基于注意力机制与多尺度特征融合的行人重识别方法

针对行人重识别中因遮挡、姿态变化使模型特征无法充分表达行人信息的问题，提出了基于注意力机制与多尺度特征融合的行人重识别方法。该方法首先使用改进的骨干网络（R-ResNet50）提取图像特征；其次，抽取网络不同尺度的特征层嵌入注意力机制（DANet）,使模型更关注于重点信息；最后，对提取出的关键特征进行多尺度特征融合，实现特征间的优势互补，并使用联合交叉熵损失、难样本采样三元组损失和中心损失的多损失函数策略对网络模型进行训练。实验结果表明，本文所提方法在Market1501、DukeMTMC-ReID数据集上的首位命中率（Rank-1）和平均精度均值（mAP）分别达到了92.7%、80.4%和86.4%、71.0%，模型提取的特征更具有判别性，识别率更高。

Person re-identification method based on attention mechanism and multi-scale feature fusion

For the problem that model features cannot fully express the person information due to occlusion and posture change in person re-identification, the person re-identification method based on the attention mechanism and multi-scale features fusion is proposed. In this method, firstly the improved backbone network (R-ResNet50) was used to extract image features; secondly, the feature layers of the network at different scales was extracted to embed in the attention mechanism (DANet), so that the model paid more attention to the key information; finally, the extracted key features were fused with multi-scale features to achieve complementary advantages among features, and the multi-loss function strategy of cross entropy loss, difficult sample triplet loss and center loss was used to train the network model. The experimental results show that the Rank-1 and mAP of this method on the Market1501 and DukeMTMC-ReID dataset are 92.7%, 80.4% and 86.4%, 71.0% respectively, so the features extracted from the model are more discriminant and the recognition rate is higher.