Adversarial sample generation method based on target detection model feature vector migration

Abstract

The invention discloses an adversarial sample generation method based on target detection model feature vector migration. The adversarial sample generation method comprises the following steps of S1, convolutional neural network feature vector migration; step S2, generating adversarial noise; and step S3, evaluating an adversarial sample attack effect. According to the method, the attack effect of the confrontation sample in deep learning models such as target detection is stronger, and better mobility is achieved on the premise that attack concealment is considered. The adversarial sample generation method disclosed by the invention reveals the important role of the feature vector in the adversarial mechanism of the target detection model, verifies the attack threat of the adversarial sample disclosed by the invention, and can heuristically explore the research in the field of a robust target detection algorithm, so that a new defense mechanism is designed; and the method is of great significance to application of the target detection model in actual life.

Description

technical field [0001] The invention relates to the fields of computer vision and deep learning, and in particular to a method for generating confrontation samples based on the transfer of feature vectors of target detection models. Background technique [0002] With the rapid development of artificial intelligence technology, more and more deep learning models play a greater role in production and life. Deep learning has a wide range of applications in many fields, such as: face recognition, fingerprint unlocking, smart medical care, smart input methods, voice assistants, and more. Although deep learning has shown strong advantages in various application scenarios, the emergence of adversarial examples has caused great obstacles and security risks to the promotion of deep learning. Adversarial examples are a special kind of input that can make the model misjudge with serious consequences. Taking the image classification task as an example, for a certain input image, the d...

AI Technical Summary

Problems solved by technology

In order to improve the attack capability of adversarial samples, attackers often adopt the method of increasing the adversarial noise, but this will lead to distortion of adversarial samples, poor concealment and reduced attack usability

Images