Small pixel confrontation sample defense method for image recognition process
A technology against samples and image recognition, applied in character and pattern recognition, instruments, computer parts, etc., can solve problems such as image recognition errors, and achieve simple and practical effects
Active Publication Date: 2020-10-16
UNIV OF ELECTRONICS SCI & TECH OF CHINA
View PDF6 Cites 1 Cited by
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
[0007] In view of the above-mentioned technical problems, the purpose of the present invention is to solve the problem of wrong image recognition after adding disturbance and camouflage in the small pixel image recognition process
Method used
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View moreImage
Smart Image Click on the blue labels to locate them in the text.
Smart ImageViewing Examples
Examples
Experimental program
Comparison scheme
Effect test
Embodiment
[0034] S1: In the data preparation stage, certain elemental data need to be collected as a training data set.
[0035] S2: Data preprocessing stage, which screens useful data in the original data and discards useless information (such as removing label information, only keeping label numbers, etc.), making the model easier to train;
[0036] S3: Model pre-training stage. In this stage, the conventional neural network model training method is used. The element data set and data set label are used as data input, which are transmitted to the machine learning framework for supervised learning, and an original model is trained as an unenhanced model. the model;
[0037] S4: The stage of adversarial sample generation. In this stage, a small pixel image adversarial sample generation algorithm (such as C&W, DeepFool) is used to generate a large number of adversarial samples for the training data set.
[0038] S5: The feature extraction stage of adversarial samples. In this stage, sta...
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More PUM
Login to View More Abstract
The invention relates to the field of machine learning, provides a small pixel confrontation sample defense method for an image recognition process, and aims to solve the problem of image recognitionerrors after disturbance camouflage is added in the small pixel image recognition process. The main scheme includes steps of carrying out model training on the original sample set O; obtaining an unenhanced classification model, performing adversarial sample generation on each picture in the training data set to obtain an adversarial sample set A, performing disturbance value statistics, performing statistics on the proportion of each disturbance value, calculating the disturbance value distribution of each adversarial sample, and obtaining a disturbance distribution histogram of the adversarial samples; using a disturbance distribution histogram of the adversarial sample to simulate a disturbance law of an adversarial sample generation algorithm, and training the DUNet model to obtain a noise reduction input layer; and splicing the noise reduction input layer and the unenhanced small pixel image classification model to obtain an enhanced model.
Description
technical field [0001] A model defense method for small pixel adversarial samples, which is used for the defense of adversarial samples, belongs to the field of machine learning. Background technique [0002] In recent years, machine learning has developed rapidly. With the continuous advancement of machine learning technology, machine learning has gained a lot of attention in areas closely related to people's daily life, such as malicious email detection, malicious program detection, image recognition, face recognition, image classification, and driverless driving. Very wide range of applications. Machine learning has gradually penetrated into people's daily life and has become a key technology to improve people's living standards. However, while machine learning has brought great help to people's learning and life, there are still many security issues in machine learning algorithms. In early systems that used machine learning for network intrusion detection and malicious...
Claims
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More Application Information
Patent Timeline
Login to View More Patent Type & Authority Applications(China)
IPC IPC(8): G06K9/62
CPCG06F18/254G06F18/241G06F18/214
Inventor 牛伟纳张小松任仲蔚丁康一谢科张瑾昀曹蓉
Owner UNIV OF ELECTRONICS SCI & TECH OF CHINA