CN112801185A - Network security situation understanding and evaluating method based on improved neural network - Google Patents
Network security situation understanding and evaluating method based on improved neural network Download PDFInfo
- Publication number
- CN112801185A CN112801185A CN202110117235.9A CN202110117235A CN112801185A CN 112801185 A CN112801185 A CN 112801185A CN 202110117235 A CN202110117235 A CN 202110117235A CN 112801185 A CN112801185 A CN 112801185A
- Authority
- CN
- China
- Prior art keywords
- data
- neural network
- layer
- network
- improved
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/10—Text processing
- G06F40/12—Use of codes for handling textual entities
- G06F40/126—Character encoding
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/10—Text processing
- G06F40/12—Use of codes for handling textual entities
- G06F40/151—Transformation
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
-
- 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
- G06N3/082—Learning methods modifying the architecture, e.g. adding, deleting or silencing nodes or connections
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Biomedical Technology (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Error Detection And Correction (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a network security situation understanding and evaluating method based on an improved neural network, which improves and optimizes a model framework by adopting a composite optimal unit and using Relu6 as an activation function of a convolutional layer; meanwhile, the distribution characteristics of the checked test sample data are fully considered, network training is carried out after the balance of the test data set is adjusted, and the accuracy rate of situation understanding and evaluation is improved; according to the characteristic that the network data information has continuity in time, the current network information is analyzed, and meanwhile, the time series data is effectively subjected to data fusion by combining the network data before the attack; the operation time of the traditional situation awareness research method is short when the attribute features are more and the sample size is larger.
Description
Technical Field
The invention relates to a network security situation understanding and evaluating method, in particular to a network security situation understanding and evaluating method based on an improved neural network, and belongs to the technical field of network security.
Background
With the rapid development of 5G technologies, blockchains and artificial intelligence technologies, network attack events are increasing, and with the great increase of the number of network terminal devices, network data becomes more huge and complex, and network security situation awareness becomes more important.
Most of the traditional network security situation understanding and evaluation methods are methods based on mathematical models, rule-based reasoning, probability statistics and the like, and although the methods have respective advantages, the methods also have certain defects, for example, the methods based on the mathematical models lack a unified standard in weight establishment, the methods based on the rule-based reasoning have higher computational complexity, the methods based on the probability statistics need a large amount of storage space, and gradient explosion is easy to generate.
With the huge and complicated network data and the rapid increase of the data volume of the network information, the traditional machine learning algorithm has less obvious advantages when processing multi-attribute data and large sample data, has slightly poor deep learning performance in the aspects of feature selection, classification processing, running time and the like, does not need to manually select attribute features for deep learning, a learner only needs to load the data information into a neural network, the network can perform autonomous learning, and the deep learning can show better effect with the increase of the data volume and the increase of the attribute features.
Disclosure of Invention
The invention aims to provide a network security situation understanding and evaluating method based on an improved neural network.
In order to solve the technical problems, the invention adopts the technical scheme that: a network security situation understanding and evaluating method based on an improved neural network comprises the following steps:
step 1: data preprocessing: the method comprises the following steps:
step 1-1: data type conversion: converting more than 1 symbolic data into digital identification data;
step 1-2: one-hot encoding: carrying out one-hot coding on each digital identification type data;
step 1-3: data dimension expansion: expanding the characteristic dimension of the one-hot coding to a minimum square number larger than the characteristic dimension of the one-hot coding, wherein the expanded dimension is filled with preset numbers; converting the one-hot codes after data dimension expansion into a characteristic diagram represented by a square matrix;
step 2: building an improved convolutional neural network: the convolutional neural network comprises an improved convolutional neural network composite optimal structure unit above level 1, a Droupout layer, a Flatten layer, a full connection layer, a BatchNomalization layer, a full connection layer and a Softmax layer; the composite optimal structural unit includes first to third convolution layers and a pooling layer; the convolution kernel size of the first convolution layer is 1xM, the convolution kernel size of the second convolution layer is Nx1, and the convolution kernel size of the third convolution layer is 1x 1; the number of channels of the first to third convolution layers is the same; each convolutional layer comprises an activation function;
and step 3: training an improved convolutional neural network: inputting training set data, and training parameters of the improved convolutional neural network;
and 4, step 4: understanding and evaluating network security posture: inputting prediction set data into the improved convolutional neural network trained in the step 1, and understanding and evaluating the network security situation.
Further, the maximum pooling is adopted for the composite optimal unit in the step 3, and 2x2 pooling and 1x1 pooling are used in a crossed manner.
Further, the activation function of each convolutional layer is Relu 6:
f(x)＝min(6,max(0,x)) (1)
wherein x is the output value of the corresponding convolution layer; f (x) is the output value of the activation function.
Further, before data preprocessing, test set sample equalization processing is carried out;
still further, the training set data includes sample data in Normal, Dos, Probing, R2L, and U2R states; and the test set sample equalization processing is to perform undersampling processing on sample data in a Dos state and perform oversampling processing on the sample data in the states of Probing, R2L and U2R.
Furthermore, the number N of channels of each convolution layer in the improved neural network is more than or equal to 3; the input sample data corresponding to each channel in each convolution layer come from different time states of the network, and the time sequence relation of the input sample data is adjusted through a time factor; the step 3 comprises a time factor training step.
The technical effect obtained by adopting the technical scheme is as follows:
1. by improving and optimizing the model frame, the network parameters are reduced, and the situation understanding and evaluation accuracy is improved;
2. the method avoids the short running time of the traditional situation awareness research method when the attribute characteristics are more and the sample size is larger;
3. according to the characteristic that the network data information has continuity in time, the current network information is analyzed, and meanwhile, the time series data is effectively subjected to data fusion by combining the network data before the attack;
4. the invention fully considers the data characteristics, carries out network training after adjusting the test data set, and further improves the accuracy of situation understanding and evaluation.
Drawings
FIG. 1 is a schematic diagram of a standard multiple input channel convolutional layer;
FIG. 2 is a classical schematic of a convolution decomposition technique;
FIG. 3 is a schematic diagram of depth direction convolution;
FIG. 4 is a schematic diagram of a point convolution;
FIG. 5 is a diagram of a composite optimal structural element of the present invention;
fig. 6 is a flow chart of an embodiment of the present invention.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings and specific embodiments.
A network security situation understanding and evaluating method based on an improved neural network comprises the following steps:
step 1: data preprocessing: the method comprises the following steps:
step 1-1: data type conversion: converting more than 1 symbolic data into digital identification data;
step 1-2: one-hot encoding: carrying out one-hot coding on each digital identification type data;
step 1-3: data dimension expansion: expanding the characteristic dimension of the one-hot coding to a minimum square number larger than the characteristic dimension of the one-hot coding, wherein the expanded dimension is filled with preset numbers; converting the one-hot codes after data dimension expansion into a characteristic diagram represented by a square matrix;
step 2: building an improved convolutional neural network: the convolutional neural network comprises an improved convolutional neural network composite optimal structure unit above level 1, a Droupout layer, a Flatten layer, a full connection layer, a BatchNomalization layer, a full connection layer and a Softmax layer; the composite optimal structural unit includes first to third convolution layers and a pooling layer; the convolution kernel size of the first convolution layer is 1xM, the convolution kernel size of the second convolution layer is Nx1, and the convolution kernel size of the third convolution layer is 1x 1; the number of channels of the first to third convolution layers is the same; each convolutional layer comprises an activation function;
and step 3: training an improved convolutional neural network: inputting training set data, and training parameters of the improved convolutional neural network;
and 4, step 4: understanding and evaluating network security posture: inputting prediction set data into the improved convolutional neural network trained in the step 1, and understanding and evaluating the network security situation.
The maximum pooling is adopted by the composite optimal unit in the step 3, and 2x2 pooling and 1x1 pooling are used alternately.
The activation function of each convolutional layer is Relu 6:
f(x)＝min(6,max(0,x)) (1)
wherein x is the output value of the corresponding convolution layer; f (x) is the output value of the activation function.
The training set data comprises sample data in Normal, Dos, Probing, R2L and U2R states; before data preprocessing, the sample data in the Dos state is subjected to undersampling processing, and the sample data in the Probing, R2L and U2R states is subjected to oversampling processing.
The number N of channels of each convolution layer in the improved neural network is more than or equal to 3; the input sample data corresponding to each channel in each convolution layer come from different time states of the network, and the time sequence relation of the input sample data is adjusted through a time factor; the step 3 comprises a time factor training step.
In this embodiment, a KDDCup-99 data set is taken as an example for technical solution description. In the KDDCup-99 dataset, 3 Protocol types (Protocol _ type), 70 Service types (Service), 11 network connection states (Flag) and 5 major labels (Label) in the original dataset need to be converted from symbolic data to numeric identifiers. The data feature type output by the convolutional neural network is a matrix, for original 41-dimensional data, the size of a data feature graph after the data feature graph is converted into the matrix is small, the feature extraction of the deep convolutional neural network is not facilitated, more original information is lost during pooling, and the training of a final model is greatly influenced. To ensure that the information of the data set is fully utilized and to adapt to the deep convolutional network, one-hot (OH) encoding is used herein, and K is used as a register to encode K different states.
After the original data set is subjected to one-hot encoding, the characteristic dimension of the data attribute is changed from 41 to 128 dimensions, the label dimension is changed from 1 to 5 dimensions, and for a convolutional neural network, in order to operate a convolutional kernel conveniently, it is generally desirable that the characteristic matrix of the input network is a square matrix as much as possible. For this purpose, dimension expansion is performed on the attribute features subjected to the one-hot coding, and the specific implementation method is to supplement 16-dimensional '0' after 128-dimensional to widen the attribute features to 144-dimensional, and the attribute features are converted into a feature map represented by a 12x12 square matrix when the attribute features are input into a convolutional neural network.
As shown in fig. 1, for an input size of H1xW1xC1, a convolution kernel size of MxN, and an output feature size of H2xW2xC2, the parameters of the standard convolutional layer are:
P＝MxNxC1xC2 (1)
convolution decomposition is to combine convolution steps in different directions and then perform convolution, and is essentially to split convolution kernels, for example, two convolution 3 × 3 can replace one convolution layer of 5 × 5, and fig. 2 is a classic schematic diagram of the convolution decomposition technology. The embodiment decomposes in the X direction and the Y direction, so that the network has lower calculated amount and fewer parameters, and simultaneously, the depth of the network is deepened, and the expression capability of a nonlinear expansion model of the network is improved.
For a network with an input size of H1xW1xC1, a convolution kernel size of MxN, and an output characteristic diagram size of H2xW2xC2, after the convolution kernel is decomposed into two convolution kernels, i.e., 1xM and Nx1, in the X and Y directions, the parameters are respectively:
PM＝1xMxC1xC2 (2)
PN＝Nx1xC1xC2 (3)
the total parameter quantities are:
PS1＝(M+N)xC1xC2 (4)
the standard convolutional neural network and the decomposed convolutional neural network parameter are compared as follows:
depth Separable Convolution (Depthwise Separable Convolution) is a Convolution that operates in two steps, first in the depth direction, i.e., Depthwise Convolution, as shown in FIG. 3, and second in point Convolution, i.e., Pointwise Convolution, as shown in FIG. 4. The convolution channel correlation and the space channel correlation are separated, and the depth separable convolution not only reduces the calculation complexity and the model parameter quantity, but also does not cause great influence on the model precision.
The point Convolution Pointwise contribution is used for establishing data connection between different channels at the same position, and the output channel and the characteristic diagram of the Convolution kernel in the step are consistent with the final output channel. For a network with an input size of H1xW1xC1, a convolution kernel size of MxN, and an output signature size of H2xW2xC2, the parameters of this step are:
PP＝C1x1x1xC2 (6)
combining the feature maps obtained in the two steps with a downsampling technology to establish a composite optimal structure unit, as shown in fig. 5, the composite optimal structure unit comprises 4 layers of mechanisms, wherein the first three layers are convolution networks, the fourth layer is a pooling layer, the size of a first layer of convolution kernel is 1xM, and the number of channels is adjusted according to the network effect; the size of the convolution kernel of the second layer is Nx1, and the number of channels needs to be kept consistent with that of the first layer; the size of the convolution kernel of the third layer is 1x1, and the number of channels is consistent with the number of output channels; the fourth layer is a pooling layer, maximum pooling is adopted in the actual training, and 2x2 pooling and 1x1 pooling are used alternately, so that the network parameters are reduced to the maximum extent on the premise of fully extracting network characteristics, and the training time and memory consumption of the model are reduced.
According to data characteristics, the most suitable network depth is selected, the network of the embodiment is 14 layers, wherein 1-8 layers are two optimal structural units, corresponding pooling layer parameters are 2x2 and 1x1 respectively, and then a Droupout layer, a Flatten layer, a full connection layer, a BatchNomalization layer, a full connection layer and a Softmax layer are provided.
In the embodiment, Relu6 is used as an activation function, and the Relu6 function limits the maximum output value of Relu, namely when X is larger than or equal to 6, the function gradient is also 0. Experiments prove that Relu6 enables the network of the invention to present the best learning effect.
The network security situation perception in common data sets mainly has 5 states, namely Normal, Dos, Probing, R2L and U2R, wherein the number of Dos attacks is far higher than the two attack numbers of Probing and R2L, and is more than thousand times of U2R, in a convolutional neural network, the more input samples are, the better the model can show, but the appearance of unbalanced data makes the model not reach an ideal learning result.
In the embodiment, Dos data is subjected to undersampling processing, namely, part of Dos data is lost, and Probing, R2L and U2R data are subjected to oversampling processing, namely, the proportion of the Probing, R2L, U2R and other data is increased.
The traditional network security situation understanding and evaluating only analyzes the current network state, and does not fully consider the network state before the network attack occurs.
For a convolutional neural network, setting different Timer values means that the actual meaning of convolutional input is different in dimensionality, usually, the convolutional neural network input information can be one-dimensional (such as natural language processing), two-dimensional (gray level graph) and three-dimensional (RGB color graph), the time sequence provided by the text enables the characteristic graph input by the convolutional neural network to be N (N is more than or equal to 3) channels, for each channel convolved by the N channels, separate two-dimensional convolution can be performed on each channel, then convolution outputs at corresponding positions of each channel are summed, and the obtained result is the output of the layer of convolution.
In the process of increasing the input channel, the accuracy of network situation understanding and evaluation is increased and then reduced, and the main reason is that in the first stage, along with the increase of the Timer, the characteristic diagram of the input network contains more network data, and the neural network can learn more network information, so that the model shows better and better evaluation results. When the Timer reaches a certain value, the accuracy of the model reaches the highest, and the characteristic diagram of the input convolutional neural network is 12x12 xN. In the second stage, as the Timer increases, the network contains too much network information, which results in an overfitting phenomenon in the model training stage. Therefore, the network is trained to reach the optimal Timer value.
Claims (6)
1. A network security situation understanding and evaluating method based on an improved neural network is characterized by comprising the following steps: the method comprises the following steps:
step 1: data preprocessing: the method comprises the following steps:
step 1-1: data type conversion: converting more than 1 symbolic data into digital identification data;
step 1-2: one-hot encoding: carrying out one-hot coding on each digital identification type data;
step 1-3: data dimension expansion: expanding the characteristic dimension of the one-hot coding to a minimum square number larger than the characteristic dimension of the one-hot coding, wherein the expanded dimension is filled with preset numbers; converting the one-hot codes after data dimension expansion into a characteristic diagram represented by a square matrix;
step 2: building an improved convolutional neural network: the convolutional neural network comprises an improved convolutional neural network composite optimal structure unit above level 1, a Droupout layer, a Flatten layer, a full connection layer, a BatchNomalization layer, a full connection layer and a Softmax layer; the composite optimal structural unit includes first to third convolution layers and a pooling layer; the convolution kernel size of the first convolution layer is 1xM, the convolution kernel size of the second convolution layer is Nx1, and the convolution kernel size of the third convolution layer is 1x 1; the number of channels of the first to third convolution layers is the same; each convolutional layer comprises an activation function;
and step 3: training an improved convolutional neural network: inputting training set data, and training parameters of the improved convolutional neural network;
and 4, step 4: understanding and evaluating network security posture: inputting prediction set data into the improved convolutional neural network trained in the step 1, and understanding and evaluating the network security situation.
2. The improved neural network based network security situation understanding and evaluating method of claim 1, wherein: the maximum pooling is adopted by the composite optimal unit in the step 3, and 2x2 pooling and 1x1 pooling are used alternately.
3. The improved neural network based network security situation understanding and evaluating method of claim 2, wherein: the activation function of each convolutional layer is Relu 6:
f(x)＝min(6,max(0,x)) (1)
wherein x is the output value of the corresponding convolution layer; f (x) is the output value of the activation function.
4. The improved neural network based network security situation understanding and evaluating method of claim 3, wherein: before data preprocessing, test set sample equalization is performed.
5. The improved neural network-based network security situation understanding and evaluating method of claim 4, wherein: the training set data comprises sample data in Normal, Dos, Probing, R2L and U2R states; and the test set sample equalization processing is to perform undersampling processing on sample data in a Dos state and perform oversampling processing on the sample data in the states of Probing, R2L and U2R.
6. The improved neural network-based network security situation understanding and evaluating method according to any one of claims 1-5, wherein the number N of channels of each convolutional layer in the improved neural network is greater than or equal to 3; the input sample data corresponding to each channel in each convolution layer come from different time states of the network, and the time sequence relation of the input sample data is adjusted through a time factor; the step 3 comprises a time factor training step.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110117235.9A CN112801185B (en) | 2021-01-28 | 2021-01-28 | Network security situation understanding and evaluating method based on improved neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110117235.9A CN112801185B (en) | 2021-01-28 | 2021-01-28 | Network security situation understanding and evaluating method based on improved neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112801185A true CN112801185A (en) | 2021-05-14 |
| CN112801185B CN112801185B (en) | 2023-02-24 |
Family
ID=75812452
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110117235.9A Active CN112801185B (en) | 2021-01-28 | 2021-01-28 | Network security situation understanding and evaluating method based on improved neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112801185B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114970694A (en) * | 2022-05-12 | 2022-08-30 | 河北师范大学 | Network security situation assessment method and model training method thereof |
| CN115834174A (en) * | 2022-11-15 | 2023-03-21 | 北京天融信网络安全技术有限公司 | Network security situation prediction method and device based on timing diagram neural network |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107294795A (en) * | 2017-08-02 | 2017-10-24 | 上海上讯信息技术股份有限公司 | A kind of network security situation prediction method and equipment |
| CN109379379A (en) * | 2018-12-06 | 2019-02-22 | 中国民航大学 | Based on the network inbreak detection method for improving convolutional neural networks |
| US20200103894A1 (en) * | 2018-05-07 | 2020-04-02 | Strong Force Iot Portfolio 2016, Llc | Methods and systems for data collection, learning, and streaming of machine signals for computerized maintenance management system using the industrial internet of things |
| CN110956111A (en) * | 2019-11-22 | 2020-04-03 | 苏州闪驰数控系统集成有限公司 | Artificial intelligence CNN, LSTM neural network gait recognition system |
| CN111213160A (en) * | 2017-11-14 | 2020-05-29 | 谷歌有限责任公司 | Efficient convolutional neural network |
-
2021
- 2021-01-28 CN CN202110117235.9A patent/CN112801185B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107294795A (en) * | 2017-08-02 | 2017-10-24 | 上海上讯信息技术股份有限公司 | A kind of network security situation prediction method and equipment |
| CN111213160A (en) * | 2017-11-14 | 2020-05-29 | 谷歌有限责任公司 | Efficient convolutional neural network |
| US20200103894A1 (en) * | 2018-05-07 | 2020-04-02 | Strong Force Iot Portfolio 2016, Llc | Methods and systems for data collection, learning, and streaming of machine signals for computerized maintenance management system using the industrial internet of things |
| CN109379379A (en) * | 2018-12-06 | 2019-02-22 | 中国民航大学 | Based on the network inbreak detection method for improving convolutional neural networks |
| CN110956111A (en) * | 2019-11-22 | 2020-04-03 | 苏州闪驰数控系统集成有限公司 | Artificial intelligence CNN, LSTM neural network gait recognition system |
Non-Patent Citations (3)
| Title |
|---|
| NAZANIN MOARREF ET AL.: "A Transient Stability Prediction Method based on Multi-Channel Convolutional Neural Networks Using Time Series of PMU Measurements", 《IEEE》 * |
| 任德志: "基于机器学习的安全态势感知特征提取技术研究", 《中国优秀博硕士学位论文全文数据库(硕士)信息科技辑》 * |
| 张任川 等: "应用改进卷积神经网络的网络安全态势预测方法", 《计算机工程与应用》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114970694A (en) * | 2022-05-12 | 2022-08-30 | 河北师范大学 | Network security situation assessment method and model training method thereof |
| CN115834174A (en) * | 2022-11-15 | 2023-03-21 | 北京天融信网络安全技术有限公司 | Network security situation prediction method and device based on timing diagram neural network |
| CN115834174B (en) * | 2022-11-15 | 2023-06-09 | 北京天融信网络安全技术有限公司 | Network security situation prediction method and device based on time sequence diagram neural network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112801185B (en) | 2023-02-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112149316B (en) | Aero-engine residual life prediction method based on improved CNN model | |
| CN113159051B (en) | Remote sensing image lightweight semantic segmentation method based on edge decoupling | |
| Yang et al. | Lightweight feature fusion network for single image super-resolution | |
| CN110517329B (en) | Deep learning image compression method based on semantic analysis | |
| CN112801185B (en) | Network security situation understanding and evaluating method based on improved neural network | |
| CN110502988A (en) | Group positioning and anomaly detection method in video | |
| CN113269077B (en) | Underwater acoustic communication signal modulation mode identification method based on improved gating network and residual error network | |
| CN112489164B (en) | Image coloring method based on improved depth separable convolutional neural network | |
| CN108256482A (en) | A kind of face age estimation method that Distributed learning is carried out based on convolutional neural networks | |
| CN107240136A (en) | A kind of Still Image Compression Methods based on deep learning model | |
| CN112364719A (en) | Method for rapidly detecting remote sensing image target | |
| CN110599502B (en) | Skin lesion segmentation method based on deep learning | |
| CN109857871A (en) | A kind of customer relationship discovery method based on social networks magnanimity context data | |
| CN114693064B (en) | Building group scheme generation performance evaluation method | |
| CN113362242B (en) | Image restoration method based on multi-feature fusion network | |
| CN110728186B (en) | Fire detection method based on multi-network fusion | |
| CN116403063A (en) | No-reference screen content image quality assessment method based on multi-region feature fusion | |
| CN109918542A (en) | A kind of convolution classification method and system for relationship diagram data | |
| CN115546032A (en) | Single-frame image super-resolution method based on feature fusion and attention mechanism | |
| CN115471016A (en) | Typhoon prediction method based on CISSO and DAED | |
| CN112487938A (en) | Method for realizing garbage classification by utilizing deep learning algorithm | |
| CN113313180B (en) | Remote sensing image semantic segmentation method based on deep confrontation learning | |
| CN116306780B (en) | Dynamic graph link generation method | |
| CN113343760A (en) | Human behavior recognition method based on multi-scale characteristic neural network | |
| CN115272776A (en) | Hyperspectral image classification method based on double-path convolution and double attention and storage medium |
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 |