Convolutional Neural Networks for Vehicle Re-identification with Adversarial Loss

Since the concept of the generative adversarial networks (GAN) was put forward in 2014, it has attracted great attention from researchers to continuously improve and apply. With the advent of the era of big data, the structure of deep convolutional neural networks has become more complex, and the feature learning and feature expression capabilities have improved compared with traditional machine learning methods. As a newly emerging research direction, vehicle re-identification (ReID) is of great significance. However, because of the images that can be obtained in practical applications often encounter artificially intentional or occlusion due to conditional constraints and characteristics of vehicle re-identification, the existing recognition rate is still not satisfactory. In order to solve this problem, this paper uses the adversarial idea in the generative adversarial networks combined the convolutional neural networks (CNN) to improve the baseline algorithm. In this paper, a lot of experiments are carried out on the standard dataset Vehicle ID. By comparing and analyzing the experimental results with the existing baseline algorithm, the proposed method embodies better recognition performance.