Pedestrian re-identification system adversarial sample generation method and system

Abstract

The invention discloses a pedestrian re-identification system adversarial sample generation method and system, and the method comprises the steps: S1, inputting an original image into a generator based on a residual network, and generating adversarial disturbance; S2, adding the adversarial disturbance to the original picture bit by bit to generate a coarse adversarial sample I', performing feature connection on the coarse adversarial sample I' and the original picture, and inputting the coarse adversarial sample I' into a multi-stage discriminator to generate a binary mask graph capable of controlling the number of adversarial disturbance points; S3, performing bitwise multiplication on the binary mask pattern and the anti-disturbance; s4, inputting the adversarial samples into a pedestrian re-identification model to be attacked, and taking a return value of the model as input of a feature confusion loss function, an adversarial learning loss function, a smooth classification confusion function and a multi-scale structure similarity loss function; and S5, iteratively carrying out training processes from S1 to S4 for multiple times, and updating parameters of the generator and themulti-stage discriminator.

Description

technical field [0001] The present invention relates to the field of computer vision based on deep learning, in particular to a method and system for generating adversarial samples for pedestrian re-identification systems based on feature confusion and multi-stage adversarial generation networks. Background technique [0002] In recent years, deep neural networks have achieved widespread success in computer vision tasks. As one of the most important computer vision tasks, person re-identification is to match individuals in the camera by extracting and measuring distinguishable features from pairs of images. Due to the excellent performance of advanced methods in recent years (such as the research work "Beyond part models: Person retrieval with refined part pooling and a strong convolutional baseline (ECCV)" by Sun et al. in 2018), pedestrian re-identification is very important in public security video surveillance or criminal identification. began to be widely used. [000...

AI Technical Summary

Problems solved by technology

Therefore, the explosive increase in the number of object categories is the fundamental challenge for deceiving visual classification systems that class methods cannot transfer efficiently.

Images