-06 | 2019-07-23 | Nuance Communications, Inc. | Learning front-end speech recognition parameters within neural network training |
-
2017
- 2017-09-20 CN CN201710853173.1A patent/CN107633058B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102821002A (en) * | 2011-06-09 | 2012-12-12 | ***通信集团河南有限公司信阳分公司 | Method and system for network flow anomaly detection |
| CN102546624A (en) * | 2011-12-26 | 2012-07-04 | 西北工业大学 | Method and system for detecting and defending multichannel network intrusion |
| CN103179105A (en) * | 2012-10-25 | 2013-06-26 | 四川省电力公司信息通信公司 | Intelligent Trojan horse detecting device based on behavior features in network flows and method thereof |
| CN106096060A (en) * | 2016-07-06 | 2016-11-09 | 吴本刚 | Ocean network security risk system of defense |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107633058A (en) | 2018-01-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107633058B (en) | Deep learning-based data dynamic filtering system and method | |
| US10140553B1 (en) | Machine learning artificial intelligence system for identifying vehicles | |
| US20240037141A1 (en) | Content tagging | |
| CN109948428B (en) | GPU cluster deep learning edge computing system facing to sensing information processing | |
| CN110458107B (en) | Method and device for image recognition | |
| CN113365147B (en) | Video editing method, device, equipment and storage medium based on music card point | |
| CN111090763A (en) | Automatic picture labeling method and device | |
| US20210089825A1 (en) | Systems and methods for cleaning data | |
| CN111783712A (en) | Video processing method, device, equipment and medium | |
| CN111882059A (en) | Data processing method and device, electronic equipment and storage medium | |
| CN111177328B (en) | Question-answer matching system and method, question-answer processing device and medium | |
| CN110149529A (en) | Processing method, server and the storage medium of media information | |
| Zhang et al. | A mobile application for cat detection and breed recognition based on deep learning | |
| CN117033657A (en) | Information retrieval method and device | |
| CN113724132B (en) | Image style migration processing method and device, electronic equipment and storage medium | |
| CN113869398B (en) | Unbalanced text classification method, device, equipment and storage medium | |
| US20230161568A1 (en) | Ai differentiation based hw-optimized intelligent software development tools for developing intelligent devices | |
| CN116363457B (en) | Task processing, image classification and data processing method of task processing model | |
| CN112200862A (en) | Training method of target detection model, target detection method and device | |
| CN115035463B (en) | Behavior recognition method, behavior recognition device, behavior recognition equipment and storage medium | |
| CN116311492A (en) | Gesture recognition method and system based on depth camera and contour extraction | |
| CN111950591B (en) | Model training method, interaction relation recognition device and electronic equipment | |
| CN115587297A (en) | Method, apparatus, device and medium for constructing image recognition model and image recognition | |
| CN114492366A (en) | Binary file classification method, computing device and storage medium | |
| CN111461253A (en) | Automatic feature extraction system and method |
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 |