基于全尺度和注意力融合学习的车辆再识别方法

Vehicle Re-identification Method Based on Omni-scale and Attention Fusion Learning

车辆再识别(Re-identification)是计算机视觉领域的研究热点之一,其关键在于车辆辨别性特征的提取.为了更好地提取此类特征,本文提出了一种基于全尺度和注意力融合学习的特征提取方法,该方法通过多个感受野获取不同尺度的特征,并将提取到的不同尺度特征融合;同时为了在特征提取过程中重点关注辨别性特征,特引入注意力机制,增强特征的表达能力.经实验证明,该方法在VeRi-776主流数据集上的Rank-1和mAP均优于其他主流方法.

Re-identification of vehicles is one of the research hotspots in the field of computer vision,The key of Re-identification is the extraction of vehicle discriminative features.In order to better extract such features,this paper proposes a feature extraction method based on full-scale and attention fusion learning.This method acquires features of different scales through multiple receptive fields,and merges the extracted features of different scales.In order to focus on discriminative features in the feature extraction process,we introduce attention mechanism to enhance the expressive ability of features.Experiments have proved that this method is superior to other mainstream methods in both Rank-1 and mAP on the VeRi-776 mainstream dataset.