融合注意力与多层次特征提取的行人再识别方法

Pedestrian re-identification method fusing attention and multi-level feature extraction

在行人再识别任务中,仅使用全局特征并不能很好地表示行人特征。提出一种将局部特征与全局特征相融合并嵌入注意力机制的方法。采用2个网络分支结构,分别进行局部和全局特征提取,并且在提取行人特征过程中,嵌入空间注意力机制和通道注意力机制,通过关注图像的局部和全局信息,增强特征表示能力。在数据处理阶段,引入随机加噪与随机擦除的数据增强方法,解决行人再识别过程中物体遮挡及噪声干扰问题,提高模型鲁棒性。在3个数据集上进行实验,将几种先进的行人再识别方法相比较,结果表明,该方法的均值平均精度(mAP)值和Rank—1值更高。

In pedestrian re-identification tasks,only using global features cannot represent pedestrian features very well.A method is proposed to fuse local features and global features and embed them into the attention mechanism.In this method,two network branches are used to extract local and global features,respectively.In the process of extracting pedestrian features,spatial attention mechanism and channel attention mechanism are embedded,and the feature representation capability is enhanced by paying attention to the local and global information of the image.In the data processing stage,random noise and random erasure are introduced to enhance the data to solve the problem of object occlusion and noise interference in the process of pedestrian re-identification,and improve the robustness of the model.Experimental results on three datasets show that the mAP and Rank—1 values of the proposed method are higher than those of several advanced pedestrian re-identification methods.