CN112507338B - Improved system based on deep learning semantic segmentation algorithm - Google Patents
Improved system based on deep learning semantic segmentation algorithm Download PDFInfo
- Publication number
- CN112507338B CN112507338B CN202011521918.2A CN202011521918A CN112507338B CN 112507338 B CN112507338 B CN 112507338B CN 202011521918 A CN202011521918 A CN 202011521918A CN 112507338 B CN112507338 B CN 112507338B
- Authority
- CN
- China
- Prior art keywords
- module
- virus
- deep learning
- characteristic
- segmentation algorithm
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F21/00—Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
- G06F21/50—Monitoring users, programs or devices to maintain the integrity of platforms, e.g. of processors, firmware or operating systems
- G06F21/55—Detecting local intrusion or implementing counter-measures
- G06F21/56—Computer malware detection or handling, e.g. anti-virus arrangements
- G06F21/562—Static detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/24—Querying
- G06F16/245—Query processing
-
- 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
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Computer Security & Cryptography (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Computer Hardware Design (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Evolutionary Computation (AREA)
- Mathematical Physics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Databases & Information Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Virology (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses an improved system based on a deep learning semantic segmentation algorithm, which comprises a main processor module, a deep learning module, an information output module, a wireless transmission module, a receiving terminal module, a safety inspection module, a data submission module, a power supply module, a segmentation algorithm module, a virus searching and killing module and a data recording module, wherein the deep learning module is used for searching and killing viruses; the output end of the main processor module, the deep learning module and the segmentation algorithm module are sequentially connected, the power supply output end of the power supply module is connected with the power supply input end of the main processor module, and the data inspection module is installed outside the main processor module; the power supply output end of the power supply module is connected with the power supply input end of the main processor module, and the data inspection module is fixedly installed outside the main processor module; the output end of the main processor module is connected with the input end of the information output module.
Description
Technical Field
The invention belongs to the technical field of learning algorithm range, and particularly relates to an improved system based on a deep learning semantic segmentation algorithm.
Background
Deep learning is the intrinsic law and expression level of the learning sample data, and the information obtained in the learning process is very helpful for the interpretation of data such as characters, images and sounds. The final aim of the method is to enable the machine to have the analysis and learning capability like a human, and to recognize data such as characters, images and sounds. Deep learning is a complex machine learning algorithm, and achieves the effect in speech and image recognition far exceeding the prior related art.
However, when a common system is used, the data inspection module is internally lacked, so that when the segmentation algorithm is carried out, data are easy to generate errors, and therefore, the working efficiency is low, and meanwhile, antivirus software is lacked for real-time protection, which is not beneficial to the safety of the system.
Disclosure of Invention
The invention aims to: in order to solve the above-mentioned problems, an improved system of deep learning semantic segmentation algorithm is provided.
The invention is realized by at least one of the following technical schemes.
An improved system based on a deep learning semantic segmentation algorithm comprises a main processor module, a deep learning module, an information output module, a wireless transmission module, a receiving terminal module, a safety inspection module, a data submission module, a power supply module, a segmentation algorithm module, a virus searching and killing module and a data recording module; the output end of the main processor module, the deep learning module and the segmentation algorithm module are sequentially connected, the power supply output end of the power supply module is connected with the power supply input end of the main processor module, and the data inspection module is installed outside the main processor module;
the output end of the main processor module, the information output module, the wireless transmission module, the receiving terminal module, the safety inspection module and the input end of the virus searching and killing module are sequentially connected;
the output end of the virus searching and killing module is connected with the input end of the data inspection module; the output end of the safety inspection module is connected with the input end of the data submission module, and the output end of the data submission module is connected with the input end of the data recording module.
Preferably, the symbol checking module, the logic checking module, the keyword setting module and the information receiving module are arranged in the data checking module; the input ends of the symbol checking module, the logic checking module, the keyword setting module and the information receiving module are all connected with the output end of the data checking module.
Preferably, the firewall module, the cloud virus library upgrading module and the intelligent warning module are arranged outside the virus searching and killing module; the input ends of the firewall module, the cloud virus library upgrading module and the intelligent warning module are all connected with the output end of the virus checking and killing module; the output end of the virus checking and killing module is connected with the input ends of the firewall module, the cloud virus library upgrading module and the intelligent warning module.
Preferably, the deep learning module adopts a deep learning method of a convolutional neural network, the neural network is an LcNet-5 model formed by a convolutional layer, a pooling layer, a down-sampling layer and a full-connection layer, and supervised training learning is performed by using an error gradient back propagation algorithm.
Preferably, the LcNet-5 model introduces local connection, weight sharing and pooling; each neuron of the LcNet-5 model is connected with the pixel of one region only, and the region connected with the neuron is called local receptive field; the weight sharing is realized by directly assigning the length and width parameters in the receptive field to a convolution kernel, and the parameters forming the convolution kernel are the size of the receptive field; the fixed size regions on the convolution output are averaged or maximized to form pooled posing.
Preferably, the segmentation algorithm module adopts a superpixel segmentation algorithm, and performs superpixel segmentation on the original image to obtain a superpixel region formed by aggregating pixels with similar characteristics, and spatial position information between the superpixels is blended in the segmentation process to further improve the segmentation effect.
Preferably, the firewall module comprises a virtual machine, a virus signature code and a virus detection module, wherein the virtual machine is used for detecting whether a virus is infected or not;
the way to judge whether the virus is infected is as follows:
scanning a detected object by using a specific virus code contained in each virosome through a characteristic code scanning method, if a specific virus code is found in the detected object, then the virus represented by the virus code exists, analyzing the characteristic virus code of the virus and storing the characteristic virus code in a virus code library file, wherein the characteristic code scanning method is divided into characteristic code scanning and characteristic word scanning, the characteristic code refers to the characteristic virus code, the characteristic word scanning is to extract key characteristic words from the virus body to form a characteristic word library, the scanned object, namely the characteristic words, are compared with the characteristic codes during scanning, and if the characteristic words are matched, the scanned object, namely the characteristic words, is judged to be infected with the virus; aiming at the encrypted virus, the feature code scanning method cannot scan and detect the feature code, so a set of Windows running environment is virtualized by software, the virus is opened and attacks the running environment under the environment, namely a virtual machine, the encrypted virus is finally decrypted during the execution, and the virus is confirmed by feature inspection after the decryption.
Preferably, a human-computer interaction module is arranged outside the data inspection module, and the human-computer interaction module comprises an operation keyboard, a handwriting input board and a control mouse, and the operation keyboard, the handwriting input board and the control mouse are connected with a USB interface inside the human-computer interaction module.
Preferably, the general processor module is a CPU Intel i5.
Preferably, the sign checking module, the logic checking module and the keyword setting module all use an LMS matching algorithm.
In summary, compared with the prior art, the invention has the beneficial effects that:
1. in the invention, the data inspection module can inspect the data to be processed by the segmentation algorithm module and the deep learning module in the working process of the segmentation algorithm module and the deep learning module, thereby ensuring the tidiness of the data, avoiding the wrong data from being processed together, improving the working efficiency of the system, strengthening the processing capacity in the system and lightening the working operation burden of workers.
2. In the invention, the virus killing module is matched with the safety inspection module to carry out virus killing on the processed data in real time, so that the conditions of data damage and loss caused by virus mixing into the data are prevented, the antiviral capability of the system is improved, and the safe and smooth operation of the system is ensured.
Drawings
FIG. 1 is a block diagram of an improved system based on a deep learning semantic segmentation algorithm according to the present invention;
FIG. 2 is a block diagram of a data verification module according to the present invention;
FIG. 3 is a block diagram of a virus searching and killing module according to the present invention;
the labels in the figure are: 1-a main processor module, 2-a deep learning module, 3-an information output module, 4-a wireless transmission module, 5-a receiving terminal module, 6-a security inspection module, 7-a data submission module, 8-a power supply module, 9-a segmentation algorithm module, 10-a data inspection module, 11-a virus checking and killing module, 12-a data recording module, 13-a symbol inspection module, 14-a logic inspection module, 15-a keyword setting module, 16-an information receiving module, 17-a firewall module, 18-a cloud virus library upgrading module and 19-an intelligent warning module.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
The improved system based on the deep learning semantic segmentation algorithm shown in fig. 1 comprises a main processor module 1, a deep learning module 2, an information output module 3, a wireless transmission module 4, a receiving terminal module 5, a security inspection module 6, a data submission module 7, a power supply module 8, a segmentation algorithm module 9, a data inspection module 10, a virus checking and killing module 11, a data recording module 12, a symbol inspection module 13, a logic inspection module 14, a keyword setting module 15, an information receiving module 16, a firewall module 17, a cloud virus library upgrading module 18 and an intelligent warning module 19.
The output end of the main processor module 1 is connected with the input end of the deep learning module 2, and the output end of the deep learning module 2 is connected with the input end of the segmentation algorithm module 9.
The segmentation algorithm module 9 adopts a superpixel segmentation algorithm, the superpixel is an irregular area formed by adjacent pixels, and the pixels are similar in texture, color, brightness and the like; the original image is subjected to superpixel segmentation, so that a plurality of superpixel regions formed by aggregating pixels with similar characteristics can be obtained; when the super pixels are used as basic units for segmentation, the influence of noise points in the image on the segmentation result can be reduced, and in addition, because the number of the super pixels is relatively small, the spatial position information among the super pixels can be integrated in the segmentation process, so that the segmentation effect is improved; the super-pixel segmentation method based on the graph theory takes the pixels in the image as nodes in a weighted undirected graph, edges in the graph represent the relation of adjacent pixels, the weight of the edges is represented by the similarity or the difference between the pixels, and finally different rules are used for dividing the pixels to complete segmentation.
If the size of an input picture is W X H, if a full-connection network is used to generate an X X Y featuremap, X X Y F parameters are needed, if the length and width of the original picture are in the level of 10X 2, and the size of W X H is as much as that of X Y, the number of the parameters needed by such a layer of network is 108 to 1012, and the number of the parameters is reduced for each pixel on the output layer featuremap, so that each pixel of the original picture is connected, and each link needs one parameter; however, since the images are generally locally related, if each pixel of the output layer is only connected with one local part of the input layer picture, the number of required parameters is greatly reduced; assuming that each pixel of the output layer is connected to only one small square of W X H on the input picture, i.e. the pixel value of the output layer is calculated only from the pixel values in the small square of F X F of the original picture, the number of required parameters for each pixel of the output layer is reduced from the original W X Y to F.
The deep learning module 2 adopts a deep learning method of a convolutional neural network, the response of each layer in the convolutional neural network is obtained by applying a group of neurons with the same parameters to local region excitation on different positions of the previous layer, and therefore, the deep learning module is insensitive to position, shape and scale; the convolutional neural network is an LcNet-5 model consisting of a convolutional layer, a pooling layer, a downsampling layer and a full-connection layer, and supervised training learning is performed by using an error gradient back propagation algorithm; this network model has enjoyed great success in handwritten digit recognition; the LcNet-5 model comprises the ideas of local connection, weight sharing and pooling; in the neural network of the LcNet-5 model, each node in each layer is connected with all the points in the upper layer, which results in a large number of parameters in the network, thus increasing the complexity of training the network model; in order to reduce the parameter quantity in the network, each neuron is only connected with pixels of one region, and the region connected with the neuron is called a local receptive field; in addition, because the statistical characteristics of the image obtained by using convolution operation are irrelevant to the position, weight sharing is introduced, and the parameter of a convolution kernel is the size of a receptive field; in order to further reduce network parameters and improve the generalization capability of a network model, a fixed-size area on convolution output can be averaged or maximized, and the aggregation operation is pooling posing; the introduction of local connectivity, weight sharing and pooling in CNNs greatly reduces the parameters of the network, making training learning for deeper networks possible as well.
The general processor module is a CPU Intel i5; the information output module uses an output module in python; the wireless transmission module uses Wifi wireless transmission; the receiving terminal module uses an input module in python; the security inspection module uses prefix tree sensitive word filtering; the data submitting module and the data recording module both use logging logs; the power supply module uses a 600W power supply.
The power supply output end of the power supply module 8 is connected with the power supply input end of the main processor module 1, and the data inspection module 10 is fixedly installed outside the main processor module 1.
As shown in fig. 2, the data checking module 10 is internally provided with the symbol checking module 13, the logic checking module 14, the keyword setting module 15 and the information receiving module 16, and an output end of the data checking module 10 is connected to input ends of the symbol checking module 13, the logic checking module 14, the keyword setting module 15 and the information receiving module 16.
The sign checking module 13, the logic checking module 14 and the keyword setting module 15 all use an LMS matching algorithm.
The output end of the main processor module 1 is connected with the input end of the information output module 3, the output end of the information output module 3 is connected with the input end of the wireless transmission module 4, and the output end of the wireless transmission module 4 is connected with the input end of the receiving terminal module 5.
As shown in fig. 3, the firewall module 17, the cloud virus library upgrading module 18, and the intelligent warning module 19 are disposed outside the virus antivirus module 11, and an output end of the virus antivirus module 11 is connected to the firewall module 17, the cloud virus library upgrading module 18, and an input end of the intelligent warning module 19.
The firewall module 17 includes a function of determining whether a virus is infected, a virus signature scan and a virtual machine,
the way to judge whether the virus is infected is as follows:
the detected object is scanned by a characteristic code scanning method by using a specific virus code contained in each virosome. If a specific virus code is found in the detected object, the virus represented by the virus code is found, the characteristic virus code of the virus is analyzed and stored in a virus code library file, the characteristic code scanning method is divided into characteristic code scanning and characteristic word scanning, the characteristic code refers to the characteristic virus code, the characteristic word scanning is to extract key characteristic words from the virus body, the characteristic words comprise, for example, royal, games and the like to form a characteristic word library, the scanned object, namely the characteristic word, is compared with the characteristic code during scanning, and if the characteristic words are matched, the scanned object, namely the characteristic word, is judged to be infected with the virus; aiming at the viruses passing through encryption, the feature code scanning method cannot scan and detect the feature codes, so a set of Windows operating environment is virtualized by software, the viruses are opened and attack the operating environment under the environment, namely a virtual machine, the encrypted viruses are finally decrypted during the execution, and the viruses are confirmed through feature verification (shown in a graph 1) after the decryption.
TABLE 1 characterisation test
The output end of the receiving terminal module 5 is connected with the input end of the safety inspection module 6, the output end of the safety inspection module 6 is connected with the input end of the virus searching and killing module 11, and the output end of the virus searching and killing module 11 is connected with the input end of the data inspection module 10; the output end of the safety inspection module 6 is connected with the input end of the data submission module 7, and the output end of the data submission module 7 is connected with the input end of the data recording module 12.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents and improvements made within the spirit and principle of the present invention are intended to be included within the scope of the present invention.
Claims (9)
1. An improved system based on a deep learning semantic segmentation algorithm is characterized in that: the system comprises a main processor module (1), a deep learning module (2), an information output module (3), a wireless transmission module (4), a receiving terminal module (5), a safety inspection module (6), a data submission module (7), a power supply module (8), a segmentation algorithm module (9), a virus checking and killing module (11) and a data recording module (12); the output end of the main processor module (1), the deep learning module (2) and the segmentation algorithm module (9) are sequentially connected, the power output end of the power supply module (8) is connected with the power input end of the main processor module (1), and a data inspection module (10) is installed outside the main processor module (1);
the output end of the main processor module (1), the information output module (3), the wireless transmission module (4), the receiving terminal module (5), the safety inspection module (6) and the input end of the virus checking and killing module (11) are sequentially connected;
the output end of the virus searching and killing module (11) is connected with the input end of the data inspection module (10); the output end of the safety inspection module (6) is connected with the input end of the data submission module (7), and the output end of the data submission module (7) is connected with the input end of the data recording module (12);
a symbol checking module (13), a logic checking module (14), a keyword setting module (15) and an information receiving module (16) are arranged in the data checking module (10); the input ends of the symbol checking module (13), the logic checking module (14), the keyword setting module (15) and the information receiving module (16) are all connected with the output end of the data checking module (10).
2. The improved system based on deep learning semantic segmentation algorithm according to claim 1, characterized in that: a firewall module (17), a cloud virus library upgrading module (18) and an intelligent warning module (19) are arranged outside the virus checking and killing module (11); the input ends of the firewall module (17), the cloud virus library upgrading module (18) and the intelligent warning module (19) are all connected with the output end of the virus searching and killing module (11); the output end of the virus checking and killing module (11) is connected with the input ends of the firewall module (17), the cloud virus library upgrading module (18) and the intelligent warning module (19).
3. The improved system based on deep learning semantic segmentation algorithm according to claim 2, characterized in that: the deep learning module (2) adopts a deep learning method of a convolutional neural network, the neural network is an LcNet-5 model formed by a convolutional layer, a pooling layer, a down-sampling layer and a full-connection layer, and supervised training learning is carried out by using an error gradient back propagation algorithm.
4. The improved system based on deep learning semantic segmentation algorithm according to claim 3, characterized in that: the LcNet-5 model introduces local connection, weight sharing and pooling; each neuron of the LcNet-5 model is connected with the pixel of one region only, and the region connected with the neuron is called local receptive field; the weight sharing is realized by directly assigning the length and width parameters in the receptive field to a convolution kernel, and the parameters forming the convolution kernel are the size of the receptive field; the fixed size regions on the convolution output are averaged or maximized to form pooled posing.
5. The improved system based on deep learning semantic segmentation algorithm according to claim 4, characterized in that: the segmentation algorithm module (9) adopts a superpixel segmentation algorithm, and performs superpixel segmentation on the original image to obtain a superpixel region formed by aggregating pixels with similar characteristics, and spatial position information among the superpixels is blended in the segmentation process to further improve the segmentation effect.
6. The improved system for deep learning semantic segmentation algorithm according to claim 5, characterized in that: the firewall module (17) comprises a virtual machine, a virus characteristic code scanning module and a virus detection module, wherein the virtual machine is used for judging whether a virus is infected or not;
the way to judge whether the virus is infected is as follows:
scanning a detected object by using a specific virus code contained in each virosome through a characteristic code scanning method, if a specific virus code is found in the detected object, then the virus represented by the virus code exists, analyzing the characteristic virus code of the virus and storing the characteristic virus code in a virus code library file, wherein the characteristic code scanning method is divided into characteristic code scanning and characteristic word scanning, the characteristic code refers to the characteristic virus code, the characteristic word scanning is to extract key characteristic words from the virus body to form a characteristic word library, the scanned object, namely the characteristic words, are compared with the characteristic codes during scanning, and if the characteristic words are matched, the scanned object, namely the characteristic words, is judged to be infected with the virus; aiming at the encrypted virus, the feature code scanning method cannot scan and detect the feature code, so a set of Windows running environment is virtualized by software, the virus is opened and attacks the running environment under the environment, namely a virtual machine, the encrypted virus is finally decrypted during the execution, and the virus is confirmed by feature inspection after the decryption.
7. The improved system based on deep learning semantic segmentation algorithm according to claim 6, characterized in that: the data inspection module (10) is externally provided with a human-computer interaction module, the human-computer interaction module comprises an operation keyboard, a handwriting input board and a control mouse, and the operation keyboard, the handwriting input board and the control mouse are connected with an internal USB interface of the human-computer interaction module.
8. The improved system based on deep learning semantic segmentation algorithm according to claim 7, characterized in that: the general processor module (1) is a CPU Intel i5.
9. The improved system based on deep learning semantic segmentation algorithm according to claim 8, characterized in that: the sign checking module (13), the logic checking module (14) and the keyword setting module (15) all use LMS matching algorithm.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011521918.2A CN112507338B (en) | 2020-12-21 | 2020-12-21 | Improved system based on deep learning semantic segmentation algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011521918.2A CN112507338B (en) | 2020-12-21 | 2020-12-21 | Improved system based on deep learning semantic segmentation algorithm |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112507338A CN112507338A (en) | 2021-03-16 |
| CN112507338B true CN112507338B (en) | 2023-02-14 |
Family
ID=74922837
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011521918.2A Active CN112507338B (en) | 2020-12-21 | 2020-12-21 | Improved system based on deep learning semantic segmentation algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112507338B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113409322B (en) * | 2021-06-18 | 2022-03-08 | 中国石油大学(华东) | Deep learning training sample enhancement method for semantic segmentation of remote sensing image |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109711413A (en) * | 2018-12-30 | 2019-05-03 | 陕西师范大学 | Image, semantic dividing method based on deep learning |
| CN109740615A (en) * | 2018-12-29 | 2019-05-10 | 武汉大学 | A kind of minimizing technology of pair of attack resistance sample disturbance |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019014487A1 (en) * | 2017-07-12 | 2019-01-17 | The Regents Of The University Of California | Detection and prevention of adversarial deep learning |
-
2020
- 2020-12-21 CN CN202011521918.2A patent/CN112507338B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109740615A (en) * | 2018-12-29 | 2019-05-10 | 武汉大学 | A kind of minimizing technology of pair of attack resistance sample disturbance |
| CN109711413A (en) * | 2018-12-30 | 2019-05-03 | 陕西师范大学 | Image, semantic dividing method based on deep learning |
Non-Patent Citations (1)
| Title |
|---|
| 深度学习模型的中毒攻击与防御综述;陈晋音 等;《信息安全学报》;20200731;第5卷(第4期);第14-26页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112507338A (en) | 2021-03-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zhang et al. | Adversarial examples: Opportunities and challenges | |
| Bozkir et al. | Catch them alive: A malware detection approach through memory forensics, manifold learning and computer vision | |
| Baptista et al. | A novel malware detection system based on machine learning and binary visualization | |
| Ito et al. | Web application firewall using character-level convolutional neural network | |
| Fang et al. | Android malware familial classification based on dex file section features | |
| CN110704840A (en) | Convolutional neural network CNN-based malicious software detection method | |
| Tang et al. | Dynamic API call sequence visualisation for malware classification | |
| Zhao et al. | A review of computer vision methods in network security | |
| Lazzarini et al. | A stacking ensemble of deep learning models for IoT intrusion detection | |
| CN113159045A (en) | Verification code identification method combining image preprocessing and convolutional neural network | |
| Khan et al. | Malware classification framework using convolutional neural network | |
| CN112507338B (en) | Improved system based on deep learning semantic segmentation algorithm | |
| Conti et al. | A few-shot malware classification approach for unknown family recognition using malware feature visualization | |
| Shen et al. | Feature fusion-based malicious code detection with dual attention mechanism and BiLSTM | |
| Wang et al. | Improving adversarial robustness of deep neural networks by using semantic information | |
| Liu et al. | Towards transferable unrestricted adversarial examples with minimum changes | |
| Van Dao et al. | An Attention Mechanism for Combination of CNN and VAE for Image-Based Malware Classification | |
| Sharif et al. | A deep learning based technique for the classification of malware images | |
| Warnecke et al. | Don’t paint it black: White-box explanations for deep learning in computer security | |
| CN111917781A (en) | Intelligent internal malicious behavior network attack identification method and electronic equipment | |
| Qiang | Research on software vulnerability detection method based on improved CNN model | |
| Marcinowski | Top interpretable neural network for handwriting identification | |
| CN114021136A (en) | Back door attack defense system for artificial intelligence model | |
| Wen et al. | A image texture and BP neural network basec malicious files detection technique for cloud storage systems | |
| CN113361422A (en) | Face recognition method based on angle space loss bearing |
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 |